Received: 11 January 2017 Revised: 16 November 2017 Accepted: 16 November 2017
DOI: 10.1002/env.2488
RESEARCH ARTICLE
Nonparametric estimation of multivariate quantiles
M. Coblenz
1
R. Dyckerhoff
2
O. Grothe
1
1
Institute of Operations Research,
Karlsruhe Institute of Technologie (KIT),
76131 Karlsruhe, Germany
2
Institute of Econometrics and Statistics,
University of Cologne, 50923 Cologne,
Germany
Correspondence
M. Coblenz, Institute of Operations
Research, Karlsruhe Institute of
Technologie (KIT), 76131 Karlsruhe,
Germany.
Email: maximilian.coblenz@kit.edu
In many applications of hydrology, quantiles provide important insights in the
statistical problems considered. In this paper, we focus on the estimation of mul-
tivariate quantiles based on copulas. We provide a nonparametric estimation
procedure for a notion of multivariate quantiles, which has been used in a series
of papers. These quantiles are based on particular level sets of copulas and admit
the usual probabilistic interpretation that a p-quantile comprises a probability
mass p. We also explore the usefulness of a smoothed bootstrap in the estima-
tion process. Our simulation results show that the nonparametric estimation
procedure yields excellent results and that the smoothed bootstrap can be bene-
ficially applied. The main purpose of our paper is to provide an easily applicable
method for practitioners and applied researchers in domains such as hydrology
and coastal engineering.
KEYWORDS
copulas, multivariate quantiles in hydrology, smoothed bootstrap
1 INTRODUCTION
It is important to assess and quantify risk in complex environments. A statistical approach to do so is by using quan-
tiles. They provide an easy way to measure extreme events and their corresponding probabilities. Up to now, quantiles
are largely used in a univariate setting. However, the increase in complexity and the availability of larger data sets have
motivated researchers to transfer the concept into dimensions higher than one and to propose different definitions of mul-
tivariate quantiles (e.g., Di Bernardino, Laloë, Maume-Deschamps, & Prieur, 2013; Salvadori, Durante, & Perrone, 2013;
Serfling, 2002). Multivariate quantiles are especially useful when it is not possible or reasonable to combine the random
variables involved into one single variable of interest, that is, when no univariate analysis on the combined variables is
possible. This is particularly the case in application domains such as hydrology, where, for example, flood peak and flood
volume may not be meaningfully combined into one variable of interest.
Historically, hydrology and coastal engineering are areas where first examples of multivariate quantile approaches
gained attention in applications and could provide a more realistic picture than univariate approaches (for early refer-
ences, see Salvadori, 2004; Salvadori & De Michele, 2004; Yue & Rasmussen, 2002). Additionally, international guidelines
in these fields led researchers to consider multivariate approaches more closely (Salvadori, Durante, De Michele, Bernardi,
& Petrella, 2016). For example, Chebana and Ouarda (2011) used multivariate quantiles in a bivariate setting of fre-
quency analysis of floods. They modeled dependencies of flood volume and flood peak via a copula approach and
analyzed the resulting combinations for a given risk level. Requena, Mediero, and Garrote (2013) used multivariate
quantiles based on copulas in hydrologic dam design. Salvadori, Durante, Tomasicchio, and D'Alessandro (2015) used
copula-based multivariate quantiles to measure the probability of structural failure in coastal and offshore engineering,
measured by return period and design quantile. Recently, multivariate quantiles are also present in applications of
financial risk management. There, they lead to multivariate extensions of risk measures such as value at risk and
Environmetrics. 2018;e2488. wileyonlinelibrary.com/journal/env Copyright © 2018 John Wiley & Sons, Ltd. 1 of 23
https://doi.org/10.1002/env.2488