Technische Universität München Fakultät für Mathematik. Properties of extreme-value copulas

Size: px
Start display at page:

Download "Technische Universität München Fakultät für Mathematik. Properties of extreme-value copulas"

Transcription

1 Technische Universität München Fakultät für Mathematik Properties of extreme-value copulas Diplomarbeit von Patrick Eschenburg Themenstellerin: Betreuer: Prof. Claudia Czado, Ph.D. Eike Christian Brechmann Abgabetermin: 13. Mai 213

2 Hiermit erkläre ich, dass ich die Diplomarbeit selbständig angefertigt und nur die angegebenen Quellen und Hilfsmittel verwendet habe. München, den 13. Mai 213 i

3 Acknowledgments First of all, I would like to express my sincere gratitude to Prof. Claudia Czado for her excellent supervision of this thesis. I am very thankful for the helpful guidance that I received during the last months. Furthermore, I would like to thank my supervisor Eike Christian Brechmann for his great support, his commitment and many valuable ideas that helped to develop this thesis. Last, but certainly not least, I want to express my deepest gratitude to my parents for their unlimited support and helpful advice throughout my years of study. A deep thank-you also goes to my girlfriend Christina for her patience and encouragement during the last couple of months. ii

4 Contents List of Figures List of Tables List of Abbreviations and Symbols vi vii viii 1 Introduction 1 2 Preliminaries Copulas Tail dependence function Extreme-value copulas Representation of extreme-value copulas Bivariate extreme-value copulas Dependence measures Measures of association Tail dependence Families of bivariate extreme-value copulas Parametric extreme-value copula families Relationships between extreme-value copulas Asymmetry in extreme-value copulas Simulating from extreme-value copulas Asymmetry curve of extreme-value copulas Measure of non-exchangeability Estimation of the measure of non-exchangeability A note on the empirical copula process Generating asymmetric extreme-value copulas Non-parametric tests of symmetry Empirical measure of non-exchangeability A distribution-free test of symmetry Statistical inference on extreme-value copulas Parametric estimation Non-parametric estimation of the Pickands dependence function Estimators based on non-extreme bivariate samples Estimators based on extreme bivariate samples iii

5 iv Contents Moment estimators Weighted minimum distance estimation Corrections of the non-parametric estimators Extreme-value dependence tests Ghoudi, Khoudraji, and Rivest test for extreme-value dependence Extreme-value dependence test based on the copula process Comparison of the extreme-value dependence tests Model selection Graphical model selection Akaike and Bayesian information criteria Goodness-of-fit testing for extreme-value copulas Simulation study Non-parametric estimation of the Pickands dependence function Parametric estimation methods for extreme-value copulas Application Modeling scheme for extreme-value copulas Example Conclusion & Outlook 154 A Additional calculations 156 A.1 Derivatives of bivariate extreme-value copulas A.2 Proof of Theorem A.3 Derivation of the rank-based estimator in (5.73) B Additional figures 162 B.1 Similarity of extreme-value copulas B.2 Dependence measures of extreme-value copulas Bibliography 168

6 List of Figures 2.1 Illustration of an asymmetric copula Illustration of tail dependence Illustration of Example Tail dependence function and stable tail dependence function Pickands dependence function Connection between Pickands- and tail dependence function Illustration of the upper tail dependence coefficient Dependence function of the Marshall-Olkin copula Dependence function, scatter and contour plots of the Hüsler-Reiss copula Dependence function, scatter and contour plots of the t-ev copula Dependence function, scatter and contour plots of the Gumbel copula Dependence function, scatter and contour plots of the Tawn copula Dependence function, scatter and contour plots of the Galambos copula Dependence function, scatter and contour plots of the Joe copula Dependence function, scatter and contour plots of the BB5 copula Relationships between extreme-value copulas Similarity of the Pickands dependence functions of extreme-value copulas Pickands dependence functions with similar dependence characteristics Kendall s τ and λ u of the t-ev copula Relationship between τ and λ u of the t-ev copula Approximation of the t-ev copula Relationship between τ and λ u of the BB5 copula Relationship between τ and λ u of the symmetric Tawn copula Relationship between τ and λ u of the symmetric Joe copula Illustration of symmetry for the Gumbel copula Asymmetry curve of extreme-value copulas Illustration of Table Maximally non-exchangeable extreme-value copulas Estimation of the measure of non-exchangeability Power of the non-parametric symmetry test for n = 5 observations Power of the non-parametric symmetry test for n = 1 observations Power of the non-parametric symmetry test for n = 2 observations Theoretical functions h for the symmetry test Greatest convex minorant of non-parametric estimators v

7 vi List of Figures 6.1 Mean integrated squared errors of the non-parametric estimators Variation in parametric estimation methods for one-parameter families Variation in parametric estimation methods for two-parameter families Variation in parametric estimation methods for three-parameter families Modeling scheme for extreme-value copulas Scatter- and empirical contour plot of the pairs (Cs,Sc) and (Cs, Sc) Empirical density estimate of t for the pair (Cs, Sc) Fit of the estimated Pickands dependence functions Fit of the estimated Pickands dependence functions Scatter plot of the pair (Cs, Sc) with superimposed asymmetry curve. 152 B.1 Similarity of extreme-value copulas for τ = B.2 Similarity of extreme-value copulas for τ = B.3 Dependence measures of the Hüsler-Reiss copula B.4 Dependence measures of the t-ev copula B.5 Dependence measures of the Gumbel copula B.6 Dependence measures of the Tawn copula with θ = B.7 Dependence measures of the Tawn copula with θ = B.8 Dependence measures of the Galambos copula B.9 Dependence measures of the BB5 copula B.1 Dependence measures of the Joe copula with δ = B.11 Dependence measures of the Joe copula with δ =

8 List of Tables 3.1 Dependence coefficients of the families in Figure Pickands dependence functions of extreme-value copula families Approximation of the t-ev copula Approximation of mode{h(t)} Simulation study for empirical version of measure of non-exchangeability Parameter sets for the non-parametric symmetry test Rejection rates for the symmetry test for n = 5 observations Rejection rates for the symmetry test for n = 1 observations Rejection rates for the symmetry test for n = 2 observations Overview of non-parametric estimators of the Pickands dependence function Non-parametric estimators of the Pickands dependence function Overview of extreme-value dependence tests Mean integrated squared error of the non-parametric estimators Inversion of Blomqvist s β Comparison of parametric estimation methods P-values of the extreme-value dependence test for the pair (Cs, Sc) Estimated measures of the pair (Cs, Sc) Parametric estimation of the pair (Cs, Sc) vii

9 List of Abbreviations and Symbols BIPIT DA R Z Γ( ) Bivariate probability integral transform Domain of attraction Real numbers Integral numbers Gamma function X Random variable x Observation/realization of the random variable X x Sample mean, mean of the n observations of the random variable X n Sample size, number of observations of the random variable X (Û1, Û2) Vector of pseudo-observations X = (X 1,..., X d ) Vector of random variables x = (x 1,..., x d ) Vector of observations of the random variables X 1,..., X d d Dimension, number of variables F ( ) f( ) F 1 ( ),..., F d ( ) f 1 ( ),..., f d ( ) F ( ) Distribution function Density function Marginal distribution functions Marginal density functions Conditional distribution function E[ ] Expectation Var( ) Variance Cov(, ) Covariance Corr(, ) Correlation 1( ) Indicator function U[a, b] Uniform distribution on [a, b] Φ ρ ( ) Cdf of the standard normal distribution with correlation coefficient ρ Φ ( ) Inverse of the standard normal distribution function T ν ( ) Cdf of Student s t distribution with ν degrees of freedom C( ) Ĉ( ) C( ) Joint survival function Survival copula distribution function Copula distribution function viii

10 List of Abbreviations and Symbols ix c( ) C n ( ) C n C φ ( ) C κ1,κ 2 ( ) C φ,a ( ) Π d ( ) θ l( ) l ( ) l( ) K( ) A( ) Â n ( ) Â n,c ( ) Copula density function Empirical copula Empirical copula process Archimedean copula distribution function Copula generated by the device of Khoudraji Archimax copula distribution function d-dimensional independence copula Parameter or parameter vector of the copula C( ) Upper tail dependence function Stable tail dependence function Lower tail dependence function Kendall s distribution function Pickands dependence function Rank-based estimator of Pickands dependence function Corrected version of rank-based estimator of Pickands dependence function ˆρ n ρ S ˆτ τ β ˆβ n ˆλ u n λ u ˆλ l n λ l EV Empirical version of Spearman s ρ S Spearman s ρ Empirical version of Kendall s Tau Kendall s Tau Blomqvist s β Empirical version of Blomqvist s β Empirical version of upper tail dependence coefficient Upper tail dependence coefficient Empirical version of lower tail dependence coefficient Lower tail dependence coefficient Class of bivariate extreme-value copulas

11 Chapter 1 Introduction While the modeling of the dependence structure of random variables is one of the main topics in probability theory and statistics, the dependencies among extreme events have been of special interest lately, fuelled by events like the financial crisis or natural catastrophes. It was already pointed out by Coles and Tawn [1991] that most extreme events are in character multivariate. This means that extreme events are characterized by the joint occurrence of marginal extremes. As an example consider the financial sector, where substantial risks for companies arise if several extremal losses happen simultaneously. Consequently, the dependence of extremes has to play an important role in the modeling of joint extremes. In the classical approach of multivariate extreme-value theory, joint extremes are modeled by the asymptotic limit of the joint distribution of componentwise maxima of independent and identically distributed (i.i.d.) samples (Resnick [1987], Galambos [1987]). Joe [1997] investigates possible dependence structures of the limiting distributions, but a tremendous gain in flexibility is achieved if one uses copulas to model the dependence. Copulas provide a functional link between multivariate distribution functions and their univariate margins (Sklar [1959]). This allows to model the underlying dependence structure separately from the margins, a result which explains the growing popularity of these functions (see Nelsen [26] for a monograph on copulas). Extreme-value copulas combine the areas of multivariate extreme-value theory and copulas. They arise as asymptotic limits of copulas of component-wise maxima in i.i.d. samples and, thus, can be considered as appropriate models for the dependence structure between extreme events. However, compared to the research on copulas, the study of extreme-value copulas is quite behind. Recently, Gudendorf and Segers [21] have conducted a survey which grasps the most important aspects of extreme-value copulas and provides a bibliography for further literature. The aim of this thesis is to gather the literature available on extreme-value copulas and investigate their properties in detail to gain a comprehensive picture of this copula class. Therefore we will put special focus on statistical methods that are particularly related to the extreme-value copula class. The thesis is organized as follows. In Chapter 2, we will lay the theoretical foundation for this thesis. Basic concepts of copulas will be discussed. As mentioned, most results for modeling joint extremes have been derived in the context of multivariate extremes. We will combine these results with the concept of copulas and derive the representation of extreme-value copulas step by step. In the bivariate case, extreme- 1

12 2 Chapter 1 Introduction value copulas can be uniquely characterized by a one-dimensional mapping, called the Pickands dependence function. In Chapter 3, the most popular extreme-value copulas will be introduced and illustrated by means of this function. Further, connections between the different copula families will be highlighted. In contrast to some other copula classes (e.g. the popular Archimedean copula class), extreme-value copulas are capable of modeling asymmetry. In Chapter 4 we will discuss this aspect in detail and present a construction principle to generate asymmetric extreme-value copulas. Chapter 5 is concerned with statistical inference on extreme-value copulas. Besides estimation methods, tests for the hypothesis of an extreme-value dependence structure will be introduced, which should be considered before using this copula class. In Chapter 6 the estimation methods are compared in a Monte Carlo study. Chapter 7 discusses extreme-value copulas from a practical perspective. A summarizing modeling scheme for extreme-value copulas will be presented and illustrated by means of an example. Chapter 8 concludes, points out recent fields of research and potential fields of improvement for the future.

13 Chapter 2 Preliminaries This chapter provides the stochastic framework and basic definitions for the thesis. A sound understanding of the underlying theory is a crucial prerequisite before working with the matter. Therefore we will first build up the theory of extreme-value copulas in three constructive steps: after introducing the concept of copulas as the state-of-the-art methodology in dependence modeling, we will present tail dependence functions as an important tool, and finally give the common representation for the class of extremevalue copulas. 2.1 Copulas According to Genest and Favre [27], a main limitation of the traditional approach to multivariate dependence modeling is the fact that the individual behaviour of the margins has to be described by a single parametric family. As will become clear in the following section, working with copulas overcomes this limitation by splitting the analysis of the joint distribution of a multivariate random vector into two parts: the analysis of the margins - which is carried out independently for each random variable - and the choice of a copula which describes the dependence structure as a function of the margins. The copula thus provides a link (latin word: copula) between univariate margins and the joint distribution and implicitly defines a multivariate distribution function. Further it allows a margin-free analysis of multivariate dependence structures. We will first give a formal definition of copulas and then discuss some important properties. Main references for this chapter are Nelsen [26], Embrechts et al. [23] and Joe [1997]. As most of our work will be performed in two dimensions, we first define the bivariate copula. An extension to the higher dimensional case works in analogous fashion, however requires a little more theoretical background regarding multivariate distribution functions for which we refer to the given references. Definition 2.1 (Bivariate Copula) A bivariate (or two-dimensional) copula is a function C : [, 1] 2 [, 1] with the following properties: (i) For every u,v in [, 1] C(u, ) = = C(, v) 3

14 4 Chapter 2 Preliminaries and C(u, 1) = u and C(1, v) = v. (ii) C is 2-increasing, i.e. for every u 1, u 2, v 1, v 2 in [,1] s.t. u 1 u 2 and v 1 v 2, C(u 2, v 2 ) C(u 2, v 1 ) C(u 1, v 2 ) + C(u 1, v 1 ). Properties (i) and (ii) define a bivariate copula as a distribution function on [, 1] 2 with uniformly distributed margins (Nelsen [26], Embrechts et al. [23]). Thus a bivariate copula defines a bivariate distribution function on [, 1] 2 with uniformly distributed margins, i.e. for random variables U 1, U 2 C it holds that U 1 U[, 1] and U 2 U[, 1]. If the matter of interest is the survival probability of U 1, U 2 beyond points u 1, u 2, i.e. P (U 1 > u 1, U 2 > u 2 ), we have the following definition. Definition 2.2 (Joint survival function) Let U := (U 1, U 2 ) be a random vector with copula C. Then the joint survival function C of U is defined as C(u 1, u 2 ) = P (U 1 > u 1, U 2 > u 2 ). Note that this is not a copula. However there exists a survival copula Ĉ and the connection between the survival function C and the survival copula Ĉ is given by C(u 1, u 2 ) = Ĉ(1 u 1, 1 u 2 ). The ideas mentioned so far can be transferred to the multidimensional case as follows. Definition 2.3 (Copula) A d-dimensional copula is a function C : [, 1] d [, 1] with the following properties: (i) For every u := (u 1,..., u d ) [, 1] d and C(u) = if at least one coordinate of u is, if all coordinates of u are 1 except u k, k {1,..., d}, then C(u) = u k. (ii) C is d-increasing (for details see Nelsen [26], Embrechts et al. [23]). Besides the fact that a copula is a multivariate distribution function, the popularity in applications is mostly due to Sklar s Theorem (Sklar [1959]). This central theorem for copulas provides the link between joint distribution functions and their univariate margins and hence builds the foundation for most copula applications. Theorem 2.4 (Sklar) Let F be a d-dimensional distribution function with margins F 1,..., F d.then there exists a d-dimensional copula C such that for all x = (x 1,..., x d ) (R {, }) d : F (x 1,..., x d ) = C(F 1 (x 1 ),..., F d (x d )). If F 1,..., F d are all continuous, then the copula C is unique.

15 2.1 Copulas 5 The theorem illustrates the separation of a multivariate distribution function into univariate margins and the copula C, with C modeling the dependence structure. We will always consider continuous marginals in this thesis and hence assume the uniqueness of the respective copula. Precisely, consider the case of a random vector X := (X 1,..., X d ) with continuous margins F 1,..., F d and joint distribution function F. Then the distribution of the random vector X can be written as F (x 1,..., x d ) = P (X 1 x 1,..., X d x d ) = C(F 1 (x 1 ),..., F d (x d )). We say that C is the unique copula of X. Note that the margins of a copula have to be uniform. This is achieved by the probability integral transformation X i F i (X i ), formally defined below. Theorem 2.5 (Probability integral transform) Let X be a random variable with continuous distribution F. Then the random variable U = F(X) is uniformly distributed on (,1). Proof. F is a distribution function and thus non-decreasing. Denote by F : (, 1) (, ), F (u) := inf{x : F (x) u} the generalised inverse of F. Then P (U u) = P (F (X) u) = P (X F (u)) = F (F (u)) = u, < u < 1, which is the distribution function of a uniformly distributed random variable on (,1), so that U U(, 1). As a copula is a distribution function, it is possible to derive the copula density c which is needed in many applications, e.g. in maximum likelihood estimation. For a d-dimensional copula C it is defined as c(u 1,..., u d ) = d C(u 1,..., u d ) u 1... u d, (2.1) given that all partial derivatives exist. It follows from the above definition that if C is the unique copula of a d-dimensional random vector X F with continuous margins F i, i = 1,..., d, the probability density f of X is given by d f(x) = c(f 1 (x 1 ),..., F d (x d )) f i (x i ). So just as the joint distribution function, the joint density function can be separated into the marginal densities and the copula density. This motivates the following theorem. We will also give a proof of this theorem to illustrate some aspects of the theory presented so far. Theorem 2.6 Let X := (X 1,..., X d ) be a vector of continuous random variables. Then X 1,..., X d are independent if and only if C is the independence copula Π d (u) = u 1... u d. i=1

16 6 Chapter 2 Preliminaries Proof. We will carry out the proof in two steps. First, it has to be shown that the independence copula is indeed a copula. We will prove this in the bivariate case. Second, the statement itself will be proven. (1) For the independence copula Π 2 to be a bivariate copula, the requirements of Definition 2.1 have to be fulfilled. Property (i) follows directly from the definition of the independence copula, property (ii) can be calculated as Π 2 (u 2, v 2 ) Π 2 (u 2, v 1 ) Π 2 (u 1, v 2 )+ Π 2 (u 1, v 1 ) = u 2 v 2 u 2 v 1 u 1 v 2 +u 1 v 1 = u 2 (v 2 v 1 ) u 1 (v 2 v 1 ) = (u 2 u 1 )(v 2 v 1 ), where the last inequality follows from the assumptions u 2 u 1 and v 2 v 1. The same results hold in the multivariate case. (2) Assume that the X i are independent. Then F (x) = F 1 (x 1 )... F d (x d ). From Sklar (1959) we know that F can be expressed via a copula C as F (x) = C(F 1 (x 1 ),..., F d (x d )). Thus C(F 1 (x 1 ),..., F d (x d )) = F 1 (x 1 )... F d (x d ) C = Π d. The uniqueness of the solution is guaranteed by the continuity of the margins. C(u) = Π d (u) c(u) = π d (u) = d Π d (u) d = 1 f(x) = 1 f i (x i ) X i u 1... u d i=1 independent for i 1,..., d We now state a couple of important properties of copulas. Theorem 2.7 (Fréchet-Hoeffding bounds) Consider the following functions defined on the unit cube [, 1] d : M d (u) = min{u 1,..., u d }, (2.2) W d (u) = max{u u d d + 1, }. Then for any u [, 1] d and any d-dimensional copula C it holds that W d (u) C(u) M d (u). The function M d is a copula in any dimension d. In contrast W d is not a copula for d 3, but it still provides the best available lower boundary for C in any dimension (Nelsen [26], Theorem ). The functions M and W also have descriptive interpretations, as the copula M describes perfect co-monotonicity between the random variables while W corresponds to the case of counter-monotonicity. Together with the independence copula Π the stated functions cover the three extreme scenarios of possible dependence structures. Theorem 2.8 Let (X 1,..., X d ) be a vector of continuous random variables with copula C. If α 1,..., α d are strictly increasing functions, then (α 1 (X 1 ),..., α d (X d )) also has copula C. The fact that copulas are invariant under strictly increasing transformations of the random variables provides a useful tool in applications and further explains the popularity of copulas in multivariate dependence modelling. A question that will concern us in this thesis is the matter of symmetry of copulas. We will give a short definition here (Durante and Papini [29]), an exhaustive treatment of the topic will take place in Chapter 4. An illustration is given in Figure 2.1.

17 2.2 Tail dependence function U1 U U1 U2 Figure 2.1: Scatter plots of a symmetric copula (left) and an asymmetric copula (right). Definition 2.9 (Symmetric copula) Let X, Y be continuous random variables with copula C. X and Y are called exchangeable if and only if they are identically distributed, i.e. F X = F Y, and the copula C is symmetric, i.e. C(u, v) = C(v, u). 2.2 Tail dependence function Tail dependence is a measure of comovement in the tails of multivariate distribution functions, i.e. of joint extreme values of random variables. Informally spoken, it gives the probability of large (small) values of one random variable co-occuring with large (small) values of other random variables. The relevance to extreme value theory is thus obvious. In the following we will introduce the concept of tail dependence functions of Joe et al. [21]. Figure 2.2 gives a graphical motivation for the topic in the two-dimensional case. Let (U 1, U 2 ) C, where C is a bivariate copula. Scatter plots of two possible copula models are displayed in Figure 2.2. We are interested in comovements of extremely large (small) values, i.e. points in the far right upper (left lower) corner. The points in the boxes indicate that the left plot displays tail independence as there is no connection between large (or small) values of the respective variables while in the right plot, especially in the upper right corner, there is an obvious coexistence of large values of both variables. For a formal definition, let F be the distribution function of some random vector X = (X 1,..., X d ) with copula C. In analogy to Definition 2.2, where the bivariate case was considered, we denote by C the joint survival function of C which is given by C(u 1,..., u d ) = P (U 1 > u 1,..., U d > u d ) =: Ĉ(1 u 1,..., 1 u d ). (2.3)

18 8 Chapter 2 Preliminaries U1 U U1 U2 Figure 2.2: Scatter plots of the independence copula (left) and a copula displaying upper tail dependence (right). The function Ĉ is called the survival copula of C. The probability of joint occurrence of extreme values is then given by the tail dependence functions. Definition 2.1 (Tail dependence functions) The lower and upper tail dependence functions denoted by l and l, respectively, are defined as l(w, C) = lim s s 1 C(sw 1,..., sw d ) w = (w 1,..., w d ) R d + (2.4) l(w, C) = lim s s 1 C(1 sw 1,..., 1 sw d ) w = (w 1,..., w d ) R d + (2.5) Any result for one of the tail dependence functions can be transferred to the other one by (2.3), from which it follows that Ĉ(uw 1,..., uw d ) = P (1 U 1 u 1,..., 1 U d U d ) = C(1 uw 1,..., 1 uw d ) and as a consequence l(w, Ĉ) = l(w, C). Thus in the following we will focus on l(w, C) as upper tail dependence functions are of special relevance to extreme-value copulas. For notational ease we will simply speak of the tail dependence function and denote it by l whenever no risk of confusion arises. Proposition 2.11 (Properties of tail dependence functions) (l1) l is grounded, i.e. l(w) = if at least one w i =, i {1,..., d}. (l1) l is d-increasing, i.e. l(w 1,..., w i + ε i,..., w d ) l(w 1,..., w i,..., w d ) ε i, 1 i d. (l1) l is homogeneous of order one, i.e. l(sw) = sl(w) for any s.

19 2.2 Tail dependence function 9 (l1) l(w) = for all w if and only if l(k) = for some positive k = (k 1,..., k d ). A proof can be found in Joe et al. [21]. Note that properties (l1) and (l2) are very similar to those of copulas introduced in Section 2.1. We now turn our attention to an important aspect of Definition 2.1. The vector w defines the relative scale at which the components of the copula C go to extremes. This is best explained in an example. Example 1 Consider the bivariate case, i.e. w R 2 +. (1) Choose w 1 = w 2 =.5. Then l(.5,.5) = lim s s 1 C(s.5, s.5) = lim s s 1 P (U 1 1.5s, U 2 1.5s), i.e. the tail dependence function gives the probability that U 1 and U 2 take on extremely high values beyond a threshold lim(1.5s), while the threshold values s are identical. (2) Now choose w 1 w 2, for instance w 1 =.8, w 2 =.2. Then l(.8,.2) = lim s s 1 C(s.8, s.2) = lim s s 1 P (U 1 1.8s, U 2 1.2s). In this case the tail dependence function gives the joint probability of U 1 and U 2 exceeding different thresholds, since U 1 may take on lower values than U 2. This is illustrated in Figure 2.3. While technically the definition holds for s, s was first fixed at the values.4,.2 and.1 for better illustration (s means that the boxes move further into the upper right corner). For fixed s, the edges of the boxes correspond to the threshold values of U 1 and U 2 in the tail dependence function. The left plot shows case (1) where the boxes are symmetric as both threshold values are identical. In case (2), U 1 may take on smaller values than U 2, which is reflected in the asymmetric boxes in the right plot including a broader range of U 1 values. As s is continuous, repeating this procedure for a wide range of s values, here s [,.6], results in the lines through the corners of the boxes which display the threshold values for U 1 and U 2 for the respective values of s. Case (2) of Example 1 can be interpreted as measuring dependence along an axis other than the diagonal. This is an important extension which will be considered again in Chapter 4 during the investigation of asymmetry.

20 1 Chapter 2 Preliminaries U1 U U1 U2 Figure 2.3: Illustration of Example 1 for w = (.5,.5) (left) and w = (.8,.2) (right). 2.3 Extreme-value copulas Representation of extreme-value copulas In this section we will derive the most common representation of extreme-value copulas which is given via the so-called Pickands dependence function. Due to the results in Section 2.1, copulas are a growingly popular way of expressing classical distribution functions via their margins. However, the main theoretical results for this chapter were already provided by Haan and Resnick [1977] and Pickands [1981] in the context of multivariate extremes. They show that the class of extreme-value distributions coincides with the class of max-stable distributions and give a handy representation of this class using the tail dependence function. A transfer of this theory to the concept of copulas is given by Gudendorf and Segers [21], which we will mainly follow and complement here. Consider a sample of n independent d-dimensional random vectors X (i) := (X i1,..., X id ), i {1,..., n} with joint marginal distribution F, marginals F i and copula C F. Being interested in extremes we denote the vector of component-wise maxima by M n = (M n1,..., M nd ), where M nj = max i=1,...,n {X ij}. Note that M n will normally not be a sample point, as no observation takes on the extremes in all margins simultaneously. However, this is the vector that was historically considered for the investigation of multivariate extremes and many important statistical tools can be derived from the study of this vector (Beirlant et al. [24]).

21 2.3 Extreme-value copulas 11 Proposition 2.12 Let C F be the copula of the random vector X = (X 1,..., X d ). Then the copula C Mn of M n is given by C Mn (u 1,..., u d ) = C F (u 1/n 1,..., u 1/n d ) n, (u 1,..., u d ) [, 1] d. (2.6) Proof. First we show that the margins F n,i of M n are given by F n i : F ni (x i ) = P (M ni x i ) = P (X 1i x i,..., X di x i ) = P (X 1i x i )... P (X di x i ) = F i (x i ) n, where we use the independence of the sample. Thus we see that for u i = F ni (x i ) C Mn (u 1,..., u d ) = C Mn (F n1 (x 1 ),..., F nd (x d ) = P (M n1 x 1,..., M nd x d ) = P (X 11 x 1,..., X n1 x 1,..., X 1d x d,..., X nd x d ) = P (X 11 x 1,..., X 1d x d,..., X n1 x 1,..., X nd x d ) iid = P (X 11 x 1,..., X d1 x d ) n = C F (F 1 (x 1 ),..., F d (x d )) n = C F (F n,1 (x 1 ) 1/n,..., F nd (x d ) 1/n ) n = C F (u 1/n 1,..., u 1/n d ) n. We are interested in the asymptotic distribution of {M n }, that means we have to check if the copula C Mn converges to some copula C. If so, this limiting copula is called an extreme-value copulas. Definition 2.13 (Extreme-value copula) A copula C is called an extreme-value copula if there exists a copula C F such that, for n, C F (u 1/n 1,..., u 1/n d ) n C(u 1,..., u d ) (u 1,..., u d ) [, 1] d (2.7) The copula C F is said to be in the domain of attraction of the copula C (written C F DA(C) ). This definition formally introduces extreme-value copulas as asymptotic limits of componentwise maxima, however it does not provide a handy statistical tool. In order to be able to work with this class of copulas, a different representation is sought. By definition, extreme-value copulas are the copulas of extreme-value distributions. A major setback of multivariate extreme-value distributions in dimension d 2, however, is that the class of possible limiting dependence structures is infinite and therefore cannot be captured in a finite-dimensional parametric family (Beirlant et al. [24]). Thus unlike the univariate case, where the GEV (Generalised Extreme-Value) distribution covers the whole range of possible limit functions and provides a handsome tool with just three parameters, multivariate extreme-value distributions have to be indexed in another way. The most popular approach is closely linked to the concept of max-stability.

22 12 Chapter 2 Preliminaries Definition 2.14 (Max-stability of copulas) A copula C is called max-stable if it satisfies the relationship C(u 1,..., u d ) = C(u 1/m 1,..., u 1/m for every integer m 1 and (u 1,..., u d ) [, 1] d. This definition gives rise to the following theorem. d ) m Theorem 2.15 A copula C is an extreme-value copula if and only if it is max-stable. Proof. Let C Mn denote the copula of M n. We first rewrite C n as C Mn (u 1,..., u d ) (2.6) = C F (u 1/n 1,..., u 1/n d ) n ( ) = C F (u 1/mk 1,..., u 1/mk d ) mk = C F ((u 1/m 1 ) 1/k,..., (u 1/m ) 1/k ) mk d (2.6) = C Mk (u 1/m 1,..., u 1/m d ) m, (2.8) where in ( ) we used the substitution n = mk for fixed integer m and we let k in the last step. Now using the fact that C is an extreme-value copula it follows from Definition 2.13 that C(u 1,..., u d ) (2.7) = lim C F (u 1/n 1,..., u 1/n n d (2.6) = lim n C Mn (u 1,..., u d ) ) n (2.8) = lim C Mk (u 1/m 1,..., u 1/m d k = C(u 1/m 1,..., u 1/m d ) m. This follows directly from the definition of max-stability. Knowing the class of extreme-value distributions, we are interested in an applicable representation. Therefore we need the following definition. ) m Definition 2.16 (Max-id and max-stable distribution function) A distribution function F on R d is max-infinitely divisible (max-id) if for every t there exists a distribution F t on R d such that F = F t t, i.e. F 1/t is a distribution function. max-stable, if for every t > there exist functions a i (t) >, b i (t), i = 1,..., d such that F t (x) = F (a 1 (t)x 1 + b 1 (t),..., a d (t)x d + b d (t)),. (2.9)

23 2.3 Extreme-value copulas 13 For every max-id distribution function F it is further known that there exists a measure µ such that F (x) = exp( µ([, )\[, x)), x [, ], (2.1) (see Resnick [1987], Proposition 5.8). Due to this representation the measure µ is called exponent measure. It is clear from (2.9) that if F is max-stable, F t is a distribution function for every t > and therefore every max-stable distribution is also max-id. Thus every max-stable distribution can also be represented via an exponent measure. By definition, the set of copulas of max-stable distributions is exactly the class of maxstable copulas, which coincides with the class of extreme-value copulas according to Theorem We will see in the following theorem, which is found in Gudendorf and Segers [21], that the representation of extreme-value copulas is closely related to this representation via an exponent measure. Theorem 2.17 A d-variate copula C is an extreme-value copula if and only if there exists a finite Borel measure H on the unit simplex d 1 = {(u 1,..., u d ) [, ) d : j u j = 1}, called spectral measure, such that C(u 1,..., u d ) = exp{ l ( log u 1,..., log u d )}, (u 1,..., u d ) (, 1] d, (2.11) where the stable tail dependence function l : [, ) d [, ) is given by l (x 1,..., x d ) = max {w jx j }dh(w 1,..., w d ), (x 1,..., x d ) [, ) d. j=1,...,d d 1 The spectral measure H is arbitrary except for the d moment constraints w j dh(w 1,..., w d ) = 1, j {1,..., d}. d 1 Proof. The proof of this theorem is mainly based on the representation of max-stable distributions via spectral measures and requires transformation to polar coordinates. It can be found, for instance, in Resnick [1987] or Beirlant et al. [24]. To provide a more intuitive interpretation of the stable tail dependence function 1 other than via the spectral measure H, we use the following Proposition. Proposition 2.18 The domain of attraction equation (2.7) is equivalent to lim t t 1 (1 C F (1 tx 1,..., 1 tx d ) = l (x 1,..., x d ). 1 Note that the nomenclature of the terms tail dependence function and stable tail dependence function is very diverse in the literature, even in the sources cited in this section. For instance, Beirlant et al. [24] defines a different tail dependence function in a way that the stable tail dependence function arises as its limit for values in the far corners. In Joe et al. [21], the stable tail dependence function is called exponent function, expressing the use in the representation of extreme-value copulas via an exponent measure. In other articles, l is also referred to as upper tail copula (Haug et al. [211],Schmidt and Stadtmüller [26]).

24 14 Chapter 2 Preliminaries Proof. In (2.7), we rewrite the extreme-value copula C on the right-hand side in terms of its stable tail dependence function as in (2.11). This yields lim C F (u 1/n 1,..., u 1/n n d ) n = exp{ l ( log u 1,..., log u d )} lim log(c F (u 1/n 1,..., u 1/n n d ) n ) = l ( log u 1,..., log u d ) (1) lim n n(1 C F (exp{ 1 n log u 1},..., exp{ 1 n log u d})) = l ( log u 1,..., log u d ) (2) lim n n(1 C F (1 + 1 n log u 1,..., n log u d)) = l ( log u 1,..., log u d ) lim t t 1 (1 C F (1 tx 1,..., 1 tx d )) = l (x 1,..., x d ) (2.12) with the substitutions x i = log u i, i {1,..., d} and t = 1 in the last step. Further n we used the linear expansions of the logarithm and the exponential function in (1) : log(x) 1 x for x and in (2) : exp( 1 x) x for n. n n As 1 C F (1 u 1,..., 1 u d ) = P ( d i=1 {U i > 1 u i }), the stable tail dependence function l (x 1,..., x d ) is proportional to the limit lim P ( d i=1 {U i > 1 tx i }). Precisely, t recall from (2.12) that ( d ) l (x 1,..., x d ) = lim t 1 (1 C F (1 tx 1,..., 1 tx d )) = lim t 1 P {U i > 1 tx i }. t t i=1 (2.13) In contrast to that, the tail dependence function l as introduced in (2.5) is defined as l(x 1,..., x d ) = lim t t 1 C F (tx 1,..., tx d ) = lim t ( d ) t 1 P {U i > 1 tx i }, i=1 which expresses joint exceedance of the thresholds of all margins U i. This is graphically illustrated in Figure 2.4. A connection between the two functions is given by the inclusion-exclusion-principle and formulated in Joe et al. [21]: l (x 1,..., x d ) = ( 1) S 1 l(x 1,..., x j ; j S; C S ) S I,S for all non-empty subsets S I, where I := {1,..., d} denotes the index set and C S the copula of the S -dimensional margin {U i, i S}. For S = 1, define l(x i ) = x i. As an example, in the bivariate case this leads to the important relation l (x 1, x 2 ) = x 1 + x 2 l(x 1, x 2 ). (2.14) In the following we will state two important properties of the stable tail dependence function.

25 2.3 Extreme-value copulas U1 U U1 U2 Figure 2.4: Left panel: The stable tail dependence function l is proportional to the probability that at least one component is above a certain threshold. Right panel: For the tail dependence function l both components have to be large simultaneously. Proposition 2.19 The stable tail dependence function l as in (2.12) has the following properties. (i) l is homogeneous of order 1, i.e. l (ax) = al (x), a >, (x 1,..., x d ) [, ) d. (2.15) (ii) max{x 1,..., x d } l (x 1,..., x d ) x x d. Proof. (i) l (ax) (2.12) = lim t t 1 (1 C F (1 atx 1,..., 1 atx d )) = lim n n{1 C F (1 ax 1 n,..., 1 ax d n )} = lim s as{1 C F (1 ax 1 as,..., 1 ax d as )} = lim s as{1 C F (1 x 1 s,..., 1 x d s )} = lim t at 1 (1 C F (1 tx 1,..., 1 tx d )) (2.12) = al (x). (ii) From (2.13) it is known that l is proportional to the union of the probabilities P (U i > 1 tx i ), i = 1,..., d. As U U[, 1], each of these probabilities equals P (U i > 1 tx i ) = 1 P (U i 1 tx i ) = tx i.

26 16 Chapter 2 Preliminaries Thus we can establish the following boundaries: ( d ) max{tx 1,..., tx d } P {U i > 1 tx i } tx tx d. (2.16) i=1 Multiplying (2.16) by 1/t and letting t gives max{x 1,..., x d } l (x 1,..., x d ) x x d, (2.17) see also (2.13). The homogeneity property (i) in Proposition (2.19) allows to restrict the stable tail dependence function to the unit simplex d 1 := {(w 1,..., w d ) [, 1] d : w w d = 1}. This idea is originally suggested in Pickands [1981]. Therefore consider the real a := 1 x x d as a scaling factor in (2.15). The vector x := ax is then a normed version of x as it holds that x i := ax i = x i x x d [, 1], i = 1,..., d Thus one can define a function on the unit simplex d 1 by and d x i = 1. A( x) := l (ax) = al (x). (2.18) The function A(x) is the restriction of the stable tail dependence function to the unit simplex. That idea was originally formulated in Pickands [1981], which motivates the following definition. Definition 2.2 (Pickands dependence function) The Pickands dependence function A : d 1 [1/d, 1] is defined as A(w 1,..., w d ) := l (x 1,..., x d ) d i=1 x, where w j := i for (x 1,..., x d ) [, ) d \{}. i=1 x j d i=1 x, j = 1,..., d, (2.19) i Note that the variable w d = 1 w 1... w d 1 is often suppressed and A is written only as a function of (w 1,..., w d 1 ). The fact that the Pickands dependence function maps to [1/d, 1] follows from using (2.17) and (2.19). According to (2.17) it holds that max{x 1,..., x d } l (x 1,..., x d ) x x d max{x 1,..., x d } l (x 1,..., x d ) x x d x x }{{ d } A(w 1,...,w d 1 ) x x d x x d (2.2)

27 2.3 Extreme-value copulas 17 In case that all x i, i = 1..., d, are identical, the left-hand side in (2.2) equals 1/d. Otherwise it holds that the fraction is larger than 1/d, from which the claim follows. Definition 2.2 captures the connection between the Pickands dependence function and the stable tail dependence function. Together with (2.11) this gives rise to the most common representation of extreme-value copulas which are expressed via A as {( d ) ( C(u 1,..., u d ) = exp log u i A i=1 log u 1 d i=1 log u,..., i log u d 1 d i=1 log u i Proposition 2.21 (Properties of the Pickands dependence function) (A1) A is convex. (A2) max{w 1,..., w d 1 } A(w 1,..., w d 1 ) 1 (w 1,..., w d 1 ) d 1. )}. (2.21) Proof. A proof of Lemma 2.21 can be found in Joe [1997] (p.175). It is shown that (A2) has to hold in order for C in (2.21) to define a copula. Further it follows that in this case A, if it exists, is positive definite and therewith convex. In the next section we will discuss these properties in more detail in the important bivariate case Bivariate extreme-value copulas Most of the literature available on extreme-value copulas concentrates on the bivariate case. This has mainly two reasons. First, higher dimensional copulas of dimension d 3 have certain limitations (e.g. limited number of available families, lack of flexibility) and are therefore often modeled by pair-copula constructions built up by bivariate copulas (Aas et al. [29]). Second, for bivariate extreme-value copulas, the Pickands dependence function reduces to a one-dimensional mapping which completely characterizes C and facilitates many statistical procedures. In the bivariate case, the stable tail dependence function can be restricted to the unit simplex, which we parameterise following Gudendorf and Segers [21]: 1 = {(1 t, t) : t [, 1]}. (2.22) We know from Theorem 2.17 (precisely (2.11)) that we can express a bivariate extremevalue copula as C(u 1, u 2 ) = exp{ l ( log(u 1 ), log(u 2 ))}. Defining w 1 := 1 t and w 2 := t, according to (2.19) the Pickands dependence function is then defined as A(1 t, t) = l ( log(u 1 ), log(u 2 )) log(u 1 ) log(u 2 ) homog. = l (log(u 1 ), log(u 2 )) (log(u 1 ) + log(u 2 )) homog. = l ( log(u1 ) log(u 1 u 2 ), log(u 2 ) log(u 1 u 2 ) ). (2.23)

28 18 Chapter 2 Preliminaries As mentioned by Segers [212b], it is common to suppress one argument of the Pickands dependence function due to the restriction on the unit simplex. As it is more convenient to work with A(t) than A(1 t), we will work with where A(t) (2.23) = l ( log(u1 ) log(u 1 u 2 ), log(u 2 ) log(u 1 u 2 ) ) = l (1 t, t), (2.24) t = log(u 2) log(u 1 u 2 ). (2.25) We will now define bivariate extreme-copulas. Theorem 2.22 (Bivariate extreme-value copula) A bivariate copula C is an extreme-value copula if and only if { ( )} log(u2 ) C(u 1, u 2 ) = exp (log(u 1 ) + log(u 2 ))A log(u 1 u 2 ) = (u 1 u 2 ) A(log(u 2)/ log(u 1 u 2 )), (u 1, u 2 ) (, 1] 2 \{(1, 1)}, (2.26) where A : [, 1] [1/2, 1] is convex and satisfies max{1 t, t} A(t) 1 t [, 1]. Proof. Multivariate extreme-value copulas are defined in (2.11), which in the bivariate case reduces to C(u 1, u 2 ) = exp{ l ( log(u 1 ), log(u 2 ))}. (2.27) Using (2.23) we obtain (2.28) in (2.27) yields A(t) = l ( log(u 1 ), log(u 2 )) log(u 1 ) log(u 2 ) l ( log(u 1 ), log(u 2 )) = log(u 1 u 2 )A(t). (2.28) C(u 1, u 2 ) = exp{log(u 1 u 2 )A(t)}, where t = log(u 2 )/ log(u 1 u 2 ) according to (2.25). Note that this representation is not unique. By parameterising the unit simplex as 1 = {(t, 1 t) : t [, 1]}, one gets the argument log(u 1 )/ log(u 1 u 2 ) instead of log(u 2 )/ log(u 1 u 2 ) in (2.26). Several authors use this parametrization (Hürlimann [23], Abdous and Ghoudi [25]). This aspect should be considered when working with bivariate extreme-value copulas. Before discussing the properties of the Pickands dependence function A in more detail, we give the conditional copulas and the density for bivariate extreme-value copulas which we will need later, e.g. for contour plots of the parametric families or maximum likelihood estimation. The conditional copula of the random variable U 2 given U 1 = u 1 will be denoted by C 2 1 ( u 1 ) and is given by C 2 1 (u 2 u 1 ) := C(u 1, u 2 ) u 1 = C(u 1, u 2 ) u 1 (A(t) ta (t)), (2.29)

29 2.3 Extreme-value copulas 19 A(t) independence co monotonicity t Figure 2.5: Pickands dependence function: the admissible range is coloured in grey. The dashed upper line corresponds to the independence copula while the solid line denotes max{t, 1 t}, which corresponds to perfectly dependent variables. The dotted lines show two examples of typical dependence functions. where t = log(u 2 )/ log(u 1 u 2 ) and we assume that A exists. Equivalently, we get c(u 1, u 2 ) = C(u 1, u 2 ) u 1 u 2 C 1 2 (u 1 u 2 ) := C(u 1, u 2 ) u 2 = C(u 1, u 2 ) u 2 (A(t) + (1 t)a (t)). (2.3) Recall that the bivariate copula density c equals the first derivative of the copula with respect to both arguments (see (2.1)) and is thus given by [ ( )] A(t) 2 + (1 2t)A (t)a(t) (1 t)t A (t) 2 A (t), log(u 1 u 2 ) (2.31) where we again use t = log(u 2 )/ log(u 1 u 2 ) and assume the existence of A. For the parametric families discussed later this is always the case. A complete calculation of these expressions can be found in Appendix A.1. There it can also be seen that the derivation order does not play a role for the density as the Hessian matrix of the function t(u 1, u 2 ) := log(u 2 )/ log(u 1 u 2 ) is symmetric. The original definition of the Pickands dependence function (Pickands [1981]) was the following: A(t) = l (1 t, t), t [, 1], as in (2.24). By property (A2) in Proposition 2.21, a bivariate Pickands dependence function lies within the shaded area indicated in Figure 2.5. Further (A1) requires A to be convex. The solid and dashed lines in Figure 2.5 give examples of two possible Pickands dependence functions. 2.5 further illustrates the fact that the class of multivariate

30 2 Chapter 2 Preliminaries extreme-value copulas (and therewith extreme-value distributions) cannot be captured by a single parametric family. Instead any function A satisfying (A1) and (A2) is a possible Pickands dependence function. Consequently the class of bivariate extremevalue copulas is indexed by the set of convex functions defined on the unit interval. Still some important parametric families arise quite naturally and will be presented in Chapter 3. The boundaries of the Pickands dependence function A have special meanings: The upper line A(t) = 1 in Figure 2.5 corresponds to the case of independence: inserting A(t) = 1 into (2.26) yields C(u 1, u 2 ) = u 1 u 2 = Π 2 (u 1, u 2 ), the independence copula. The lower boundary A(t) = max{1 t, t} in Figure 2.5 corresponds to the case of perfect dependence: C(u 1, u 2 ) = max{u 1, u 2 } = M 2 (u 1, u 2 ). An interpretation of A is given as follows. It holds that A(t) = l (1 t, t) = 1 l(1 t, t) t [, 1]. (2.32) The argument of the Pickands dependence function thus corresponds to the argument (1 t, t) of the tail dependence functions. For l, this is exactly the vector w in (2.5). Recall that l measures the dependence along the axis (1 s(1 t), 1 st) in terms of the joint exceedance probability of both random variables. Informally spoken, the more points fall on the dependence axis or scatter around the upper right corner of the dependence axis, the higher the dependence. Now bearing in mind that A(t) = 1 corresponds to independence between the random variables, it follows from (2.32) that a higher dependence along an axis results in a smaller value of A. This is illustrated in Figure 2.6. The value A(1/2) is of special importance, as l(1, 1) measures the dependence along the diagonal. This interpretation also plays an important role in the analysis of asymmetry. In Chapter 3, popular parametric families of bivariate extreme-value copulas will be introduced.

A measure of radial asymmetry for bivariate copulas based on Sobolev norm

A measure of radial asymmetry for bivariate copulas based on Sobolev norm A measure of radial asymmetry for bivariate copulas based on Sobolev norm Ahmad Alikhani-Vafa Ali Dolati Abstract The modified Sobolev norm is used to construct an index for measuring the degree of radial

More information

Financial Econometrics and Volatility Models Copulas

Financial Econometrics and Volatility Models Copulas Financial Econometrics and Volatility Models Copulas Eric Zivot Updated: May 10, 2010 Reading MFTS, chapter 19 FMUND, chapters 6 and 7 Introduction Capturing co-movement between financial asset returns

More information

GENERAL MULTIVARIATE DEPENDENCE USING ASSOCIATED COPULAS

GENERAL MULTIVARIATE DEPENDENCE USING ASSOCIATED COPULAS REVSTAT Statistical Journal Volume 14, Number 1, February 2016, 1 28 GENERAL MULTIVARIATE DEPENDENCE USING ASSOCIATED COPULAS Author: Yuri Salazar Flores Centre for Financial Risk, Macquarie University,

More information

Modelling Dependence with Copulas and Applications to Risk Management. Filip Lindskog, RiskLab, ETH Zürich

Modelling Dependence with Copulas and Applications to Risk Management. Filip Lindskog, RiskLab, ETH Zürich Modelling Dependence with Copulas and Applications to Risk Management Filip Lindskog, RiskLab, ETH Zürich 02-07-2000 Home page: http://www.math.ethz.ch/ lindskog E-mail: lindskog@math.ethz.ch RiskLab:

More information

Modelling Dependent Credit Risks

Modelling Dependent Credit Risks Modelling Dependent Credit Risks Filip Lindskog, RiskLab, ETH Zürich 30 November 2000 Home page:http://www.math.ethz.ch/ lindskog E-mail:lindskog@math.ethz.ch RiskLab:http://www.risklab.ch Modelling Dependent

More information

Extreme Value Analysis and Spatial Extremes

Extreme Value Analysis and Spatial Extremes Extreme Value Analysis and Department of Statistics Purdue University 11/07/2013 Outline Motivation 1 Motivation 2 Extreme Value Theorem and 3 Bayesian Hierarchical Models Copula Models Max-stable Models

More information

Overview of Extreme Value Theory. Dr. Sawsan Hilal space

Overview of Extreme Value Theory. Dr. Sawsan Hilal space Overview of Extreme Value Theory Dr. Sawsan Hilal space Maths Department - University of Bahrain space November 2010 Outline Part-1: Univariate Extremes Motivation Threshold Exceedances Part-2: Bivariate

More information

Multivariate Distribution Models

Multivariate Distribution Models Multivariate Distribution Models Model Description While the probability distribution for an individual random variable is called marginal, the probability distribution for multiple random variables is

More information

EVANESCE Implementation in S-PLUS FinMetrics Module. July 2, Insightful Corp

EVANESCE Implementation in S-PLUS FinMetrics Module. July 2, Insightful Corp EVANESCE Implementation in S-PLUS FinMetrics Module July 2, 2002 Insightful Corp The Extreme Value Analysis Employing Statistical Copula Estimation (EVANESCE) library for S-PLUS FinMetrics module provides

More information

Dependence. Practitioner Course: Portfolio Optimization. John Dodson. September 10, Dependence. John Dodson. Outline.

Dependence. Practitioner Course: Portfolio Optimization. John Dodson. September 10, Dependence. John Dodson. Outline. Practitioner Course: Portfolio Optimization September 10, 2008 Before we define dependence, it is useful to define Random variables X and Y are independent iff For all x, y. In particular, F (X,Y ) (x,

More information

Bivariate generalized Pareto distribution

Bivariate generalized Pareto distribution Bivariate generalized Pareto distribution in practice Eötvös Loránd University, Budapest, Hungary Minisymposium on Uncertainty Modelling 27 September 2011, CSASC 2011, Krems, Austria Outline Short summary

More information

Tail Dependence of Multivariate Pareto Distributions

Tail Dependence of Multivariate Pareto Distributions !#"%$ & ' ") * +!-,#. /10 243537698:6 ;=@?A BCDBFEHGIBJEHKLB MONQP RS?UTV=XW>YZ=eda gihjlknmcoqprj stmfovuxw yy z {} ~ ƒ }ˆŠ ~Œ~Ž f ˆ ` š œžÿ~ ~Ÿ œ } ƒ œ ˆŠ~ œ

More information

Lecture Quantitative Finance Spring Term 2015

Lecture Quantitative Finance Spring Term 2015 on bivariate Lecture Quantitative Finance Spring Term 2015 Prof. Dr. Erich Walter Farkas Lecture 07: April 2, 2015 1 / 54 Outline on bivariate 1 2 bivariate 3 Distribution 4 5 6 7 8 Comments and conclusions

More information

On the Estimation and Application of Max-Stable Processes

On the Estimation and Application of Max-Stable Processes On the Estimation and Application of Max-Stable Processes Zhengjun Zhang Department of Statistics University of Wisconsin Madison, WI 53706, USA Co-author: Richard Smith EVA 2009, Fort Collins, CO Z. Zhang

More information

ESTIMATING BIVARIATE TAIL

ESTIMATING BIVARIATE TAIL Elena DI BERNARDINO b joint work with Clémentine PRIEUR a and Véronique MAUME-DESCHAMPS b a LJK, Université Joseph Fourier, Grenoble 1 b Laboratoire SAF, ISFA, Université Lyon 1 Framework Goal: estimating

More information

Bivariate Rainfall and Runoff Analysis Using Entropy and Copula Theories

Bivariate Rainfall and Runoff Analysis Using Entropy and Copula Theories Entropy 2012, 14, 1784-1812; doi:10.3390/e14091784 Article OPEN ACCESS entropy ISSN 1099-4300 www.mdpi.com/journal/entropy Bivariate Rainfall and Runoff Analysis Using Entropy and Copula Theories Lan Zhang

More information

A simple graphical method to explore tail-dependence in stock-return pairs

A simple graphical method to explore tail-dependence in stock-return pairs A simple graphical method to explore tail-dependence in stock-return pairs Klaus Abberger, University of Konstanz, Germany Abstract: For a bivariate data set the dependence structure can not only be measured

More information

Nonparametric Estimation of the Dependence Function for a Multivariate Extreme Value Distribution

Nonparametric Estimation of the Dependence Function for a Multivariate Extreme Value Distribution Nonparametric Estimation of the Dependence Function for a Multivariate Extreme Value Distribution p. /2 Nonparametric Estimation of the Dependence Function for a Multivariate Extreme Value Distribution

More information

On the Conditional Value at Risk (CoVaR) from the copula perspective

On the Conditional Value at Risk (CoVaR) from the copula perspective On the Conditional Value at Risk (CoVaR) from the copula perspective Piotr Jaworski Institute of Mathematics, Warsaw University, Poland email: P.Jaworski@mimuw.edu.pl 1 Overview 1. Basics about VaR, CoVaR

More information

Copulas. Mathematisches Seminar (Prof. Dr. D. Filipovic) Di Uhr in E

Copulas. Mathematisches Seminar (Prof. Dr. D. Filipovic) Di Uhr in E Copulas Mathematisches Seminar (Prof. Dr. D. Filipovic) Di. 14-16 Uhr in E41 A Short Introduction 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 The above picture shows a scatterplot (500 points) from a pair

More information

Statistics of Extremes

Statistics of Extremes Statistics of Extremes Anthony Davison c 211 http://stat.epfl.ch Multivariate Extremes 19 Componentwise maxima.................................................. 194 Standardization........................................................

More information

Dependence. MFM Practitioner Module: Risk & Asset Allocation. John Dodson. September 11, Dependence. John Dodson. Outline.

Dependence. MFM Practitioner Module: Risk & Asset Allocation. John Dodson. September 11, Dependence. John Dodson. Outline. MFM Practitioner Module: Risk & Asset Allocation September 11, 2013 Before we define dependence, it is useful to define Random variables X and Y are independent iff For all x, y. In particular, F (X,Y

More information

When is a copula constant? A test for changing relationships

When is a copula constant? A test for changing relationships When is a copula constant? A test for changing relationships Fabio Busetti and Andrew Harvey Bank of Italy and University of Cambridge November 2007 usetti and Harvey (Bank of Italy and University of Cambridge)

More information

Non parametric estimation of Archimedean copulas and tail dependence. Paris, february 19, 2015.

Non parametric estimation of Archimedean copulas and tail dependence. Paris, february 19, 2015. Non parametric estimation of Archimedean copulas and tail dependence Elena Di Bernardino a and Didier Rullière b Paris, february 19, 2015. a CNAM, Paris, Département IMATH, b ISFA, Université Lyon 1, Laboratoire

More information

Bayesian Point Process Modeling for Extreme Value Analysis, with an Application to Systemic Risk Assessment in Correlated Financial Markets

Bayesian Point Process Modeling for Extreme Value Analysis, with an Application to Systemic Risk Assessment in Correlated Financial Markets Bayesian Point Process Modeling for Extreme Value Analysis, with an Application to Systemic Risk Assessment in Correlated Financial Markets Athanasios Kottas Department of Applied Mathematics and Statistics,

More information

Bayesian Modelling of Extreme Rainfall Data

Bayesian Modelling of Extreme Rainfall Data Bayesian Modelling of Extreme Rainfall Data Elizabeth Smith A thesis submitted for the degree of Doctor of Philosophy at the University of Newcastle upon Tyne September 2005 UNIVERSITY OF NEWCASTLE Bayesian

More information

Accounting for extreme-value dependence in multivariate data

Accounting for extreme-value dependence in multivariate data Accounting for extreme-value dependence in multivariate data 38th ASTIN Colloquium Manchester, July 15, 2008 Outline 1. Dependence modeling through copulas 2. Rank-based inference 3. Extreme-value dependence

More information

Contents 1. Coping with Copulas. Thorsten Schmidt 1. Department of Mathematics, University of Leipzig Dec 2006

Contents 1. Coping with Copulas. Thorsten Schmidt 1. Department of Mathematics, University of Leipzig Dec 2006 Contents 1 Coping with Copulas Thorsten Schmidt 1 Department of Mathematics, University of Leipzig Dec 2006 Forthcoming in Risk Books Copulas - From Theory to Applications in Finance Contents 1 Introdcution

More information

CONTAGION VERSUS FLIGHT TO QUALITY IN FINANCIAL MARKETS

CONTAGION VERSUS FLIGHT TO QUALITY IN FINANCIAL MARKETS EVA IV, CONTAGION VERSUS FLIGHT TO QUALITY IN FINANCIAL MARKETS Jose Olmo Department of Economics City University, London (joint work with Jesús Gonzalo, Universidad Carlos III de Madrid) 4th Conference

More information

How to select a good vine

How to select a good vine Universitetet i Oslo ingrihaf@math.uio.no International FocuStat Workshop on Focused Information Criteria and Related Themes, May 9-11, 2016 Copulae Regular vines Model selection and reduction Limitations

More information

New Classes of Multivariate Survival Functions

New Classes of Multivariate Survival Functions Xiao Qin 2 Richard L. Smith 2 Ruoen Ren School of Economics and Management Beihang University Beijing, China 2 Department of Statistics and Operations Research University of North Carolina Chapel Hill,

More information

Tail dependence in bivariate skew-normal and skew-t distributions

Tail dependence in bivariate skew-normal and skew-t distributions Tail dependence in bivariate skew-normal and skew-t distributions Paola Bortot Department of Statistical Sciences - University of Bologna paola.bortot@unibo.it Abstract: Quantifying dependence between

More information

The extremal elliptical model: Theoretical properties and statistical inference

The extremal elliptical model: Theoretical properties and statistical inference 1/25 The extremal elliptical model: Theoretical properties and statistical inference Thomas OPITZ Supervisors: Jean-Noel Bacro, Pierre Ribereau Institute of Mathematics and Modeling in Montpellier (I3M)

More information

Estimation of multivariate critical layers: Applications to rainfall data

Estimation of multivariate critical layers: Applications to rainfall data Elena Di Bernardino, ICRA 6 / RISK 2015 () Estimation of Multivariate critical layers Barcelona, May 26-29, 2015 Estimation of multivariate critical layers: Applications to rainfall data Elena Di Bernardino,

More information

Tail Dependence Functions and Vine Copulas

Tail Dependence Functions and Vine Copulas Tail Dependence Functions and Vine Copulas Harry Joe Haijun Li Aristidis K. Nikoloulopoulos Revision: May 29 Abstract Tail dependence and conditional tail dependence functions describe, respectively, the

More information

Models and estimation.

Models and estimation. Bivariate generalized Pareto distribution practice: Models and estimation. Eötvös Loránd University, Budapest, Hungary 7 June 2011, ASMDA Conference, Rome, Italy Problem How can we properly estimate the

More information

Estimating the tail-dependence coefficient: Properties and pitfalls

Estimating the tail-dependence coefficient: Properties and pitfalls Estimating the tail-dependence coefficient: Properties and pitfalls Gabriel Frahm Markus Junker Rafael Schmidt May 26, 2006 Abstract The concept of tail dependence describes the amount of dependence in

More information

Lecture 2 One too many inequalities

Lecture 2 One too many inequalities University of Illinois Department of Economics Spring 2017 Econ 574 Roger Koenker Lecture 2 One too many inequalities In lecture 1 we introduced some of the basic conceptual building materials of the course.

More information

Explicit Bounds for the Distribution Function of the Sum of Dependent Normally Distributed Random Variables

Explicit Bounds for the Distribution Function of the Sum of Dependent Normally Distributed Random Variables Explicit Bounds for the Distribution Function of the Sum of Dependent Normally Distributed Random Variables Walter Schneider July 26, 20 Abstract In this paper an analytic expression is given for the bounds

More information

The Instability of Correlations: Measurement and the Implications for Market Risk

The Instability of Correlations: Measurement and the Implications for Market Risk The Instability of Correlations: Measurement and the Implications for Market Risk Prof. Massimo Guidolin 20254 Advanced Quantitative Methods for Asset Pricing and Structuring Winter/Spring 2018 Threshold

More information

Multivariate Measures of Positive Dependence

Multivariate Measures of Positive Dependence Int. J. Contemp. Math. Sciences, Vol. 4, 2009, no. 4, 191-200 Multivariate Measures of Positive Dependence Marta Cardin Department of Applied Mathematics University of Venice, Italy mcardin@unive.it Abstract

More information

Probability Distributions and Estimation of Ali-Mikhail-Haq Copula

Probability Distributions and Estimation of Ali-Mikhail-Haq Copula Applied Mathematical Sciences, Vol. 4, 2010, no. 14, 657-666 Probability Distributions and Estimation of Ali-Mikhail-Haq Copula Pranesh Kumar Mathematics Department University of Northern British Columbia

More information

Copulas and dependence measurement

Copulas and dependence measurement Copulas and dependence measurement Thorsten Schmidt. Chemnitz University of Technology, Mathematical Institute, Reichenhainer Str. 41, Chemnitz. thorsten.schmidt@mathematik.tu-chemnitz.de Keywords: copulas,

More information

Estimating Bivariate Tail: a copula based approach

Estimating Bivariate Tail: a copula based approach Estimating Bivariate Tail: a copula based approach Elena Di Bernardino, Université Lyon 1 - ISFA, Institut de Science Financiere et d'assurances - AST&Risk (ANR Project) Joint work with Véronique Maume-Deschamps

More information

THE MODELLING OF HYDROLOGICAL JOINT EVENTS ON THE MORAVA RIVER USING AGGREGATION OPERATORS

THE MODELLING OF HYDROLOGICAL JOINT EVENTS ON THE MORAVA RIVER USING AGGREGATION OPERATORS 2009/3 PAGES 9 15 RECEIVED 10. 12. 2007 ACCEPTED 1. 6. 2009 R. MATÚŠ THE MODELLING OF HYDROLOGICAL JOINT EVENTS ON THE MORAVA RIVER USING AGGREGATION OPERATORS ABSTRACT Rastislav Matúš Department of Water

More information

A Conditional Approach to Modeling Multivariate Extremes

A Conditional Approach to Modeling Multivariate Extremes A Approach to ing Multivariate Extremes By Heffernan & Tawn Department of Statistics Purdue University s April 30, 2014 Outline s s Multivariate Extremes s A central aim of multivariate extremes is trying

More information

Modelling and Estimation of Stochastic Dependence

Modelling and Estimation of Stochastic Dependence Modelling and Estimation of Stochastic Dependence Uwe Schmock Based on joint work with Dr. Barbara Dengler Financial and Actuarial Mathematics and Christian Doppler Laboratory for Portfolio Risk Management

More information

A Brief Introduction to Copulas

A Brief Introduction to Copulas A Brief Introduction to Copulas Speaker: Hua, Lei February 24, 2009 Department of Statistics University of British Columbia Outline Introduction Definition Properties Archimedean Copulas Constructing Copulas

More information

Construction and estimation of high dimensional copulas

Construction and estimation of high dimensional copulas Construction and estimation of high dimensional copulas Gildas Mazo PhD work supervised by S. Girard and F. Forbes Mistis, Inria and laboratoire Jean Kuntzmann, Grenoble, France Séminaire Statistiques,

More information

Clearly, if F is strictly increasing it has a single quasi-inverse, which equals the (ordinary) inverse function F 1 (or, sometimes, F 1 ).

Clearly, if F is strictly increasing it has a single quasi-inverse, which equals the (ordinary) inverse function F 1 (or, sometimes, F 1 ). APPENDIX A SIMLATION OF COPLAS Copulas have primary and direct applications in the simulation of dependent variables. We now present general procedures to simulate bivariate, as well as multivariate, dependent

More information

Lehrstuhl für Statistik und Ökonometrie. Diskussionspapier 87 / Some critical remarks on Zhang s gamma test for independence

Lehrstuhl für Statistik und Ökonometrie. Diskussionspapier 87 / Some critical remarks on Zhang s gamma test for independence Lehrstuhl für Statistik und Ökonometrie Diskussionspapier 87 / 2011 Some critical remarks on Zhang s gamma test for independence Ingo Klein Fabian Tinkl Lange Gasse 20 D-90403 Nürnberg Some critical remarks

More information

Robustness of a semiparametric estimator of a copula

Robustness of a semiparametric estimator of a copula Robustness of a semiparametric estimator of a copula Gunky Kim a, Mervyn J. Silvapulle b and Paramsothy Silvapulle c a Department of Econometrics and Business Statistics, Monash University, c Caulfield

More information

Technische Universität München. Zentrum Mathematik. Modeling dependence among meteorological measurements and tree ring data

Technische Universität München. Zentrum Mathematik. Modeling dependence among meteorological measurements and tree ring data Technische Universität München Zentrum Mathematik Modeling dependence among meteorological measurements and tree ring data Diplomarbeit von Michael Pachali Themenstellerin: Prof. Claudia Czado, Ph.D. Betreuer:

More information

Construction of asymmetric multivariate copulas

Construction of asymmetric multivariate copulas Construction of asymmetric multivariate copulas Eckhard Liebscher University of Applied Sciences Merseburg Department of Computer Sciences and Communication Systems Geusaer Straße 0627 Merseburg Germany

More information

Gaussian processes. Chuong B. Do (updated by Honglak Lee) November 22, 2008

Gaussian processes. Chuong B. Do (updated by Honglak Lee) November 22, 2008 Gaussian processes Chuong B Do (updated by Honglak Lee) November 22, 2008 Many of the classical machine learning algorithms that we talked about during the first half of this course fit the following pattern:

More information

Tail Approximation of Value-at-Risk under Multivariate Regular Variation

Tail Approximation of Value-at-Risk under Multivariate Regular Variation Tail Approximation of Value-at-Risk under Multivariate Regular Variation Yannan Sun Haijun Li July 00 Abstract This paper presents a general tail approximation method for evaluating the Valueat-Risk of

More information

Multivariate Distributions

Multivariate Distributions IEOR E4602: Quantitative Risk Management Spring 2016 c 2016 by Martin Haugh Multivariate Distributions We will study multivariate distributions in these notes, focusing 1 in particular on multivariate

More information

A copula goodness-of-t approach. conditional probability integral transform. Daniel Berg 1 Henrik Bakken 2

A copula goodness-of-t approach. conditional probability integral transform. Daniel Berg 1 Henrik Bakken 2 based on the conditional probability integral transform Daniel Berg 1 Henrik Bakken 2 1 Norwegian Computing Center (NR) & University of Oslo (UiO) 2 Norwegian University of Science and Technology (NTNU)

More information

X

X Correlation: Pitfalls and Alternatives Paul Embrechts, Alexander McNeil & Daniel Straumann Departement Mathematik, ETH Zentrum, CH-8092 Zürich Tel: +41 1 632 61 62, Fax: +41 1 632 15 23 embrechts/mcneil/strauman@math.ethz.ch

More information

First steps of multivariate data analysis

First steps of multivariate data analysis First steps of multivariate data analysis November 28, 2016 Let s Have Some Coffee We reproduce the coffee example from Carmona, page 60 ff. This vignette is the first excursion away from univariate data.

More information

Trivariate copulas for characterisation of droughts

Trivariate copulas for characterisation of droughts ANZIAM J. 49 (EMAC2007) pp.c306 C323, 2008 C306 Trivariate copulas for characterisation of droughts G. Wong 1 M. F. Lambert 2 A. V. Metcalfe 3 (Received 3 August 2007; revised 4 January 2008) Abstract

More information

Modelling Multivariate Peaks-over-Thresholds using Generalized Pareto Distributions

Modelling Multivariate Peaks-over-Thresholds using Generalized Pareto Distributions Modelling Multivariate Peaks-over-Thresholds using Generalized Pareto Distributions Anna Kiriliouk 1 Holger Rootzén 2 Johan Segers 1 Jennifer L. Wadsworth 3 1 Université catholique de Louvain (BE) 2 Chalmers

More information

EXTREMAL DEPENDENCE OF MULTIVARIATE DISTRIBUTIONS AND ITS APPLICATIONS YANNAN SUN

EXTREMAL DEPENDENCE OF MULTIVARIATE DISTRIBUTIONS AND ITS APPLICATIONS YANNAN SUN EXTREMAL DEPENDENCE OF MULTIVARIATE DISTRIBUTIONS AND ITS APPLICATIONS By YANNAN SUN A dissertation submitted in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY WASHINGTON

More information

Copulas. MOU Lili. December, 2014

Copulas. MOU Lili. December, 2014 Copulas MOU Lili December, 2014 Outline Preliminary Introduction Formal Definition Copula Functions Estimating the Parameters Example Conclusion and Discussion Preliminary MOU Lili SEKE Team 3/30 Probability

More information

Tail dependence coefficient of generalized hyperbolic distribution

Tail dependence coefficient of generalized hyperbolic distribution Tail dependence coefficient of generalized hyperbolic distribution Mohalilou Aleiyouka Laboratoire de mathématiques appliquées du Havre Université du Havre Normandie Le Havre France mouhaliloune@gmail.com

More information

Correlation: Copulas and Conditioning

Correlation: Copulas and Conditioning Correlation: Copulas and Conditioning This note reviews two methods of simulating correlated variates: copula methods and conditional distributions, and the relationships between them. Particular emphasis

More information

Bivariate Paired Numerical Data

Bivariate Paired Numerical Data Bivariate Paired Numerical Data Pearson s correlation, Spearman s ρ and Kendall s τ, tests of independence University of California, San Diego Instructor: Ery Arias-Castro http://math.ucsd.edu/~eariasca/teaching.html

More information

Fall 2017 STAT 532 Homework Peter Hoff. 1. Let P be a probability measure on a collection of sets A.

Fall 2017 STAT 532 Homework Peter Hoff. 1. Let P be a probability measure on a collection of sets A. 1. Let P be a probability measure on a collection of sets A. (a) For each n N, let H n be a set in A such that H n H n+1. Show that P (H n ) monotonically converges to P ( k=1 H k) as n. (b) For each n

More information

Parameter addition to a family of multivariate exponential and weibull distribution

Parameter addition to a family of multivariate exponential and weibull distribution ISSN: 2455-216X Impact Factor: RJIF 5.12 www.allnationaljournal.com Volume 4; Issue 3; September 2018; Page No. 31-38 Parameter addition to a family of multivariate exponential and weibull distribution

More information

Marginal Specifications and a Gaussian Copula Estimation

Marginal Specifications and a Gaussian Copula Estimation Marginal Specifications and a Gaussian Copula Estimation Kazim Azam Abstract Multivariate analysis involving random variables of different type like count, continuous or mixture of both is frequently required

More information

MULTIDIMENSIONAL POVERTY MEASUREMENT: DEPENDENCE BETWEEN WELL-BEING DIMENSIONS USING COPULA FUNCTION

MULTIDIMENSIONAL POVERTY MEASUREMENT: DEPENDENCE BETWEEN WELL-BEING DIMENSIONS USING COPULA FUNCTION Rivista Italiana di Economia Demografia e Statistica Volume LXXII n. 3 Luglio-Settembre 2018 MULTIDIMENSIONAL POVERTY MEASUREMENT: DEPENDENCE BETWEEN WELL-BEING DIMENSIONS USING COPULA FUNCTION Kateryna

More information

A Goodness-of-fit Test for Copulas

A Goodness-of-fit Test for Copulas A Goodness-of-fit Test for Copulas Artem Prokhorov August 2008 Abstract A new goodness-of-fit test for copulas is proposed. It is based on restrictions on certain elements of the information matrix and

More information

On tail dependence coecients of transformed multivariate Archimedean copulas

On tail dependence coecients of transformed multivariate Archimedean copulas Tails and for Archim Copula () February 2015, University of Lille 3 On tail dependence coecients of transformed multivariate Archimedean copulas Elena Di Bernardino, CNAM, Paris, Département IMATH Séminaire

More information

Probability Methods in Civil Engineering Prof. Dr. Rajib Maity Department of Civil Engineering Indian Institute of Technology, Kharagpur

Probability Methods in Civil Engineering Prof. Dr. Rajib Maity Department of Civil Engineering Indian Institute of Technology, Kharagpur Probability Methods in Civil Engineering Prof. Dr. Rajib Maity Department of Civil Engineering Indian Institute of Technology, Kharagpur Lecture No. # 29 Introduction to Copulas Hello and welcome to this

More information

Copulas, a novel approach to model spatial and spatio-temporal dependence

Copulas, a novel approach to model spatial and spatio-temporal dependence Copulas, a novel approach to model spatial and spatio-temporal dependence Benedikt Gräler 1, Hannes Kazianka 2, Giovana Mira de Espindola 3 1 Institute for Geoinformatics, University of Münster, Germany

More information

Module 1. Probability

Module 1. Probability Module 1 Probability 1. Introduction In our daily life we come across many processes whose nature cannot be predicted in advance. Such processes are referred to as random processes. The only way to derive

More information

Does k-th Moment Exist?

Does k-th Moment Exist? Does k-th Moment Exist? Hitomi, K. 1 and Y. Nishiyama 2 1 Kyoto Institute of Technology, Japan 2 Institute of Economic Research, Kyoto University, Japan Email: hitomi@kit.ac.jp Keywords: Existence of moments,

More information

A New Generalized Gumbel Copula for Multivariate Distributions

A New Generalized Gumbel Copula for Multivariate Distributions A New Generalized Gumbel Copula for Multivariate Distributions Chandra R. Bhat* The University of Texas at Austin Department of Civil, Architectural & Environmental Engineering University Station, C76,

More information

Journal of Statistical Software

Journal of Statistical Software JSS Journal of Statistical Software January 2013, Volume 52, Issue 3. http://www.jstatsoft.org/ CDVine: Modeling Dependence with C- and D-Vine Copulas in R Eike Christian Brechmann Technische Universität

More information

arxiv: v1 [math.pr] 24 Aug 2018

arxiv: v1 [math.pr] 24 Aug 2018 On a class of norms generated by nonnegative integrable distributions arxiv:180808200v1 [mathpr] 24 Aug 2018 Michael Falk a and Gilles Stupfler b a Institute of Mathematics, University of Würzburg, Würzburg,

More information

Modulation of symmetric densities

Modulation of symmetric densities 1 Modulation of symmetric densities 1.1 Motivation This book deals with a formulation for the construction of continuous probability distributions and connected statistical aspects. Before we begin, a

More information

HANDBOOK OF APPLICABLE MATHEMATICS

HANDBOOK OF APPLICABLE MATHEMATICS HANDBOOK OF APPLICABLE MATHEMATICS Chief Editor: Walter Ledermann Volume VI: Statistics PART A Edited by Emlyn Lloyd University of Lancaster A Wiley-Interscience Publication JOHN WILEY & SONS Chichester

More information

Multivariate Non-Normally Distributed Random Variables

Multivariate Non-Normally Distributed Random Variables Multivariate Non-Normally Distributed Random Variables An Introduction to the Copula Approach Workgroup seminar on climate dynamics Meteorological Institute at the University of Bonn 18 January 2008, Bonn

More information

Subject CS1 Actuarial Statistics 1 Core Principles

Subject CS1 Actuarial Statistics 1 Core Principles Institute of Actuaries of India Subject CS1 Actuarial Statistics 1 Core Principles For 2019 Examinations Aim The aim of the Actuarial Statistics 1 subject is to provide a grounding in mathematical and

More information

The Mixture Approach for Simulating New Families of Bivariate Distributions with Specified Correlations

The Mixture Approach for Simulating New Families of Bivariate Distributions with Specified Correlations The Mixture Approach for Simulating New Families of Bivariate Distributions with Specified Correlations John R. Michael, Significance, Inc. and William R. Schucany, Southern Methodist University The mixture

More information

arxiv: v1 [math.pr] 9 Jan 2016

arxiv: v1 [math.pr] 9 Jan 2016 SKLAR S THEOREM IN AN IMPRECISE SETTING IGNACIO MONTES, ENRIQUE MIRANDA, RENATO PELESSONI, AND PAOLO VICIG arxiv:1601.02121v1 [math.pr] 9 Jan 2016 Abstract. Sklar s theorem is an important tool that connects

More information

Pair-copula constructions of multiple dependence

Pair-copula constructions of multiple dependence Pair-copula constructions of multiple dependence 3 4 5 3 34 45 T 3 34 45 3 4 3 35 4 T 3 4 3 35 4 4 3 5 34 T 3 4 3 5 34 5 34 T 4 Note no SAMBA/4/06 Authors Kjersti Aas Claudia Czado Arnoldo Frigessi Henrik

More information

Tail comonotonicity: properties, constructions, and asymptotic additivity of risk measures

Tail comonotonicity: properties, constructions, and asymptotic additivity of risk measures Tail comonotonicity: properties, constructions, and asymptotic additivity of risk measures Lei Hua Harry Joe June 5, 2012 Abstract. We investigate properties of a version of tail comonotonicity that can

More information

Stable Process. 2. Multivariate Stable Distributions. July, 2006

Stable Process. 2. Multivariate Stable Distributions. July, 2006 Stable Process 2. Multivariate Stable Distributions July, 2006 1. Stable random vectors. 2. Characteristic functions. 3. Strictly stable and symmetric stable random vectors. 4. Sub-Gaussian random vectors.

More information

Probabilistic Engineering Mechanics. An innovating analysis of the Nataf transformation from the copula viewpoint

Probabilistic Engineering Mechanics. An innovating analysis of the Nataf transformation from the copula viewpoint Probabilistic Engineering Mechanics 4 9 3 3 Contents lists available at ScienceDirect Probabilistic Engineering Mechanics journal homepage: www.elsevier.com/locate/probengmech An innovating analysis of

More information

PREPRINT 2005:38. Multivariate Generalized Pareto Distributions HOLGER ROOTZÉN NADER TAJVIDI

PREPRINT 2005:38. Multivariate Generalized Pareto Distributions HOLGER ROOTZÉN NADER TAJVIDI PREPRINT 2005:38 Multivariate Generalized Pareto Distributions HOLGER ROOTZÉN NADER TAJVIDI Department of Mathematical Sciences Division of Mathematical Statistics CHALMERS UNIVERSITY OF TECHNOLOGY GÖTEBORG

More information

Modelling Dropouts by Conditional Distribution, a Copula-Based Approach

Modelling Dropouts by Conditional Distribution, a Copula-Based Approach The 8th Tartu Conference on MULTIVARIATE STATISTICS, The 6th Conference on MULTIVARIATE DISTRIBUTIONS with Fixed Marginals Modelling Dropouts by Conditional Distribution, a Copula-Based Approach Ene Käärik

More information

Asymptotic behaviour of multivariate default probabilities and default correlations under stress

Asymptotic behaviour of multivariate default probabilities and default correlations under stress Asymptotic behaviour of multivariate default probabilities and default correlations under stress 7th General AMaMeF and Swissquote Conference EPFL, Lausanne Natalie Packham joint with Michael Kalkbrener

More information

The multivariate probability integral transform

The multivariate probability integral transform The multivariate probability integral transform Fabrizio Durante Faculty of Economics and Management Free University of Bozen-Bolzano (Italy) fabrizio.durante@unibz.it http://sites.google.com/site/fbdurante

More information

GENERAL MULTIVARIATE DEPENDENCE USING ASSOCIATED COPULAS

GENERAL MULTIVARIATE DEPENDENCE USING ASSOCIATED COPULAS GENERAL MULTIVARIATE DEPENDENCE USING ASSOCIATED COPULAS YURI SALAZAR FLORES University of Essex, Wivenhoe Park, CO4 3SQ, Essex, UK. ysalaz@essex.ac.uk Abstract. This paper studies the general multivariate

More information

Properties of Hierarchical Archimedean Copulas

Properties of Hierarchical Archimedean Copulas SFB 649 Discussion Paper 9-4 Properties of Hierarchical Archimedean Copulas Ostap Okhrin* Yarema Okhrin** Wolfgang Schmid*** *Humboldt-Universität zu Berlin, Germany **Universität Bern, Switzerland ***Universität

More information

Introduction to Maximum Likelihood Estimation

Introduction to Maximum Likelihood Estimation Introduction to Maximum Likelihood Estimation Eric Zivot July 26, 2012 The Likelihood Function Let 1 be an iid sample with pdf ( ; ) where is a ( 1) vector of parameters that characterize ( ; ) Example:

More information

Sklar s theorem in an imprecise setting

Sklar s theorem in an imprecise setting Sklar s theorem in an imprecise setting Ignacio Montes a,, Enrique Miranda a, Renato Pelessoni b, Paolo Vicig b a University of Oviedo (Spain), Dept. of Statistics and O.R. b University of Trieste (Italy),

More information

Operational Risk and Pareto Lévy Copulas

Operational Risk and Pareto Lévy Copulas Operational Risk and Pareto Lévy Copulas Claudia Klüppelberg Technische Universität München email: cklu@ma.tum.de http://www-m4.ma.tum.de References: - Böcker, K. and Klüppelberg, C. (25) Operational VaR

More information

Introduction to Dependence Modelling

Introduction to Dependence Modelling Introduction to Dependence Modelling Carole Bernard Berlin, May 2015. 1 Outline Modeling Dependence Part 1: Introduction 1 General concepts on dependence. 2 in 2 or N 3 dimensions. 3 Minimizing the expectation

More information