pith. sign in

arxiv: 2506.15272 · v2 · submitted 2025-06-18 · 📊 stat.ME

A penalized least squares estimator for extreme-value mixture models

Anas Mourahib , Anna Kiriliouk , Johan Segers This is my paper

Pith reviewed 2026-05-19 09:24 UTC · model grok-4.3

classification 📊 stat.ME
keywords extreme value modelsmixture modelspenalized estimationboundary parametersextreme directionsthreshold exceedancesmultivariate extremesleast squares
0
0 comments X

The pith

A penalized least squares estimator identifies boundary parameters in extreme-value mixture models by using pseudo-norm penalization on threshold exceedances.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a new estimator for the parameters of parametric mixture models used in multivariate extreme value analysis. When some variables do not participate in joint extremes, certain model parameters sit on the boundary of the allowable space. The estimator adds a pseudo-norm penalty to a least squares objective built from threshold exceedances; this penalty is meant to drive the irrelevant parameters exactly to their boundary values while leaving the rest unbiased. A companion algorithm then uses the estimated parameters to group variables into sets that can become large together. Simulations and two real datasets, river flows and stock losses, are used to check whether the approach recovers both the parameters and the groups correctly.

Core claim

The central claim is that a least squares objective augmented by a pseudo-norm penalization term, applied to exceedance data, recovers the parameters of a general extreme-value mixture model and, in particular, correctly sets to boundary values those parameters that correspond to variables not participating in an extreme event, while a data-driven procedure simultaneously identifies the groups of variables that do participate in such events.

What carries the argument

The pseudo-norm penalization term added to the least squares criterion based on threshold exceedances; it enforces boundary values for parameters linked to non-extreme directions.

If this is right

  • Parameter estimates remain accurate even when some mixture components correspond to boundary cases.
  • The data-driven grouping procedure returns sets of variables that can exceed thresholds together.
  • The same estimator can be applied directly to both environmental and financial extreme data.
  • Simulation performance improves relative to unpenalized least squares when boundary parameters are present.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The penalization idea could be tested on mixture models outside the extreme-value setting to see whether boundary identification improves more generally.
  • Risk measures that depend on joint tail behavior might become easier to compute once extreme-direction groups are identified automatically.
  • Higher-dimensional applications would reveal whether the computational cost of the penalization remains manageable as the number of variables grows.

Load-bearing premise

The penalization term pushes the appropriate parameters exactly to their boundary values without creating large bias in the estimates of the remaining parameters or in the detection of extreme-direction groups.

What would settle it

A simulation in which the estimator either leaves non-boundary parameters far from their true values or fails to recover the correct grouping of variables into extreme directions would show the method does not work as claimed.

Figures

Figures reproduced from arXiv: 2506.15272 by Anas Mourahib, Anna Kiriliouk, Johan Segers.

Figure 1
Figure 1. Figure 1: Log-transformed sample of size n = 100 from a max-stable Hüsler–Reiss model (left) and a max-stable mixture Hüsler–Reiss model (right). family of distributions. Multivariate modeling of d ⩾ 2 risk factors is a more challenging topic, since risks may exhibit some sort of dependence. The goal is not only to model rare events of each risk factor independently but to capture the extremal dependence between the… view at source ↗
Figure 2
Figure 2. Figure 2: Bivariate contours of (Σsa p s ) 1/p = 1 for given values of p. 2. write Aˆ new = (aˆ11,new, . . . , aˆd1,new, . . . , aˆ1r,new, . . . , aˆdr,new) and standardize to en￾sure unit row sums, aˆjs = ˆajs,new . X l=1,...,r aˆjl,new, j ∈ D, s ∈ R. (3.2) This yields Aˆ ∈ Md,r([0, 1]) with unit row sums; 3. let θˆ Z be the partial solution of (3.1) w.r.t. θZ, where A is fixed to Aˆ new. The resulting estimate θˆ … view at source ↗
Figure 3
Figure 3. Figure 3: Scores ED-S (top) and SMSE (bottom) defined in (4.3) and (4.4), respectively, as functions of the penalization exponent p, for the mixture logistic model perturbed as in (4.2) and sample sizes n = 1 000 (left), n = 2 000 (center), and n = 3 000 (right). Curves correspond to varying tail fractions, with approximately k/n = 0.01 (orange), k/n = 0.02 (blue), and k/n = 0.05 (green). Γ for the mixture Hüsler–Re… view at source ↗
Figure 4
Figure 4. Figure 4: Scores ED-S (top) and SMSE (bottom) defined in (4.3) and (4.4), respectively, as functions of the tail fraction k/n, for the mixture logistic model (left) and the mixture Hüsler–Reiss model (right), both perturbed as in (4.2), with penalization exponent p = 0.4. Curves correspond to n = 1 000 (orange), n = 2 000 (blue) and n = 3 000 (green). . 27 [PITH_FULL_IMAGE:figures/full_fig_p027_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Boxplots of the estimated coefficients Γij from the variogram matrix Γ for all pairs (i, j) included in some signature J of A. The true value 1 for each coefficient is presented with a horizontal red line. The tail fraction k/n is set to 0.06, the penalization exponent to p = 0.4 and the sample size to n = 1 000. 0.00 0.05 0.10 0.15 0.20 sample size n ED−S 500 1500 2500 3500 4500 [PITH_FULL_IMAGE:figures/… view at source ↗
Figure 6
Figure 6. Figure 6: The score ED-S defined in (4.3), for simulations from the mixture logistic model (blue) and the mixture Hüsler–Reiss model (orange), both perturbed as in (4.2). In both cases, a mixture logistic model is fitted using the estimation procedure in Algorithm 3. The penalization exponent is fixed to p = 0.4 and the tail fraction to k/n = 0.08. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: 31 stations from the Danube dataset clustered into K = 5 groups using an adapted version of the PAM algorithm with the pseudo-distance ˆd in (5.1) with tail fraction k/n = 0.1. Stations with the same color belong to the same cluster, and within each cluster, the station represented by a larger-sized shape indicates the cluster center. 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 fitted empirical 0.00 … view at source ↗
Figure 8
Figure 8. Figure 8: Empirical (5.1) versus fitted (5.3) extremal correlations for the results from the Danube dataset in Section 5.1 (left) and industry portfolios in Section 5.2 (right). 29 [PITH_FULL_IMAGE:figures/full_fig_p029_8.png] view at source ↗
read the original abstract

Estimating the parameters of max-stable parametric models poses significant challenges, particularly when some parameters lie on the boundary of the parameter space. This situation arises when a subset of variables exhibits extreme values simultaneously, while the remaining variables do not -- a phenomenon commonly referred to as an extreme direction. A novel estimator is proposed for the parameters of a general parametric mixture model, incorporating a threshold exceedances approach based on a pseudo-norm penalization. The latter plays a crucial role in accurately identifying parameters at the boundary of the parameter space. Additionally, the estimator comes with a data-driven algorithm to detect groups of variables corresponding to extreme directions. The performance of the estimator is assessed in terms of both parameter estimation and the identification of extreme directions through extensive simulation studies. Finally, the method is applied to two real-world datasets: discharge measurements at stations along the Danube river, and financial portfolio losses from stocks listed on the NYSE, AMEX, and NASDAQ. In both applications, the sets of variables that can become large simultaneously are identified.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a penalized least squares estimator for the parameters of a general parametric mixture model in the context of max-stable processes and extreme-value analysis. It employs a threshold-exceedances approach combined with a pseudo-norm penalization term that is intended to drive selected parameters exactly to the boundary of the parameter space (corresponding to non-extreme directions), while a separate data-driven algorithm identifies groups of variables that can exhibit simultaneous extremes. The estimator is evaluated through simulation studies assessing both parameter recovery and extreme-direction detection, and is illustrated on two real datasets: discharge measurements along the Danube river and financial portfolio losses from NYSE/AMEX/NASDAQ stocks.

Significance. If the pseudo-norm penalty successfully isolates boundary parameters without materially biasing the remaining estimates or the detection algorithm, the approach would address a practically important difficulty in fitting parametric extreme-value mixture models. The combination of simulation experiments and two distinct real-data applications (hydrology and finance) provides a reasonable empirical foundation; reproducible code or machine-checked proofs would further strengthen the contribution.

major comments (3)
  1. [§3] §3 (penalized objective): the pseudo-norm penalty is added directly to the least-squares loss constructed from threshold exceedances. Because the same objective is minimized jointly over all parameters, it is not immediate that the penalty forces boundary parameters exactly to zero while leaving non-boundary estimates asymptotically unbiased; an explicit oracle inequality or bias bound under the chosen penalty schedule is needed to support the central separation claim.
  2. [§4] §4 (simulation design): the reported success rates for extreme-direction detection are obtained after joint optimization of the penalized criterion. Without a side-by-side comparison to the unpenalized least-squares estimator (or to an oracle version that knows the boundary set in advance), it remains unclear whether the data-driven detection algorithm inherits bias from the penalty term or whether the observed performance is driven primarily by the threshold-exceedance likelihood.
  3. [§5] §5 (real-data applications): the Danube and stock-portfolio analyses identify sets of variables that can become large simultaneously, yet the manuscript does not report sensitivity of these sets to the penalty tuning parameter or to the choice of threshold. A stability analysis or bootstrap assessment of the detected extreme directions would be required to confirm that the findings are not artifacts of the penalization.
minor comments (2)
  1. [Notation] The notation for the pseudo-norm should be explicitly contrasted with the usual L1 or L0 penalties used in the extremes literature to avoid terminological confusion.
  2. [Tables 1-3] In the simulation tables, standard errors or confidence intervals for the reported bias and detection rates should be included so that readers can judge whether differences across methods are statistically meaningful.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments. We address each major comment below and describe the revisions we will implement.

read point-by-point responses

  1. Referee: [§3] §3 (penalized objective): the pseudo-norm penalty is added directly to the least-squares loss constructed from threshold exceedances. Because the same objective is minimized jointly over all parameters, it is not immediate that the penalty forces boundary parameters exactly to zero while leaving non-boundary estimates asymptotically unbiased; an explicit oracle inequality or bias bound under the chosen penalty schedule is needed to support the central separation claim.

    Authors: We agree that an explicit bias bound would strengthen the theoretical support for the separation property. The current manuscript establishes consistency of the estimator, but we will add a new proposition in Section 3 that derives a finite-sample bound on the estimation error for the non-boundary parameters under the chosen penalty schedule, showing that the bias term is of strictly lower order than the convergence rate when the penalty parameter is tuned appropriately. revision: yes

  2. Referee: [§4] §4 (simulation design): the reported success rates for extreme-direction detection are obtained after joint optimization of the penalized criterion. Without a side-by-side comparison to the unpenalized least-squares estimator (or to an oracle version that knows the boundary set in advance), it remains unclear whether the data-driven detection algorithm inherits bias from the penalty term or whether the observed performance is driven primarily by the threshold-exceedance likelihood.

    Authors: This is a fair criticism. We will expand the simulation section to include direct comparisons of extreme-direction detection rates and parameter MSE between the penalized estimator, the unpenalized least-squares estimator, and an oracle estimator that knows the true boundary set in advance. These additional results will clarify that the penalization improves detection accuracy while preserving the performance of the threshold-exceedance component. revision: yes

  3. Referee: [§5] §5 (real-data applications): the Danube and stock-portfolio analyses identify sets of variables that can become large simultaneously, yet the manuscript does not report sensitivity of these sets to the penalty tuning parameter or to the choice of threshold. A stability analysis or bootstrap assessment of the detected extreme directions would be required to confirm that the findings are not artifacts of the penalization.

    Authors: We accept this recommendation. In the revised applications section we will report the detected extreme directions for a range of penalty parameters and thresholds, together with a bootstrap stability assessment (resampling the exceedances) that quantifies the variability of the identified groups for both the Danube and financial datasets. revision: yes

Circularity Check

0 steps flagged

No significant circularity; estimator and detection algorithm are independently validated

full rationale

The paper defines a penalized least-squares estimator for max-stable mixture models that incorporates a pseudo-norm penalty to drive selected parameters to the boundary of the parameter space. This construction is presented as a novel methodological choice rather than a tautological re-expression of the data or of prior fitted quantities. Performance is assessed via separate simulation studies and real-data applications (Danube discharges and NYSE/AMEX/NASDAQ losses), which constitute external checks on bias, identification of extreme directions, and finite-sample behavior. No load-bearing step reduces by the paper's own equations to a self-citation, a fitted input renamed as a prediction, or an ansatz smuggled through prior work by the same authors. The derivation therefore remains self-contained against the stated benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Based on the abstract alone, the method rests on standard extreme-value theory assumptions for threshold exceedances and mixture models plus an unspecified tuning parameter for the pseudo-norm penalty.

free parameters (1)
  • penalty tuning parameter
    The strength of the pseudo-norm penalization must be chosen or optimized; its specific selection rule is not visible in the abstract.
axioms (1)
  • domain assumption Observations follow a parametric extreme-value mixture model whose parameters may lie on the boundary when only a subset of variables are extreme.
    Invoked throughout the abstract as the modeling framework for which the estimator is derived.

pith-pipeline@v0.9.0 · 5705 in / 1172 out tokens · 44712 ms · 2026-05-19T09:24:37.864263+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    John Wiley & Sons, 2004

    Jan Beirlant, Yuri Goegebeur, Johan Segers, and Jozef L Teugels.Statistics of extremes: theory and applications, volume 558. John Wiley & Sons, 2004

    work page 2004

  2. [2]

    Bias- corrected estimation of stable tail dependence function.Journal of Multivariate Analysis, 143:453–466, 2016

    Jan Beirlant, Mikael Escobar-Bach, Yuri Goegebeur, and Armelle Guillou. Bias- corrected estimation of stable tail dependence function.Journal of Multivariate Analysis, 143:453–466, 2016. 23

    work page 2016

  3. [3]

    Package ‘mev’

    Leo Belzile, Jennifer L Wadsworth, Paul J Northrop, Scott D Grimshaw, Jin Zhang, Michael A Stephens, and Art B Owen. Package ‘mev’. Technical report, 2024

    work page 2024

  4. [4]

    Clustering of maxima: Spatial dependencies among heavy rainfall in France.Journal of climate, 26(20):7929–7937, 2013

    Elsa Bernard, Philippe Naveau, Mathieu Vrac, and Olivier Mestre. Clustering of maxima: Spatial dependencies among heavy rainfall in France.Journal of climate, 26(20):7929–7937, 2013

    work page 2013

  5. [5]

    Flood history of the Danube tributaries Lech and Isar in the Alpine foreland of Germany.Hydrological Sciences Journal, 51(5):784–798, 2006

    Oliver Böhm and K-F Wetzel. Flood history of the Danube tributaries Lech and Isar in the Alpine foreland of Germany.Hydrological Sciences Journal, 51(5):784–798, 2006

    work page 2006

  6. [6]

    Byrd, Peihuang Lu, Jorge Nocedal, and Ciyou Zhu

    Richard H. Byrd, Peihuang Lu, Jorge Nocedal, and Ciyou Zhu. A limited memory al- gorithm for bound constrained optimization.SIAM Journal on Scientific Computing, 16(5):1190–1208, 1995

    work page 1995

  7. [7]

    An introduction to statistical modeling of extreme values, volume 208

    Stuart Coles. An introduction to statistical modeling of extreme values, volume 208. Springer, 2001

    work page 2001

  8. [8]

    Springer, 2006

    Laurens De Haan and Ana Ferreira.Extreme value theory: an introduction, volume 3. Springer, 2006

    work page 2006

  9. [9]

    Exact simulation of max-stable processes

    Clément Dombry, Sebastian Engelke, and Marco Oesting. Exact simulation of max-stable processes. Biometrika, 103(2):303–317, 2016

    work page 2016

  10. [10]

    Best attainable rates of convergence for estimators of the stable tail dependence function.Journal of Multivariate Analysis, 64(1):25–46, 1998

    Holger Drees and Xin Huang. Best attainable rates of convergence for estimators of the stable tail dependence function.Journal of Multivariate Analysis, 64(1):25–46, 1998

    work page 1998

  11. [11]

    An M-estimator for tail dependence in arbitrary dimensions

    John HJ Einmahl, Andrea Krajina, and Johan Segers. An M-estimator for tail dependence in arbitrary dimensions. The Annals of Statistics, 40(3):1764–1793, 2012

    work page 2012

  12. [12]

    A continuous updating weighted least squares estimator of tail dependence in high dimensions.Extremes, 21:205–233, 2018

    John HJ Einmahl, Anna Kiriliouk, and Johan Segers. A continuous updating weighted least squares estimator of tail dependence in high dimensions.Extremes, 21:205–233, 2018

    work page 2018

  13. [13]

    Package ‘graphicalExtremes’

    Sebastian Engelke, Adrien S Hitz, Nicola Gnecco, and Manuel Hentschel. Package ‘graphicalExtremes’. 2024

    work page 2024

  14. [14]

    Extremes of structural causal models

    Sebastian Engelke, Nicola Gnecco, and Frank Röttger. Extremes of structural causal models. arXiv preprint arXiv:2503.06536, 2025

    work page arXiv 2025

  15. [15]

    Dense classes of multivariate extreme value distributions

    Anne-Laure Fougères, Cécile Mercadier, and John P Nolan. Dense classes of multivariate extreme value distributions. Journal of Multivariate Analysis, 116: 109–129, 2013

    work page 2013

  16. [16]

    John Wiley & Sons, 2019

    Christian Francq and Jean-Michel Zakoian.GARCH models: structure, statistical inference and financial applications. John Wiley & Sons, 2019. 24

    work page 2019

  17. [17]

    Sparse representation of multivariateextremeswithapplicationstoanomalydetection

    Nicolas Goix, Anne Sabourin, and Stéphan Clémençon. Sparse representation of multivariateextremeswithapplicationstoanomalydetection. Journal of Multivariate Analysis, 161:12–31, 2017

    work page 2017

  18. [18]

    Composite likelihood estimation for the Brown–Resnick process.Biometrika, 100(2):511–518, 2013

    Raphaël Huser and Anthony C Davison. Composite likelihood estimation for the Brown–Resnick process.Biometrika, 100(2):511–518, 2013

    work page 2013

  19. [19]

    Maxima of normal random vectors: Between independence and complete dependence

    Jürg Hüsler and Rolf-Dieter Reiss. Maxima of normal random vectors: Between independence and complete dependence. Statistics and Probability Letters, 7(4): 283–286, 1989

    work page 1989

  20. [20]

    Rousseeuw.Finding groups in data: an introduction to cluster analysis

    Leonard Kaufman and Peter J. Rousseeuw.Finding groups in data: an introduction to cluster analysis. John Wiley & Sons, 2009

    work page 2009

  21. [21]

    Vignette for the tailDepFun package

    Anna Kiriliouk. Vignette for the tailDepFun package. Technical report, 2016

    work page 2016

  22. [22]

    Hypothesis testing for tail dependence parameters on the boundary of the parameter space.Econometrics and Statistics, 16:121–135, 2020

    Anna Kiriliouk. Hypothesis testing for tail dependence parameters on the boundary of the parameter space.Econometrics and Statistics, 16:121–135, 2020

    work page 2020

  23. [23]

    Climate extreme event attribution using multivariate peaks-over-thresholds modeling and counterfactual theory.The Annals of Applied Statistics, 2020

    Anna Kiriliouk and Philippe Naveau. Climate extreme event attribution using multivariate peaks-over-thresholds modeling and counterfactual theory.The Annals of Applied Statistics, 2020

    work page 2020

  24. [24]

    An estimator of the stable tail dependence function based on the empirical beta copula.Extremes, 21:581–600, 2018

    Anna Kiriliouk, Johan Segers, and Laleh Tafakori. An estimator of the stable tail dependence function based on the empirical beta copula.Extremes, 21:581–600, 2018

    work page 2018

  25. [25]

    Neural networks for parameter estimation in intractable models.Computational Statistics & Data Analysis, 185:107762, 2023

    Amanda Lenzi, Julie Bessac, Johann Rudi, and Michael L Stein. Neural networks for parameter estimation in intractable models.Computational Statistics & Data Analysis, 185:107762, 2023

    work page 2023

  26. [26]

    Multivariate sparse clustering for extremes

    Nicolas Meyer and Olivier Wintenberger. Multivariate sparse clustering for extremes. Journal of the American Statistical Association, 119(forthcoming):1–23, 2023

    work page 2023

  27. [27]

    Multivariate generalized Pareto distributions along extreme directions.Extremes, pages 1–34, 2024

    Anas Mourahib, Anna Kiriliouk, and Johan Segers. Multivariate generalized Pareto distributions along extreme directions.Extremes, pages 1–34, 2024

    work page 2024

  28. [28]

    Likelihood-based inference for max-stable processes.Journal of the American Statistical Association, 105(489): 263–277, 2010

    Simone A Padoan, Mathieu Ribatet, and Scott A Sisson. Likelihood-based inference for max-stable processes.Journal of the American Statistical Association, 105(489): 263–277, 2010

    work page 2010

  29. [29]

    Neural bayes estimators for censored inference with peaks-over-threshold models

    Jordan Richards, Matthew Sainsbury-Dale, Andrew Zammit-Mangion, and Raphaël Huser. Neural bayes estimators for censored inference with peaks-over-threshold models. Journal of Machine Learning Research, 25(390):1–49, 2024

    work page 2024

  30. [30]

    Multivariate generalized Pareto distributions

    Holger Rootzén and Nader Tajvidi. Multivariate generalized Pareto distributions. Bernoulli, 12(5):917–930, 2006. 25

    work page 2006

  31. [31]

    Inequalities for the extremal coefficients of multivariate extreme value distributions.Extremes, 5:87–102, 2002

    Martin Schlather and Jonathan Tawn. Inequalities for the extremal coefficients of multivariate extreme value distributions.Extremes, 5:87–102, 2002

    work page 2002

  32. [32]

    Determining the dependence structure of multivariate extremes.Biometrika, 107(3):513–532, 2020

    Emma S Simpson, Jennifer L Wadsworth, and Jonathan A Tawn. Determining the dependence structure of multivariate extremes.Biometrika, 107(3):513–532, 2020

    work page 2020

  33. [33]

    Simulating multivariate extreme value distributions of logistic type

    Alec Stephenson. Simulating multivariate extreme value distributions of logistic type. Extremes, 6(1):49–59, 2003

    work page 2003

  34. [34]

    Jonathan A. Tawn. Bivariate extreme value theory: models and estimation. Biometrika, 75(3):397–415, 1988

    work page 1988

  35. [35]

    Modelling multivariate extreme value distributions.Biometrika, 77(2):245–253, 1990

    Jonathan A Tawn. Modelling multivariate extreme value distributions.Biometrika, 77(2):245–253, 1990

    work page 1990

  36. [36]

    Conditional sampling for spectrally discrete max-stable random fields.Advances in Applied Probability, 43(2):461–483, 2011

    Yizao Wang and Stilian A Stoev. Conditional sampling for spectrally discrete max-stable random fields.Advances in Applied Probability, 43(2):461–483, 2011. 26 0.0 0.1 0.2 0.3 0.4 0.5 0.6 tail fraction k/n ED−S 0.02 0.06 0.1 0.14 0.18 0.0 0.1 0.2 0.3 0.4 0.5 0.6 tail fraction k/n ED−S 0.02 0.06 0.1 0.14 0.18 0 1 2 3 4 5 tail fraction k/n SMSE 0.02 0.06 0.1...

    work page 2011