On the pointwise supremum of the set of copulas with a given curvilinear section

Yao Ouyang a, Yonghui Sun a, Hua-peng Zhang b · 2024

Read On the pointwise supremum of the set of copulas with a given curvilinear section by Yao Ouyang a, Yonghui Sun a, Hua-peng Zhang b (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-20629.txt.

Description

Abstract Making use of the total variation of particular functions, we give an explicit formula for the pointwise supremum of the set of all copulas with a given curvilinear section. When the pointwise supremum is a copula is characterized. We also characterize the coincidence of the pointwise supremum and the greatest quasi-copula with the same curvilinear section.

Catalog record

CREATORYao Ouyang a, Yonghui Sun a, Hua-peng Zhang b
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE100 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-20629

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On the pointwise supremum of the set of copulas with a given curvilinear section Yao Ouyang a, Yonghui Sun a, Hua-peng Zhang b Abstract Making use of the total variation of particular functions, we give an explicit formula for the pointwise supremum of the set of all copulas with a given curvilinear section. When the pointwise supremum is a copula is characterized. We also characterize the coincidence of the pointwise supremum and the greatest quasi-copula with the same curvilinear section. Keywords: Copula, Quasi-copula, Curvilinear section, Total variation 1 Introduction In the realm of statistical modeling and data analysis, understanding and modeling the dependencies among variables is crucial for accurate predictions and informed decision-making. Traditional methods often rely on assumptions about the joint distribution of variables, which can be limiting and restrictive, especially when dealing with complex real-world data. Copula-based method [ 5 , 6 , 15 ] allows for the separate modeling of marginal distributions and the dependency structure among variables, providing a more nuanced and realistic representation of the underlying data-generating process. Thus it offers a flexible and versatile alternative to those traditional methods. Copulas, which were first introduced by Sklar in [ 14 ] , find uses in various domains such as finance and insurance. In finance, copulas have been instrumental in modeling the joint behavior of asset returns, risk management, and portfolio optimization [ 1 ] . In insurance, they have been used to assess the risk of simultaneous occurrences of extreme events [ 4 ] . Over the years, the investigation of (quasi-)copulas with a given diagonal section has attracted much attention [ 3 , 10 , 11 ] , since the diagonal section \delta_{C} provides some information about the tail dependence of a bivariate random vector and \delta_{C} itself is the distribution function of the random variable \max\{X,Y\} provided that the copula of (X,Y) is C . For a given diagonal section, the smallest copula always exists, while the greatest one may not. Recently, Kokol Bukovšek et al. [ 8 ] presented an explicit representation of the pointwise supremum of the set of all copulas with a given diagonal section. The study of (quasi-)copulas with a given diagonal section has been extended to that with a given curvilinear section [ 2 ] . For a given curvilinear section \Gamma_{\phi} , the set of all copulas (resp. quasi-copulas), denoted as \mathbb{C}_{\Gamma_{\phi}} (resp. \mathbb{Q}_{\Gamma_{\phi}} ), is nonempty. The smallest element of \mathbb{C}_{\Gamma_{\phi}} and that of \mathbb{Q}_{\Gamma_{\phi}} exist and coincide. Although the greatest element of \mathbb{Q}_{\Gamma_{\phi}} exists, denoted as A_{\Gamma_{\phi}} , the greatest element of \mathbb{C}_{\Gamma_{\phi}} may not. The present authors [ 12 ] characterized when A_{\Gamma_{\phi}} is a copula. However, the pointwise supremum of \mathbb{C}_{\Gamma_{\phi}} is unclear compared with the diagonal case. The central task of this paper is to explore an explicit formula for the pointwise supremum of \mathbb{C}_{\Gamma_{\phi}} . After recalling some necessary knowledge that will be used throughout the paper in Section 2 , we give an explicit formula for the pointwise supremum of \mathbb{C}_{\Gamma_{\phi}} in Section 3 . When the pointwise supremum of \mathbb{C}_{\Gamma_{\phi}} is a copula is characterized in Section 4 , while Section 5 is devoted to characterizing the coincidence of the pointwise supremum of \mathbb{C}_{\Gamma_{\phi}} and A_{\Gamma_{\phi}} . 2 Preliminaries In this section, we recall some basic notions and results concerning (quasi-)copulas with a given curvilinear section. Definition 2.1 . [ 9 ] A (bivariate) copula is a function C:[0,1]^{2}\to[0,1] satisfying: (C1) the boundary condition, i.e. , C(x,0)=C(0,x)=0\ {\rm and}\ C(x,1)=C(1,x)=x\ (\forall\,x\in[0,1]); (C2) the 2 -increasing property, i.e. , V_{C}(R)\geq 0 for any rectangle R=[x_{1},x_{2}]\times[y_{1},y_{2}] , where V_{C}(R)=C(x_{2},y_{2})-C(x_{2},y_{1})-C(x_{1},y_{2})+C(x_{1},y_{1}) . We will write V(\cdot) instead of V_{C}(\cdot) if there is no confusion. Definition 2.2 . [ 9 ] A (bivariate) quasi-copula is a function Q:[0,1]^{2}\to[0,1] satisfying (C1) and the following conditions: (Q1) Q is increasing in each variable; (Q2) Q is 1 -Lipschitz, i.e. , for all x_{1},x_{2},y_{1},y_{2}\in[0,1] , it holds that |Q(x_{1},y_{1})-Q(x_{2},y_{2})|\leq|x_{1}-x_{2}|+|y_{1}-y_{2}|. Note that (C1) and (C2) together imply both (Q1) and (Q2), i.e. , every copula is a quasi-copula, but not vice versa. Three important copulas are W,M and \Pi given by W(x,y)=\max\{x+y-1,0\},M(x,y)=\min\{x,y\}\ {\rm and}\ \Pi(x,y)=xy. W and M are known as the Fréchet-Hoeffding bounds for copulas and quasi-copulas since W\leq Q\leq M for any quasi-copula Q , while \Pi is known as the independence copula. Definition 2.3 . [ 10 ] A diagonal function is a function \delta:[0,1]\to[0,1] with the properties: (D1) \delta(1)=1 and \delta(t)\leq t for all t\in[0,1] ; (D2) 0\leq\delta(t_{2})-\delta(t_{1})\leq 2(t_{2}-t_{1}) for all t_{1},t_{2}\in[0,1] with t_{1}\leq t_{2} . The diagonal section \delta_{Q}:[0,1]\to[0,1] of any quasi-copula Q defined by \delta_{Q}(t)=Q(t,t) is a diagonal function. Conversely, for any diagonal function \delta , there exist quasi-copulas with diagonal section \delta , and the smallest quasi-copula B_{\delta} with diagonal section \delta is given by B_{\delta}(x,y)=\left\{\begin{array}[]{ll}y-\min\limits_{t\in[y,x]}\hat{\delta}(t)&\quad\text{if\ \ $y\leq x$}\\[5.69054pt] x-\min\limits_{t\in[x,y]}\hat{\delta}(t)&\quad\text{if\ \ $y>x$,}\end{array}\right. while the greatest quasi-copula A_{\delta} with diagonal section \delta is given by A_{\delta}(x,y)=\left\{\begin{array}[]{ll}\min\left\{y,x-\max\limits_{t\in[y,x]}\hat{\delta}(t)\right\}&\quad\text{if\ \ $y\leq x$}\\[5.69054pt] \min\left\{x,y-\max\limits_{t\in[x,y]}\hat{\delta}(t)\right\}&\quad\text{if\ \ $y>x$,}\end{array}\right. where \hat{\delta}:[0,1]\to[0,1] is defined by \hat{\delta}(t)=t-\delta(t) . Interestingly, B_{\delta} is also a copula (called the Bertino copula ), but A_{\delta} may not be. There have been characterization theorems for A_{\delta} to be a copula (see [ 10 , 11 ] ). Denote by \mathbb{C}_{\delta} the set of all copulas with given diagonal section \delta . The explicit formula of the pointwise supremum \bar{C}_{\delta} of \mathbb{C}_{\delta} has recently been given by Kokol Bukovšek et al. [ 8 ] : \bar{C}_{\delta}(x,y)=\left\{\begin{array}[]{ll}\min\left\{y,x-\dfrac{1}{2}\Big(\hat{\delta}(x)+\hat{\delta}(y)+\mathbb{V}_{y}^{x}(\hat{\delta})\Big)\right\}&\quad\text{if\ \ $y\leq x$}\\[5.69054pt] \min\left\{x,y-\dfrac{1}{2}\Big(\hat{\delta}(x)+\hat{\delta}(y)+\mathbb{V}_{x}^{y}(\hat{\delta})\Big)\right\}&\quad\text{if\ \ $y>x$,}\end{array}\right. where \mathbb{V}_{x}^{y}(\hat{\delta}) is the total variation of \hat{\delta} on [x,y] . A finite sequence \{x_{i}\}_{i=0}^{n}\subseteq[a,b] is called a partition of an interval [a,b] if it is increasing with x_{0}=a and x_{n}=b . Denote by \mathcal{P}[a,b] the set of all partitions of [a,b] . The total variation \mathbb{V}_{a}^{b}(f) of a real-valued function f on [a,b] is defined by \mathbb{V}_{a}^{b}(f)=\sup\left\{\sum_{k=1}^{n}|f(x_{k})-f(x_{k-1})|\mid\{x_{i}\}_{i=0}^{n}\in\mathcal{P}[a,b]\right\}. If \mathbb{V}_{a}^{b}(f)<+\infty , then f is differentiable almost everywhere on [a,b] and \mathbb{V}_{a}^{b}(f)\geq\int_{a}^{b}|f^{\prime}(t)|{\rm d}t , where the equality holds if and only if f is absolutely continuous [ 13 ] . We use the convention that \mathbb{V}_{b}^{a}(f)=-\mathbb{V}_{a}^{b}(f) and \mathbb{V}_{a}^{a}(f)=0 . Clearly, \mathbb{V}_{a}^{b}(f)=\mathbb{V}_{a}^{c}(f)+\mathbb{V}_{c}^{b}(f) for any c\in[a,b] , \mathbb{V}_{a}^{b}(f)=|f(b)-f(a)| if f is monotone, and \mathbb{V}_{a}^{b}(f+g)\leq\mathbb{V}_{a}^{b}(f)+\mathbb{V}_{a}^{b}(g) for any two real-valued functions f,g on [a,b] . Remark 2.4 . For any \{x_{i}\}_{i=0}^{n}\in\mathcal{P}[a,b] , it is easy to prove by induction that there exists \{y_{i}\}_{i=0}^{2m}\in\mathcal{P}[a,b] such that the sets \{x_{i}\}_{i=0}^{n} and \{y_{i}\}_{i=0}^{2m} are equal, and f(y_{2k-1})\geq\max\{f(y_{2k-2}),f(y_{2k})\} for all k\in\{1,2,\cdots,m\} . Hence, \displaystyle\sum_{k=1}^{n}|f(x_{k})-f(x_{k-1})| \displaystyle= \displaystyle\sum_{k=1}^{2m}|f(y_{k})-f(y_{k-1})| \displaystyle= \displaystyle\sum_{k=1}^{m}(f(y_{2k-1})-f(y_{2k-2}))+\sum_{k=1}^{m}(f(y_{2k-1})-f(y_{2k})) \displaystyle= \displaystyle 2\sum_{k=1}^{m}f(y_{2k-1})-2\sum_{k=1}^{m-1}f(y_{2k})-f(a)-f(b). Therefore, it holds that \mathbb{V}_{a}^{b}(f)=\sup\left\{2\sum_{k=1}^{m}f(x_{2k-1})-2\sum_{k=1}^{m-1}f(x_{2k})-f(a)-f(b)\mid\{x_{i}\}_{i=0}^{2m}\in\mathcal{P}[a,b]\right\}. The study of copulas with a given diagonal section has been extended to that with a given curvilinear section [ 2 ] . Let \Phi be the set of all functions \phi:[0,1]\to[0,1] that are continuous and strictly increasing with \phi(0)=0 and \phi(1)=1 . Obviously, {\rm id}\in\Phi , where {\rm id} is the identity map on [0,1] , i.e. , {\rm id}(x)=x for all x\in[0,1] . For any \phi\in\Phi , the curvilinear section g_{Q,\phi}:[0,1]\to[0,1] of a quasi-copula Q defined by g_{Q,\phi}(t)=Q(t,\phi(t)) has properties: (i) \max\{0,t+\phi(t)-1\}\leq g_{Q,\phi}(t)\leq\min\{t,\phi(t)\} for all t\in[0,1] ; (ii) 0\leq g_{Q,\phi}(t_{2})-g_{Q,\phi}(t_{1})\leq t_{2}-t_{1}+\phi(t_{2})-\phi(t_{1}) for all t_{1},t_{2}\in[0,1] with t_{1}<t_{2} . Note that property (ii) can be rewritten as: g_{Q,\phi}(t) is increasing and g_{Q,\phi}(t)-t-\phi(t) is decreasing. For any \phi\in\Phi , denote by \Delta_{\phi} the set of all functions \Gamma_{\phi}:[0,1]\to[0,1] with

Read the complete text