Structure learning for extremal tree models
Authors:
Preprint
Abstract
Extremal graphical models are sparse statistical models for multivariate extreme events. The underlying graph encodes conditional independencies and enables a visual interpretation of the complex extremal dependence structure. For the important case of tree models, we develop a data-driven methodology for learning the graphical structure. We show that sample versions of the extremal correlation and a new summary statistic, which we call the extremal variogram, can be used as weights for a minimum spanning tree to consistently recover the true underlying tree. Remarkably, this implies that extremal tree models can be learned in a completely non-parametric fashion by using simple summary statistics and without the need to assume discrete distributions, existence of densities, or parametric models for bivariate distributions.Keywords
All Related Versions
This publication has 38 references indexed in Scilit:
- One- versus multi-component regular variation and extremes of Markov treesAdvances in Applied Probability, 2020
- Multivariate peaks over thresholds modelsExtremes, 2017
- Exact simulation of max-stable processesBiometrika, 2016
- Conditional independence among max-stable lawsStatistics & Probability Letters, 2016
- Conditional simulation of max-stable processesBiometrika, 2012
- Extremal dependence measure and extremogram: the regularly varying caseExtremes, 2011
- A dependence measure for multivariate and spatial extreme values: Properties and inferenceBiometrika, 2003
- Dependence Measures for Extreme Value AnalysesExtremes, 1999
- Approximation of intermediate quantile processesJournal of Multivariate Analysis, 1987
- Shortest Connection Networks And Some GeneralizationsBell System Technical Journal, 1957