betaFeedback

WorksMathematics

The exact region of the Average Rectangular Volume and Spearman's rho, with an application to arachnitude

Abstract

The average rectangular volume (\ARV) of a bivariate aggregation function is the average of its volumes over all axis-parallel rectangles of the unit square; for a copula, it is the expected copula volume of the rectangle spanned by two independent uniform points. We determine the exact region of the pairs (\ARV(C),ρ(C)), where CC ranges over all bivariate copulas and ρρ is Spearman's rho. The region is compact, convex and symmetric, and its upper boundary consists of three analytically parametrized arcs. A previously proposed family of shuffles of Min generates one arc, but only on part of its range; the other two arcs are generated by singular copulas whose supports contain hyperbolic branches. The boundary is continuously differentiable away from its left endpoint \ARV=\frac{11}{120}, where it has a vertical tangent. The proof uses a folding transformation and explicit Kantorovich potentials for a family of optimal transport problems. Finally, we show that the arachnitude, a rank correlation used in actuarial practice, is an affine transformation of \ARV, so that the result also describes the joint region of Spearman's rho and arachnitude.

Introduction

In the theory of aggregation functions, the volume that a bivariate function assigns to a rectangle of the unit square measures the interaction between its two arguments on that rectangle; it is non-negative on every rectangle exactly when the function is supermodular (2-increasing), a property shared by copulas and by the functions that generate universal integrals [12, 13, 9]. Motivated by this role, the average rectangular volume (ARV) of an integrable function F:[0,1]2→RF : [0, 1]^2 \to\mathbb{R} was introduced in [3] as the average of the FF-volumes of all axis-parallel rectangles of the unit square. For supermodular aggregation functions it is a global index of the strength of the interaction between the two inputs; for a copula CC, it is the expected CC-volume of the rectangle spanned by two independent uniform points.

On the class of copulas, ARV behaves quite differently from the classical measures of concordance. It is invariant under reflections of one argument, so that it does not distinguish positive from negative dependence, and it satisfies 11120≤ARV⁡(C)≤215\frac{11}{120} \leq\operatorname{ARV}(C) \leq\frac{2}{15} [3 Proposition 3.8]. It is therefore natural to ask how much information ARV⁡\operatorname{ARV} carries about a measure of concordance such as Spearman’s rho.

For two functionals defined on copulas, this question is encoded in their exact region; determining such regions is often a challenging problem. A classical example is the relation between Kendall’s tau and Spearman’s rho, for which universal inequalities and sharp bounds have been established [6, 26]. Related bounds and exact-region problems for restricted classes of copulas and for other measures of association have been studied, for instance, in [14, 16, 17, 18, 28, 29, 10, 2, 15, 24].

Partial information on the (ARV⁡,ρ)(\operatorname{ARV},\rho) region was obtained in [3, 4]: the inequality ∣ρ(C)∣≤3ARV⁡(C)+35\lvert\rho(C)\rvert\leq3\operatorname{ARV}(C)+\frac{3}{5} holds for every copula, minimal ARV⁡\operatorname{ARV} forces ∣ρ(C)∣≤12\lvert\rho(C)\rvert\leq\frac{1}{2}, and a family of shuffles of Min traces two curves joining the extreme parts of the region, which were conjectured to be its boundary.

In this paper we give a complete description of the region (Theorem 4.1). Its upper boundary is the graph of a concave, strictly increasing function r∗r^* and consists of three analytically parametrized arcs. The shuffles of Min considered in [4] generate the rightmost arc only for 0≤a≤250 \leq a \leq\frac{2}{5}; for 25<a<12\frac{2}{5}<a<\frac{1}{2} the corresponding curve lies strictly inside the region. The remaining two arcs are generated by singular copulas whose supports contain hyperbola-type curves. The boundary is continuously differentiable away from its left endpoint, and

r∗(α)=12+4(α−11120)3/4+o((α−11120)3/4),α↓11120,r^*(\alpha)=\frac{1}{2}+4\left(\alpha-\frac{11}{120}\right)^{3/4}+o\left(\left(\alpha-\frac{11}{120}\right)^{3/4}\right),\qquad\alpha\downarrow\frac{11}{120},

so that the boundary leaves (11120,12)\left(\frac{11}{120},\frac{1}{2}\right) with a vertical tangent.

The (ARV⁡,ρ)(\operatorname{ARV},\rho) region is also of interest in practical applications. In § 5 we show that, for every copula CC and (U,V)∼C(U,V)\sim C,

45ARV⁡(C)−5=Corr⁡((2U−1)2,(2V−1)2),45\operatorname{ARV}(C)-5=\operatorname{Corr}\left((2U-1)^2,(2V-1)^2\right),

i.e. ARV is an affine transformation of the arachnitude. This index is used in actuarial practice to calibrate the dependence between risk drivers in internal capital models, since it measures whether two variables tend to be simultaneously extreme, in either direction, a feature that Spearman’s rho cannot detect [27, 1]; it is the second diagonal coefficient of the Legendre expansion of the copula density, Spearman’s rho being the first [20]. Consequently, Theorem 4.1 describes exactly which combinations of Spearman’s rho and arachnitude are compatible with some copula, and thus provides a consistency check when both coefficients are estimated from data or specified by expert judgment. For instance, a Spearman’s rho of 0.9 is compatible only with an arachnitude of at least 0.249.

The paper is organized as follows. § 2 recalls aggregation functions, copulas and ARV, introduces the folding transformation and establishes the basic geometry of the region. § 3 constructs the three copula families. § 4 states the main result and its consequences. § 5 establishes the link with arachnitude and discusses examples, while § 6 concludes the paper. The proof of the main result is deferred to Appendix § 11.

Generative AI tools were used in the research process, for the numerical exploration, for the construction of the extremal families and of the Kantorovich potentials, and for symbolic verifications; all arguments have been checked by the authors. Details are given in the declaration at the end of the paper.

Preliminary definitions and properties

We first recall the basic notions on aggregation functions and copulas and the definition of the average rectangular volume. A (bivariate) aggregation function is a function A:[0,1]2→[0,1]A : [0,1]^2 \to[0,1] that is non-decreasing in each argument and satisfies A(0,0)=0A(0,0)=0 and A(1,1)=1A(1,1)=1 [12]. For a function F:[0,1]2→RF : [0,1]^2 \to\mathbb{R} and a rectangle

R=[x1,y1]×[x2,y2],xi≤yi,R = [x_1,y_1] \times[x_2,y_2], \qquad x_i \le y_i,

the FF-volume of RR is

VF(R)=F(y1,y2)−F(x1,y2)−F(y1,x2)+F(x1,x2).V_F(R) = F(y_1,y_2) - F(x_1,y_2) - F(y_1,x_2) + F(x_1,x_2).

FF is supermodular (or 2-increasing) if VF(R)≥0V_F(R) \ge0 for every rectangle RR. A (bivariate) copula is a supermodular aggregation function with uniform margins, C(u,1)=uC(u,1)=u and C(1,v)=vC(1,v)=v for u,v∈[0,1]u,v \in[0,1] [22, 9]. Let C\mathcal{C} denote the class of bivariate copulas and let

M(u,v)=min⁡{u,v},W(u,v)=max⁡{u+v−1,0},Π(u,v)=uv.M(u,v) = \min\{u,v\}, \qquad W(u,v) = \max\{u+v-1,0\}, \qquad\Pi(u,v) = uv.

We identify a copula with the probability measure it induces on [0,1]2[0,1]^2. We write (U,V)∼C(U,V) \sim C to mean that CC is the joint distribution function of the random pair (U,V)(U,V), i.e. C(u,v)=P(U≤u,V≤v)C(u,v)=\mathbb{P}(U \le u,V \le v); in particular, UU and VV are uniformly distributed on [0,1][0,1]. In this case VC(R)=P((U,V)∈R)V_C(R)=\mathbb{P}((U,V) \in R) for every closed rectangle RR, since the margins are continuous. Spearman’s rho is

ρ(C)=12E[UV]−3=12∫[0,1]2C(u,v) du dv−3.\rho(C) = 12\mathbb{E}[UV] - 3 = 12 \int_{[0,1]^2} C(u,v)\,\mathrm{d}u\,\mathrm{d}v - 3.

For x≤y\mathbf{x} \le\mathbf{y} in [0,1]2[0,1]^2 (componentwise), write [x,y]=[x1,y1]×[x2,y2][\mathbf{x},\mathbf{y}] = [x_1,y_1] \times[x_2,y_2].

Definition 2.1 ([3]). The average rectangular volume of an integrable function F:[0,1]2→RF : [0,1]^2 \to\mathbb{R} is

ARV⁡(F)=1∣R4∣∫R4VF([x,y]) dx dy,\operatorname{ARV}(F) = \frac{1}{|\mathcal{R}_4|} \int_{\mathcal{R}_4} V_F([\mathbf{x},\mathbf{y}])\,\mathrm{d}\mathbf{x}\,\mathrm{d}\mathbf{y},

where R4={(x,y)∈[0,1]2×[0,1]2:x≤y}\mathcal{R}_4 = \{(\mathbf{x},\mathbf{y}) \in[0,1]^2 \times[0,1]^2 : \mathbf{x} \le\mathbf{y}\}.

By Fubini’s theorem [3],

ARV⁡(F)=4∫[0,1]2F(u,v)(1−2u)(1−2v) du dv.(1)\operatorname{ARV}(F) = 4 \int_{[0,1]^2} F(u,v)(1-2u)(1-2v)\,\mathrm{d}u\,\mathrm{d}v. \tag*{(1)}

For a copula, the normalized integral over R4\mathcal{R}_4 equals, by symmetry, the expected CC-volume of the rectangle RX,YR_{\mathbf{X},\mathbf{Y}} spanned by two independent uniform points X,Y\mathbf{X},\mathbf{Y} of [0,1]2[0,1]^2 [3 Remark 2.2]. Since the interval spanned by two independent uniform points of [0,1][0,1] covers uu with probability 2u(1−u)2u(1-u), it holds that

ARV⁡(C)=P((U,V)∈RX,Y)=4E[U(1−U)V(1−V)],(2)\operatorname{ARV}(C) = \mathbb{P}\left((U,V) \in R_{\mathbf{X},\mathbf{Y}}\right) = 4\mathbb{E}[U(1-U)V(1-V)], \tag*{(2)}

where (U,V)∼C(U,V) \sim C is independent of (X,Y)(\mathbf{X},\mathbf{Y}). On C\mathcal{C}, ARV is affine, invariant under the reflection of one argument, and satisfies 11120≤ARV⁡(C)≤115\frac{11}{120} \le\operatorname{ARV}(C) \le\frac{1}{15} (see [3]); under independence, ARV⁡(Π)=19\operatorname{ARV}(\Pi)=\frac{1}{9}.

The main tool of this paper is a transformation that exploits the symmetry of ARV with respect to the center of the unit square. Rectangles centered at (12,12)\left(\frac{1}{2},\frac{1}{2}\right) play a special role: for (u,v)∈[0,12]2(u,v) \in\left[0,\frac{1}{2}\right]^2, let

Ru,v=[u,1−u]×[v,1−v]R_{u,v}=[u,1-u]\times[v,1-v]

be the rectangle centered at (1/2,1/2)(1/2,1/2).

Let (U,V)∼C(U,V)\sim C and define

S=∣2U−1∣,T=∣2V−1∣.S=|2U-1|,\qquad T=|2V-1|.

Since UU and VV are uniform on [0,1][0,1], both SS and TT are again uniform on [0,1][0,1].

Definition 2.2. For a given copula CC, the folded copula of CC, denoted by CFC^{\mathrm{F}}, is the copula of the random pair (S,T)=(∣2U−1∣,∣2V−1∣)(S,T)=(|2U-1|,|2V-1|).

Notice that this construction is related to the shuffle of copulas via suitable measure-preserving transformations of the unit interval (see [8]).

By flipping one coordinate we mean replacing (U,V)(U,V) by (1−U,V)(1-U,V) or by (U,1−V)(U,1-V). Such a flip does not change the folded copula; it preserves ARV and changes the sign of ρ\rho.

For a copula DD, we write ED\mathbb{E}_D for expectation with respect to (S,T)∼D(S,T)\sim D.

References

  1. [1]T. Androschuck et al., Simulation-based capital models: testing, justifying and communicating choices. A report from the Life Aggregation and Simulation Techniques Working Party, British Actuarial Journal 22(2) (2017) 257–335.
  2. [2]J. Ansari, M. Rockel, The exact region and an inequality between Chatterjee’s and Spearman’s rank correlations, J. Multivariate Anal. 214 (2026), 105630.DOI
  3. [3]L. Anzilli, F. Durante, The average rectangular volume induced by supermodular aggregation functions, J. Math. Anal. Appl. 555 (2026), 130137.DOI
  4. [4]L. Anzilli, F. Durante, On the (ARV, ρ) region for copulas, in: Soft Methods in Probability and Statistics for Complex Data, Proceedings of SMPS 2026, Springer, to appear.
  5. [5]C. M. Cuadras, J. Augé, A continuous general multivariate distribution and its properties, Comm. Statist. Theory Methods 10(4) (1981) 339–353.DOI
  6. [6]H. E. Daniels, Rank correlation and population models, J. R. Stat. Soc. Ser. B 12 (1950) 171–181.DOI
  7. [7]F. Durante, A. Kolesárová, R. Mesiar, C. Sempi, Semilinear copulas, Fuzzy Sets Syst. 159(1) (2008) 63–76.
  8. [8]F. Durante, P. Sarkoci, C. Sempi, Shuffles of copulas, J. Math. Anal. Appl. 352 (2009) 914–921.DOI
  9. [9]F. Durante, C. Sempi, Principles of Copula Theory, CRC/Chapman & Hall, Boca Raton, 2016.DOI
  10. [10]S. Fuchs, C. Limbach, F. Schürrer, On exact regions between measures of concordance and Chatterjee’s rank correlation for lower semilinear copulas, Internat. J. Approx. Reason. 189 (2026), 109588.arxiv.org/abs/2507.23316
  11. [11]W. Gangbo, R. J. McCann, The geometry of optimal transportation, Acta Math. 177 (1996) 113–161.DOI
  12. [12]M. Grabisch, J.-L. Marichal, R. Mesiar, E. Pap, Aggregation Functions, Cambridge University Press, Cambridge, 2009.
  13. [13]E. P. Klement, R. Mesiar, E. Pap, A universal integral as common frame for Choquet and Sugeno integral, IEEE Trans. Fuzzy Syst. 18 (2010) 178–187.DOI
  14. [14]D. Kokol Bukovšek, T. Košir, B. Mojškerc, M. Omladič, Spearman’s footrule and Gini’s gamma: local bounds for bivariate copulas and the exact region with respect to Blomqvist’s beta, J. Comput. Appl. Math. 390 (2021), 113385.DOI
  15. [15]D. Kokol Bukovšek, P. Lazić, B. Mojškerc, N. Stopar, The exact region determined by Spearman’s footrule, Gini’s gamma and Kendall’s tau, J. Comput. Appl. Math. 486 (2026), 117639.arxiv.org/abs/2601.07993
  16. [16]D. Kokol Bukovšek, B. Mojškerc, On the exact region determined by Spearman’s footrule and Gini’s gamma, J. Comput. Appl. Math. 410 (2022), 114212.DOI
  17. [17]D. Kokol Bukovšek, N. Stopar, On the exact regions determined by Kendall’s tau and other concordance measures, Mediterr. J. Math. 20(3) (2023), 147.
  18. [18]D. Kokol Bukovšek, N. Stopar, On the exact region determined by Spearman’s rho and Spearman’s footrule, J. Comput. Appl. Math. 437 (2024), 115463.DOI
  19. [19]H. O. Lancaster, The structure of bivariate distributions, Ann. Math. Statist. 29(3) (1958) 719–736.DOI
  20. [20]A. J. McNeil, J. G. Nešlehová, A. D. Smith, Measures and models of non-monotonic dependence, arXiv:2512.10828, 2025.arxiv.org/abs/2512.10828
  21. [21]P. Mikušinski, H. Sherwood, M. D. Taylor, Shuffles of Min, Stochastica 13 (1992) 61–74.
  22. [22]R. B. Nelsen, An Introduction to Copulas, 2nd ed., Springer, New York, 2006.
  23. [23]S. T. Rachev, L. Rüschendorf, Mass Transportation Problems. Vol. I: Theory, Springer, New York, 1998.
  24. [24]M. Rockel, The exact region between Chatterjee’s and Blest’s rank correlations, Internat. J. Approx. Reason. 197 (2026), 109744.
  25. [25]F. Santambrogio, Optimal Transport for Applied Mathematicians, Progress in Nonlinear Differential Equations and Their Applications 87, Birkhäuser, Cham, 2015.DOI
  26. [26]M. Schreyer, R. Paulin, W. Trutschnig, On the exact region determined by Kendall’s τ and Spearman’s ρ, J. R. Stat. Soc. Ser. B 79 (2017) 613–633.
  27. [27]R. A. Shaw, A. D. Smith, G. S. Spivak, Measurement and modelling of dependencies in economic capital, British Actuarial Journal 16(3) (2011) 601–699.DOI
  28. [28]M. Tschimpke, On the exact region determined by Spearman’s ρ and Blest’s measure of rank correlation ν for bivariate extreme-value copulas, J. Multivariate Anal. 205 (2025), 105377.DOI
  29. [29]M. Tschimpke, M. Schreyer, W. Trutschnig, Revisiting the region determined by Spearman’s ρ and Spearman’s footrule φ, J. Comput. Appl. Math. 457 (2025), 116259.DOI
  30. [30]C. Villani, Optimal Transport. Old and New, Grundlehren der mathematischen Wissenschaften 338, Springer, Berlin, 2009.

Paper details

vlnc:W_kz8hbndq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 1

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

  • Mathematics

Claude

View statement
Looking for related papers…
Contents