Multivariate strong subexponential distributions: properties and applications

Charalampos D. Passalidis · 2025

Read Multivariate strong subexponential distributions: properties and applications by Charalampos D. Passalidis (2025) 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-2503-22267.txt.

Introduction

Charalampos Passalidis studies heavy-tailed probability laws that live in more than one dimension, in a two thousand twenty-five math paper. He brings in a new class, the multivariate strong subexponential ones. Such laws fit rare, huge events, like extreme insurance claims. In them, one big jump can swamp the sum of all the rest.

He proves a few basic facts for the new class. One is a form of Kesten's inequality for this case. One says the class stays closed when sums are formed. He then shows the single big jump rule for sums stopped at a random time. The terms he needs are weaker than for the old class of this kind. He also finds sharp bounds for large moves away from the mean. A use in a nonstandard insurance risk model ends the work.

Catalog record

CREATORCharalampos D. Passalidis
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE85 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-22267

Browse the full library · About the Tower of Babel

Full text

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

Multivariate strong subexponential distributions: properties and applications Charalampos D. Passalidis Abstract . In this paper we introduce and study the classes of multivariate strong and strongly subexponential distributions. Some first properties are verified, as for example a type of multivariate analogue of Kesten’s inequality, the closure property with respect to convolution, and the conditional closure property with respect to convolution roots. Next, we establish the the single big jump principle for the randomly stopped sums, under the assumption that the random vectors in the summation belong to the class of multivariate strong subexponential distributions. Here the conditions of the counting random variable are weaker in comparison with them in multivariate subexponential class. Further, we establish uniform asymptotic estimates for the precise large deviations in multivariate set up, both for random and non-random sums, when the distribution of the summands belongs to the class of multivariate strongly subexponential distributions. Finally, we provide an application to a non-standard risk model, with independent and identically distributed claim vectors, from the class of multivariate strong subexponential distributions and in the presence of constant interest force. More concretely, the common counting process of the claim vectors constitutes from inter-arrival times, that are independent but not necessarily identically distributed. Under some additional condition, on the ’heavyness’ of the counting process tail, we establish a uniform asymptotic estimate for the finite time ruin probability in this model. Keywords: multivariate heavy-tailed distributions; convolution of random vectors; randomly stopped sums; vector type precise large deviations principle; nonstandard risk model; uniformity Mathematics Subject Classification : Primary 62P05 ; Secondary 60G70. 1. Introduction As a matter of fact the heavy tailed distributions in combination with the presence of dependence, represent a hot topic in the frame of applied probability. One way for modeling simultaneously these two concepts, is through the multivariate heavy tailed distributions. Such a way leads, through relatively simple and reasonable assumptions, to reach more general results, in relation with one-dimensional distributions under the presence of specific dependence structures. From the pioneer paper [ 30 ] the multivariate regular variation, symbolically MRV , attracted the interest of various topics of applied probability, as in extreme value theory, in time-series, in risk theory, in risk management and random walks, see [ 34 ] , [ 65 ] , [ 10 ] , [ 40 ] , [ 54 ] , [ 6 ] , [ 55 ] , [ 13 ] , among others. Although the MRV , mostly in its standard form, was functioning as antidote for facing the heavy-tailed distributions in combination with the dependence, in fact it remains enough restrictive for practical purposes, especially in actuarial mathematics, where just the regular variation of marginal distributions is too restrictive, see for example in [ 38 , p. 205, Fig. 7.2] . Thus, the study of other classes of heavy-tailed distributions in multivariate set up, seems unavoidable. For the multivariate subexponential distributions we know four different approaches, by [ 16 ] (via point processes and vague convergence), [ 62 ] , [ 66 ] and by [ 46 ] . From the first three approaches, this by [ 66 ] looks to be superior, since it has some good enough properties, that are inspired by the corresponding properties of stable random vectors. However, these first three classes show to approach the subexponentiality through a linear approach of multivariate single big jump principle, while the last one, seems to approach the subexponentiality through a non-linear type of multivariate single big jump principle. Therefore, we consider that the last two approaches are functioning in complementary mode, see in [ 43 , Sec. 5] for more details. In this paper we focus in the multivariate subexponentiality \mathcal{S}_{A} approach, suggested in [ 66 ] , see definition in Section 2 below. As in the uni-variate case, in many applications, as for example random walks, randomly stopped sums, queuing theory, risk theory among others, are needed assumptions slightly weaker of subexponentiality, the same happens in the multivariate case as well. More concretely in this paper we are interested in the presence of multivariate single big jump principle in the randomly stopped sums, and asymptotic equivalencies for the precise large deviations in multivariate set up. In order to study these two problems, we shall introduce the multivariate strong and strongly subexponential distributions, that belong to class \mathcal{S}_{A} . The paper is organized as follows. In Section 2, after providing the necessary preliminary concepts for the one-dimensional and multivariate heavy-tailed distributions, we present the multivariate classes of strong (and strongly) subexponential distributions. In Section 3, we examine some closure properties of these classes, like the closure with respect to convolution, and with respect to convolution roots of random vectors distributions. In Section 4, we are interested to randomly stopped sums \displaystyle{\bf S}_{\tau}=\sum_{i=1}^{\tau}{\bf X}^{(i)}\,, (1.1) where \tau represent a (one-dimensional) discrete random variable, and the \{{\bf X}^{(i)}\,,\;i\in{\mathbb{N}}\} , represent a sequence of independent identically distributed (i.i.d.) copies of the generic random vector {\bf X} . We deal with the single big jump principle for the multivariate randomly stopped sum, namely \displaystyle\mathbf{P}[{\bf S}_{\tau}\in x\,A]\sim\mathbf{E}[\tau]\,\mathbf{P}[{\bf X}\in x\,A]\,, (1.2) as x\to\infty , see in Section 2 for definition of the set A . Under the condition that {\bf X} has distribution from the class of multivariate subexponential distributions \mathcal{S}_{A} , was established relation ( 1.2 ) under the conditions \mathbf{E}[\tau]<\infty and \mathbf{E}[(1+\varepsilon)^{\tau}]<\infty , for some \varepsilon>0 , see in [ 44 , Th. 4.2] . Here, under the restriction of the distribution of {\bf X} , to belongs in the class of multivariate strong subexponential distributions, we establish relation ( 1.2 ), under the only requirement that \mathbf{E}[\tau]<\infty and that the \tau has distribution with lighter tail than F_{A} , see Section 2 for its definition. This result is reduced into corresponding one-dimensional results under a concrete form of set A , see for more details discussions in Section 4. In Section 5, we establish a vector type precise large deviation principle, for non-random and random sums, which are defined as follows \displaystyle{\bf S}_{n}=\sum_{i=1}^{n}{\bf X}^{(i)}\,,\qquad{\bf S}_{N(t)}=\sum_{i=1}^{N(t)}{\bf X}^{(i)}\,, (1.3) where \{N(t)\,,\;t\geq 0\} , is a counting process (one-dimensional). Under the condition that the \{{\bf X}^{(i)}\,,\;i\in{\mathbb{N}}\} are i.i.d. copies of {\bf X} , with multivariate strongly subexponential distribution and independent of the \{N(t)\,,\;t\geq 0\} , which satisfies some common assumptions in one-dimensional precise large deviations, we give the asymptotic equivalencies for the \displaystyle\mathbf{P}[{\bf S}_{n}\in x\,A]\,,\;\mathbf{P}[{\bf S}_{N(t)}\in x\,A]\,, where n\to\infty (or, t\to\infty , respectively), uniformly for any x greater than a quantity, connected with the inverse function of the insensitivity function of F_{A} , see for the definitions in Section 2. These results represent generalization even in the one-dimensional subcase. Finally, in Section 6, we present an application in a multivariate non-standard risk model. We provide uniform asymptotic estimations, with respect to time, for the entrance probability of the discounted aggregate claims in some rare-set, and for the ruin probability over a finite time horizon, in a risk model with common counting process for the d -lines of business and in the presence of constant interest force. The common counting process is NOT necessarily renewal one, since we assume that the inter-arrival times, are independent but not necessarily identically distributed. These results, are

Read the complete text