Scribd
Upload
328 views

Copula Intro

This document provides an introduction to copulas, which are functions that join marginal distribution functions to describe a joint distribution. It defines copulas mathematically and discusses their key properties like being grounded and 2-increasing. Copulas allow describing the dependence between random variables independently of their marginal distributions. The document also covers basic copula terminology, Fréchet-Hoeffding bounds, nonparametric measures of dependence related to copulas like Spearman's rank correlation coefficient and Kendall's tau, and concludes by noting copulas are useful for obtaining joint distributions from marginals.

Uploaded by

ashutosh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
328 views

Copula Intro

This document provides an introduction to copulas, which are functions that join marginal distribution functions to describe a joint distribution. It defines copulas mathematically and discusses their key properties like being grounded and 2-increasing. Copulas allow describing the dependence between random variables independently of their marginal distributions. The document also covers basic copula terminology, Fréchet-Hoeffding bounds, nonparametric measures of dependence related to copulas like Spearman's rank correlation coefficient and Kendall's tau, and concludes by noting copulas are useful for obtaining joint distributions from marginals.

Uploaded by

ashutosh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

M5L11

Introduction to Copulas
1. Introduction
In this lecture, the theory of copula is introduced. The definition of copula function with
explanation, its properties, basic terminologies, Frchet-Hoeffding Bounds, Sklar theorem
and measures of dependence are discussed.
2. Introduction to Copula
Origin of the word copula is the Latin word copulare, which means to join together. In
many cases of statistical modeling, it is essential to obtain the joint pdf between two or more
random variables. Even though the marginal distributions of each of the random variables are
known, their joint distributions may not be easy to derive from these marginal distributions. If
the information on scale-free measures of dependence between random variables is available,
copula can be used to obtain their joint distribution.
3. Definition of Copula
3.1. Classical Definition
A copula, C , is a function that joins or couples multiple distribution functions to their onedimensional marginal distribution functions (Nelsen, 2006). Application of copula to
probability and statistics is achieved through Sklar Theorem (Sklar, 1959), which states that,
if H X ,Y x, y is a joint distribution function, then there exists a copula C u , v such that for

all x, y in R , ,

H X ,Y x, y C FX ( x), FY ( y )

(1)

where, FX x and FY y are the marginal distributions of X and Y respectively.


The conceptual definition of copula can be expressed in another way also. Let X and Y be a
pair of random variables with cumulative distribution functions (CDF) of FX x and FY y
respectively. Also let their joint CDF be H X ,Y x, y . Each pair, (x, y), of real numbers leads
to a point, FX x , FY y , in the unit square 0,1 0,1 and this ordered pair, in turn,

corresponds to a number, H X ,Y x, y , in [0,1]. This correspondence is indeed a function,


which is known as Copula (Nelsen, 2006).
It is worthwhile to note here that such correspondence is independent of the marginal
distributions of the random variables. In other words, any form of marginal distribution can
be coupled to get their joint distribution, which is the reason for the popularity of copula
theory in many areas of research.
3.2. Mathematical Definition

A 2-dimensional copula (2-copula) is a function C having following properties:


i.

DomC S1 S 2 , where S1 and S2 are subsets of containing 0 and 1, i.e., 0,1 .

ii.

C is grounded and 2-increasing.

iii.

For every u in S1 and v in S 2 , C u ,1 u and C 1, v v

4. Basic Terminologies
4.1. 2-increasing Function

The concept of 2-increasing function is basically 2-dimensional analog of non-decreasing


functions in one dimension. A 2-dimensional real function is 2-increasing if H -volume of
any rectangle B , i.e. VH B 0 for all rectangles B , whose vertices lie in DomH . This is
graphically explained in figures1below.

Fig. 1: (a) Rectangle B denoted as x1 , y1 x2 , y2 ; (b) Joint CDF H X ,Y x, y ;


(c) H -volume of B denoted as VH B in clockwise direction respectively
4.2. Grounded Function

Copula functions are grounded. Graphical representation of 2-dimensional grounded function


is shown in fig. 2. A function is known as zero if one of variable is zero, the value of the
function is zero. A 2-increasing function does not imply that H is non-descending in each
argument. However, if it is 2-increasing and grounded, then it also can be non-decreasing in
each argument.

H is zero at these
line segments
Fig. 2: 2-dimensional Grounded Function
4.3.

Properties of a copula function C(u,v)

A copula having domain of 2 has the following property:


1) For every u1 , u2 , v1 , v2 in if such that u1 u2 and v1 v2 , then
C u2 , v2 C u2 , v1 C u1 , v2 C u1 , v1 0

(2)

This property indicates that the copula functions are 2-increasing.


2) For every u, v in ,
C u ,0 0 C v,0

(3 and 4)

This property indicates that the copula functions are grounded


3) For every u, v in ,
C u ,1 u

(5)

C 1, v v

(6)

5. Graphical Representation of Copula


C u , v uv for 0 u , v 1 is an example of a copula function. This is the copula function

for two independent random variables. This copula function is graphically represented in 3D
function plot (Fig. 3) or in 2D contour plots (Fig. 4).

Fig. 3. Copula function in 3D function plot

Fig 4. Copula function in contour plot


6. Frchet-Hoeffding Bounds

Let C be a copula. Then for every u, v in DomC ,


max u v 1,0 C u , v min u , v

or
W u , v C u , v M u , v

where W u , v max u v 1,0 and M u , v min u , v


The graphical representation for W u , v and M u , v are given below in figures 5 and 6
respectively.

Fig. 5. Graphical representation of W u , v

Fig. 6. Graphical representation of M u , v

7. Nonparametric Measures of Association

Two different nonparametric measures of association are


a) Spearman's rank correlation coefficient or Spearman's rho ( S )
b) Kendall rank correlation coefficient or Kendalls Tau ( )

For a sample xi , y i of size n , to compute S , the samples are first transformed to their

y
x
respective ranks Ri , Ri and S is expressed as

x
i

R x Riy R y

i 1

Rix

i 1

x 2

Riy

(7)

y 2

i 1

If there is tie between two or more observations, an average of tie ranks are assigned to all
those ties.
The is expressed as

Nc Nd
1
nn 1
2

(8)

where N c and N d are the number of concordant and discordant pair respectively. If

xi x j yi y j 0 then the pair x , y and x , y is known as concordant pair and if


xi x j yi y j 0 then the pair x , y and x , y is known as discordant pair. If
xi x j yi y j 0 , i.e., if there is tie, these pairs are neither concordant or discordant.
i

Further details on these measures can be found elsewhere.


These measures are related to the copula through following expressions:

S 12 uvdC u, v 3

(9)

4 C u , v dC u , v 1

(10)

and

These expressions can be further simplified for Archimedean copulas. These will be
discussed in the next lecture.

8. Concluding Remarks

It is important to remember that copula is an effective tool to obtain joint CDF from
individual marginal distributions. In this lecture the theory of copula function is introduced.
Classical definition is discussed in detail with mathematical expressions and graphical
representations. A popular class of copula, known as Archimedean copula, will be discussed
in the next lecture.

References:

Nelsen, R. B. (2006), An introduction to copulas, Springer series in Statistics, 269 pp.,


Springer, New York.
Sklar, A. (1959), Fonctions de rpartition n dimensions et leurs marges, Publ. Inst. Stat.
Univ. Paris, 8, 229231.

You might also like