betaFeedback

WorksMathematics

Self-thresholding angular-radial models of extremal dependence

Abstract

Assessing compound risks requires a model for routine and extreme joint behaviour together, and a principled answer to where the joint tail begins. We propose the self-thresholding angular-radial (STAR) model: after marginal standardization, the radius given the angle follows a spliced distribution, with a flexible body below an angle-dependent threshold curve and a generalized Pareto tail above it. The threshold curve is estimated with all other parameters, so its uncertainty propagates to every summary, and the joint density is in closed form because the splice normalizes in one dimension for every angle. The tail scale function is the reciprocal gauge of the implied limit set, and the coefficient of extremal association is an explicit parameter, so a prior with atoms at tail independence and at the asymptotic dependence boundary yields posterior probabilities of the dependence class. At that boundary the copula of the model is asymptotically dependent, while its margins and joint tail are both lighter than exponential by a factor inversely proportional to the level. Simulations show accurate recovery of finite-level dependence summaries within and beyond the model class, and the slow separation of asymptotic dependence from strong asymptotic independence that the theory predicts. For warm-season daily maxima of fine particulate matter and ozone at a Beijing station, the pair is asymptotically independent with posterior probability 0.96 and both pollutants exceed their 0.9 quantiles on about two and a half days per season; the model gives summaries comparable to those of existing approaches without requiring the arbitrary choice of a threshold. The construction extends to d dimensions through pairwise deficits of the gauge, illustrated by adding coarse particulate matter as a third pollutant.

Introduction

Many questions about multivariate risk concern the body of a joint distribution and its tail at once: a flood defence can fail under combinations of river flow and sea level in which neither is record-breaking, and the impact of a heatwave depends on temperature and humidity together, when only one may be exceptional. Such questions require a model that describes the body and the joint tail within one framework, and a defensible answer to where the tail begins. In the univariate setting, extreme value mixture models combine a flexible body density with a generalized Pareto distribution (GPD) for threshold exceedances (Behrens et al., 2004; do Nascimento et al., 2012), with a parallel line embedding GPD tail behaviour in a single parametric family [30]; see Scarrott and MacDonald [40] and Naveau [29] for reviews. Their unifying idea is that the location of the tail is itself an object of inference. Carrying this idea to the multivariate setting raises a difficulty absent from the univariate case: the threshold is no longer a point but a boundary in Rd\mathbb{R}^d, and its shape encodes a decision about which observations count as jointly extreme, whether those in a rectangular region, those beyond a radial level, or those exceeding a more general risk functional. The visual diagnostics of univariate practice do not transfer [5], applied work defaults to a common marginal quantile for every variable [11], and the univariate threshold selection procedures of Northrop et al. [33] and Murphy et al. [26] have no multivariate counterpart.

Models for the body and tail of multivariate data follow three routes. The first blends dependence structures over the full support, through generalized Pareto copulas pieced onto a body copula [3], spliced margins with copula mixtures that deliver posterior statements about the dependence class [22], and dynamically weighted mixtures of a body and a tail copula [2]. These avoid a hard threshold, but because both components act on the full support the body influences the fitted tail and the tail remains sensitive to central observations (Hu et al., 2024). The second separates body and tail on disjoint regions, with the tail governed by the multivariate generalized Pareto distribution (mGPD) of Rootzén and Tajvidi [39]: Hu et al. (2024) combine a multivariate normal body with an mGPD above an estimated threshold vector, with a spatial analogue in Cuba et al. [10]. The separation removes the entanglement, at the price of a rectangular exceedance region, asymptotic dependence by construction, and a density discontinuous across the boundary. The third avoids a threshold altogether, extending the single-family construction of Naveau et al. [30] to several dimensions [1, 16], but offers neither likelihood-based Bayesian inference nor a boundary marking where asymptotic theory applies.

A separate and very active literature models the joint tail alone. The convergence of scaled sample clouds onto a limit set links its shape to the classical descriptions of extremal dependence [31, 32], and inference has followed for the limit set boundary [44, 41, 24], for radial generalized Pareto distributions and return sets [34] and for semiparametric angular-radial densities [23, 28], alongside the conditional approach of Heffernan and Tawn [18]. None models the distribution below the threshold, which is fixed in advance, typically at an empirical quantile, so threshold uncertainty is excluded.

This paper brings the two literatures together with the self-thresholding angular-radial (STAR) model, a generative angular-radial model that is self-thresholding in the tradition of Behrens et al. [4]: the radius conditional on the angle follows a spliced distribution, with a flexible body density below a threshold curve and a GPD above it. The threshold curve, an interpretable power-mean contour with the rectangle of Hu et al. (2024) as a limit, is estimated jointly with all other unknowns, so its uncertainty propagates into every derived quantity. Because the splice normalizes in one dimension for every angle, the joint density is in closed form. The fitted tail scale function is the reciprocal gauge of the limit set of Nolde and Wadsworth [32], so the residual tail dependence coefficient of Ledford and Tawn [20] and the angular dependence function of Wadsworth and Tawn [45] follow from their limit-set characterizations [31, 32]. Under the working scale family, the coefficient of extremal association of Coles et al. [9] is an explicit parameter, so that a three-component prior with atoms at tail independence and at the asymptotic dependence boundary returns posterior probabilities of the dependence class and a posterior for the strength of association. Beyond the limit set, we derive the exact joint and marginal tail decay of the model: both carry a factor of order 1/t1/t, so the copula is asymptotically dependent at the boundary, while on the standardized scale of the data χ(q)\chi(q) vanishes there logarithmically (Section 3.3); extensions with χ(q)\chi(q) bounded away from zero, through an explicit asymptotically dependent component, are discussed in § 7. In dd dimensions the splice and the threshold carry over, and the scale function subtracts pairwise deficits from the gauge of independence (§ 3.4); a fit to three pollutants shows what this adds to pairwise analyses and what remains open (§ 6.2). Simulations and an application to fine particulate matter and ozone in Beijing compare the model with the conditional extremes model Heffernan and Tawn, [18] and the geometric model Wadsworth and Campbell, [44].

A self-thresholding angular-radial model

Throughout, E=(E1,…,Ed)\mathbf{E}=(E_1,\ldots,E_d) denotes a random vector with standard exponential margins, obtained from the vector of interest Y\mathbf{Y} through Ej=−log⁡{1−Fj(Yj)}E_j=-\log\{1-F_j(Y_j)\} ([8]); estimation of the margins is treated in § 4.

Angular-radial decomposition

Dependence is modelled in pseudo-polar coordinates R=∥E∥1=∑j=1dEjR=\lVert\mathbf{E}\rVert_1=\sum_{j=1}^{d}E_j and W=E/R\mathbf{W}=\mathbf{E}/R, so that R>0R>0 and W\mathbf{W} lies in the unit simplex Sd−1={w∈[0,1]d:∑jwj=1}\mathcal{S}_{d-1}=\{\mathbf{w}\in[0,1]^d:\sum_j w_j=1\}. The map E↦(R,W)\mathbf{E}\mapsto(R,\mathbf{W}) has Jacobian rd−1r^{d-1}, so that any joint density factorizes as

fE(e)=fW(w)fR∣W(r∣w)r−(d−1),r=∥e∥1, w=e/r.(1)f_{\mathbf{E}}(\mathbf{e})=f_{\mathbf{W}}(\mathbf{w})f_{R\mid\mathbf{W}}(r\mid\mathbf{w})r^{-(d-1)},\qquad r=\lVert\mathbf{e}\rVert_1,\ \mathbf{w}=\mathbf{e}/r. \tag*{(1)}

The model specifies the two factors on the right of (1) directly: a flexible angular density fWf_{\mathbf{W}} describing the directions in which observations occur, and a conditional radial density fR∣Wf_{R\mid\mathbf{W}} describing how far observations travel in each direction. For clarity of exposition we present the bivariate case, identifying W\mathbf{W} with the scalar W=E1/R∈(0,1)W=E_1/R\in(0,1), and indicate the ingredients of the general case where they differ; § 3.4 gives the scale function and the dependence results in dd dimensions. This decomposition also underlies the SPAR model of Mackay and Jonathan [23] and the geometric inference of Wadsworth and Campbell [44].

The spliced radial model

Conditional on the angle, the radius follows a univariate spliced distribution, with every ingredient allowed to vary smoothly with the angle. Let ρ(w)>0\rho(w)>0 denote the threshold function, gb(⋅∣w)g_b(\cdot\mid w) a body density on (0,∞)(0,\infty) with distribution function Gb(⋅∣w)G_b(\cdot\mid w), and hGP(⋅;σ,ξ)h_{\mathrm{GP}}(\cdot;\sigma,\xi) the density of the generalized Pareto distribution with scale σ>0\sigma>0 and shape ξ\xi. The conditional radial density is

fR∣W(r∣w)={gb(r∣w),0<r≤ρ(w),{1−Gb(ρ(w)∣w)}hGP{r−ρ(w);σ(w),ξ(w)},r>ρ(w).(2)f_{R\mid W}(r\mid w)= \begin{cases} g_b(r\mid w), & 0<r\le\rho(w),\\ \{1-G_b(\rho(w)\mid w)\}h_{\mathrm{GP}}\{r-\rho(w);\sigma(w),\xi(w)\}, & r>\rho(w). \end{cases} \tag*{(2)}

The construction integrates to one for every angle, since the body contributes mass ϖ(w)=Gb(ρ(w)∣w)\varpi(w)=G_b(\rho(w)\mid w) and the tail the remaining 1−ϖ(w)1-\varpi(w). This is the central computational property of the model: a patchwork of body and tail densities separated by a curved boundary on the observation scale involves an intractable normalizing constant, the integral of the body density over the sub-threshold region, whereas here it collapses to the one-dimensional integral Gb(ρ(w)∣w)G_b(\rho(w)\mid w), available in closed form for standard body densities. The joint density follows from (1) with no further normalization. The density (2) is in general discontinuous at r=ρ(w)r=\rho(w); continuity could be enforced by tying σ(w)\sigma(w) to the body density at the threshold, as in Carreau and Bengio [7], but this would conflict with the structural scale (6). The discontinuity does not carry over to prediction: with a random threshold curve the posterior predictive density averages over ρ(⋅)\rho(\cdot) and is smooth (do Nascimento et al., 2012).

For the body we take a Gamma density with angle-dependent rate,

gb(r∣w)=λ(w)αΓ(α)rα−1exp⁡{−λ(w)r},(3)g_b(r \mid w) = \frac{\lambda(w)^\alpha}{\Gamma(\alpha)}r^{\alpha-1}\exp\{-\lambda(w)r\}, \tag*{(3)}

a choice motivated by two considerations. First, under independence in the bivariate case, R∼Gamma⁡(2,1)R \sim\operatorname{Gamma}(2,1) independently of W∼Uniform⁡(0,1)W \sim\operatorname{Uniform}(0,1), so with α=2\alpha= 2, a constant rate λ≡1\lambda\equiv1 and a uniform angular density the model reproduces independence exactly in the body and to first order in the tail, since the excess distribution of a Gamma converges to the unit-rate exponential as the threshold grows; in dimension dd the corresponding values are α=d\alpha= d and λ≡1\lambda\equiv1. Second, by the geometric theory of § 3, light-tailed sample clouds decay radially at direction-dependent exponential rates, which determine their limit set, introduced in § 2.4 [32]. Other body families, such as generalized Gamma densities or mixtures of Gamma densities (do Nascimento et al., 2012), fit in (2) without modification. Simulations with a body shape varying across the angle show that a constant shape suffices (Section S4.7).

Threshold curves

The threshold function ρ(⋅)\rho(\cdot) traces a curve in the positive orthant separating the body from the tail. Any positive function on the simplex defines a valid star-shaped boundary. We adopt a parametric family, and a nonparametric alternative would expand log⁡ρ(w)\log\rho(w) in the Bernstein basis of (5). For anchor levels u1,u2>0u_1, u_2 > 0 and exponent p>0p > 0, substituting x1=rwx_1 = rw and x2=r(1−w)x_2 = r(1-w) into the power-mean contour (x1/u1)p+(x2/u2)p=1(x_1/u_1)^p + (x_2/u_2)^p = 1 and solving for rr gives

ρ(w)={(wu1)p+(1−wu2)p}−1/p.(4)\rho(w) = \left\{\left(\frac{w}{u_1}\right)^p + \left(\frac{1-w}{u_2}\right)^p\right\}^{-1/p}. \tag*{(4)}

The exponent, estimated with all other parameters, controls the geometry of the extremal region (Figure 1): for p=1p = 1 the boundary is the straight line joining (u1,0)(u_1,0) and (0,u2)(0,u_2), for p=2p = 2 an elliptical arc, and values p∈(0,1)p \in(0,1) give boundaries concave towards the axes. As p→∞p \to\infty, ρ(w)→min⁡{u1/w,u2/(1−w)}\rho(w) \to\min\{u_1/w, u_2/(1-w)\}, whose exceedance region {x:x1>u1 or x2>u2}\{x : x_1 > u_1 \text{ or } x_2 > u_2\} is the rectangular region of Hu et al. (2024); above it they fit an mGPD, whereas (2) fits a generalized Pareto tail to the radius in each direction. We recommend p≥1p \ge1, as a concave curve can sag into the body along the diagonal and defeat the anchoring priors of § 4.2. In dd dimensions the family is ρ(w)={∑j(wj/uj)p}−1/p\rho(w) = \{\sum_j(w_j/u_j)^p\}^{-1/p}, with dd anchors and one exponent, and generalizes without change (Supplementary § 11.1).

Figure 1. Power-mean threshold curves (4), beyond which an observation is extreme, for anchors (a) u1=u2u_1 = u_2 and (b) u1>u2u_1 > u_2. The anchors fix where the curve meets the axes and the exponent its shape, a straight line at p=1p = 1 approaching the boundary of the componentwise region {x1>u1 or x2>u2}\{x_1 > u_1 \text{ or } x_2 > u_2\} as pp grows.

Angle-dependent body and tail parameters

The body rate is represented on the logarithmic scale in a low-order Bernstein basis,

log⁡λ(w)=∑k=0Kλbk(Kλk)wk(1−w)Kλ−k,(5)\log\lambda(w) = \sum_{k=0}^{K_\lambda} b_k \binom{K_\lambda}{k} w^k(1-w)^{K_\lambda-k}, \tag*{(5)}

which guarantees positivity and smoothness with interpretable coefficients.

The tail scale carries the extremal dependence of the model and is parametrized so that its strength appears as an explicit parameter. With τ(w)=min⁡(w,1−w)/max⁡(w,1−w)\tau(w) = \min(w,1-w)/\max(w,1-w), which increases from zero at the axes to one at the diagonal, we set

σ(w)=1+χˉτ(w)1+ν(w),χˉ∈[0,1],ν(w)=exp⁡{∑k=0Kνzk(Kνk)wk(1−w)Kν−k},(6)\sigma(w) = 1 + \bar{\chi}\tau(w)^{1+\nu(w)}, \qquad\bar{\chi} \in[0,1], \qquad\nu(w) = \exp\left\{\sum_{k=0}^{K_\nu} z_k \binom{K_\nu}{k} w^k(1-w)^{K_\nu-k}\right\}, \tag*{(6)}

References

  1. [1]N. Alotaibi, M. Sainsbury-Dale, P. Naveau, C. Gaetan, and R. Huser. Joint modeling of low and high extremes using a multivariate extended generalized Pareto distribution. arXiv preprint arXiv:2509.05982, 2025.arxiv.org/abs/2509.05982
  2. [2]L. M. André, J. L. Wadsworth, and A. O’Hagan. Joint modelling of the body and tail of bivariate data. Computational Statistics & Data Analysis, 189:107841, 2024.
  3. [3]S. Aulbach, V. Bayer, and M. Falk. A multivariate piecing-together approach with an application to operational loss data. Bernoulli, 18:455–475, 2012.
  4. [4]C. N. Behrens, H. F. Lopes, and D. Gamerman. Bayesian analysis of extreme events with threshold estimation. Statistical Modelling, 4:227–244, 2004.DOI
  5. [5]L. R. Belzile and A. C. Davison. Choosing the threshold in extreme value analysis. Brazilian Journal of Probability and Statistics, 2026. To appear; arXiv:2606.28540.arxiv.org/abs/2606.28540
  6. [6]R. Campbell and J. L. Wadsworth. Piecewise-linear modelling of multivariate geometric extremes. arXiv preprint arXiv:2412.05195, 2024.DOI
  7. [7]J. Carreau and Y. Bengio. A hybrid Pareto model for asymmetric fat-tailed data: the univariate case. Extremes, 12:53–76, 2009.DOI
  8. [8]S. Coles. An Introduction to Statistical Modeling of Extreme Values. Springer, London, 2001.DOI
  9. [9]S. Coles, J. Heffernan, and J. Tawn. Dependence measures for extreme value analyses. Extremes, 2:339–365, 1999.DOI
  10. [10]M. D. Cuba, M. Scott, B. P. Marchant, and D. Castro-Camilo. A bivariate spatial extreme mixture model for unreplicated heavy metal soil contamination. arXiv preprint arXiv:2402.14624, 2024.arxiv.org/abs/2402.14624
  11. [11]A. C. Davison and R. Huser. Statistics of extremes. Annual Review of Statistics and its Application, 2:203–235, 2015.
  12. [12]M. de Carvalho, M. Leonelli, and A. Rossi. Tracking change-points in multivariate extremes. arXiv preprint arXiv:2011.05067, 2020.arxiv.org/abs/2011.05067
  13. [14]C. A. T. Ferro and J. Segers. Inference for clusters of extreme values. Journal of the Royal Statistical Society: Series B, 65:545–556, 2003.DOI
  14. [15]W. Fulks and J. O. Sather. Asymptotics II: Laplace’s method for multiple integrals. Pacific Journal of Mathematics, 11:185–192, 1961.DOI
  15. [16]C. Gaetan and P. Naveau. Multivariate distributional modeling of low, moderate, and large intensities without threshold selection steps. arXiv preprint arXiv:2510.02152, 2025.arxiv.org/abs/2510.02152
  16. [17]T. E. Hanson, M. de Carvalho, and Y. Chen. Bernstein polynomial angular densities of multivariate extreme value distributions. Statistics & Probability Letters, 128:60–66, 2017.DOI
  17. [18]J. E. Heffernan and J. A. Tawn. A conditional approach for multivariate extreme values. Journal of the Royal Statistical Society: Series B, 66:497–546, 2004.
  18. [20]A. W. Ledford and J. A. Tawn. Statistics for near independence in multivariate extreme values. Biometrika, 83:169–187, 1996.DOI
  19. [21]S. Lee and J. L. Wadsworth. Geometric criteria for identifying extremal dependence and flexible modeling via additive mixtures. arXiv preprint arXiv:2512.24392, 2025.arxiv.org/abs/2512.24392
  20. [22]M. Leonelli and D. Gamerman. Semiparametric bivariate modelling with flexible extremal dependence. Statistics and Computing, 30:221–236, 2020.DOI
  21. [23]E. Mackay and P. Jonathan. Modelling multivariate extremes through angular-radial decomposition of the density function. arXiv preprint arXiv:2310.12711, 2023.arxiv.org/abs/2310.12711
  22. [24]R. Majumder, B. A. Shaby, B. J. Reich, and D. S. Cooley. Semiparametric estimation of the shape of the limiting bivariate point cloud. Bayesian Analysis, 21:1191–1217, 2026.DOI
  23. [25]T. J. Mitchell and J. J. Beauchamp. Bayesian variable selection in linear regression. Journal of the American Statistical Association, 83:1023–1032, 1988.DOI
  24. [26]C. Murphy, J. A. Tawn, and Z. Varty. Automated threshold selection and associated inference uncertainty for univariate extremes. Technometrics, 67:215–224, 2025.arxiv.org/abs/2310.17999
  25. [27]C. J. R. Murphy-Bartltrop, R. Majumder, and J. Richards. Deep learning of multivariate extremes via a geometric representation. arXiv preprint arXiv:2406.19936, 2024.
  26. [28]C. J. R. Murphy-Bartltrop, E. Mackay, and P. Jonathan. Inference for bivariate extremes via a semi-parametric angular-radial model. Extremes, 28:209–238, 2025.
  27. [29]P. Naveau. Jointly modeling the bulk and tails. In M. de Carvalho, R. Huser, P. Naveau, and B. Reich, editors, Handbook on Statistics of Extremes. Chapman & Hall/CRC, 2026.DOI
  28. [30]P. Naveau, R. Huser, P. Ribereau, and A. Hannart. Modeling jointly low, moderate, and heavy rainfall intensities without a threshold selection. Water Resources Research, 52:2753–2769, 2016.DOI
  29. [31]N. Nolde. Geometric interpretation of the residual dependence coefficient. Journal of Multivariate Analysis, 123:85–95, 2014.DOI
  30. [32]N. Nolde and J. L. Wadsworth. Linking representations for multivariate extremes via a limit set. Advances in Applied Probability, 54:688–717, 2022.arxiv.org/abs/2012.00990
  31. [33]P. J. Northrop, N. Attalides, and P. Jonathan. Cross-validatory extreme value threshold selection and uncertainty with application to ocean storm severity. Journal of the Royal Statistical Society Series C: Applied Statistics, 66(1):93–120, 2017.DOI
  32. [34]I. Papastathopoulos, L. de Monte, R. Campbell, and H. Rue. Statistical inference for radial generalized Pareto distributions and return sets in geometric extremes. arXiv preprint arXiv:2310.06130, 2023.arxiv.org/abs/2310.06130
  33. [35]M. Plummer. Cuts in Bayesian graphical models. Statistics and Computing, 25:37–43, 2015.DOI
  34. [36]S.-H. Poon, M. Rockinger, and J. A. Tawn. Extreme value dependence in financial markets: diagnostics, models, and financial implications. The Review of Financial Studies, 17(2):581–610, 2004.DOI
  35. [37]M. Ribatet, D. Cooley, and A. C. Davison. Bayesian inference from composite likelihoods, with an application to spatial extremes. Statistica Sinica, 22:813–845, 2012.
  36. [38]G. O. Roberts and J. S. Rosenthal. Examples of adaptive MCMC. Journal of Computational and Graphical Statistics, 18:349–367, 2009.DOI
  37. [39]H. Rootzén and N. Tajvidi. Multivariate generalized Pareto distributions. Bernoulli, 12:917–930, 2006.
  38. [40]C. Scarrott and A. MacDonald. A review of extreme value threshold estimation and uncertainty quantification. REVSTAT, 10:33–60, 2012.
  39. [41]E. S. Simpson and J. A. Tawn. Estimating the limiting shape of bivariate scaled sample clouds: with additional benefits of self-consistent inference for existing extremal dependence properties. Electronic Journal of Statistics, 18:4582–4611, 2024.DOI
  40. [42]E. S. Simpson and J. A. Tawn. Inference for new environmental contours using extreme value analysis. Journal of Agricultural, Biological and Environmental Statistics, 30:638–666, 2025.
  41. [43]E. S. Simpson, J. L. Wadsworth, and J. A. Tawn. Determining the dependence structure of multivariate extremes. Biometrika, 107:513–532, 2020.DOI
  42. [44]J. L. Wadsworth and R. Campbell. Statistical inference for multivariate extremes via a geometric approach. Journal of the Royal Statistical Society: Series B, 86:1243–1265, 2024.arxiv.org/abs/2208.14951
  43. [45]J. L. Wadsworth and J. A. Tawn. A new representation for multivariate tail probabilities. Bernoulli, 19:2689–2714, 2013.DOI
  44. [46]S. Zhang, B. Guo, A. Dong, J. He, Z. Xu, and S. X. Chen. Cautionary tales on air-quality improvement in Beijing. Proceedings of the Royal Society A, 473:20170457, 2017.DOI

Paper details

vlnc:W_pqje8pc7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents