See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/40634880

Multivariate Fréchet copulas and conditional value-at-risk

Article in International Journal of Mathematics and Mathematical Sciences · February 2004

DOI: 10.1155/S0161171204210158 · Source: OAI

CITATIONS READS

43 347

1 author:

Werner Hürlimann
Swiss Mathematical Society
288 PUBLICATIONS 2,073 CITATIONS

SEE PROFILE

All content following this page was uploaded by Werner Hürlimann on 04 June 2014.

The user has requested enhancement of the downloaded file.

 IJMMS 2004:7, 345–364
PII. S0161171204210158
http://ijmms.hindawi.com
© Hindawi Publishing Corp.

MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL

VALUE-AT-RISK

WERNER HÜRLIMANN

Received 18 October 2002

Based on the method of copulas, we construct a parametric family of multivariate distribu-

tions using mixtures of independent conditional distributions. The new family of multivari-
ate copulas is a convex combination of products of independent and comonotone subcopu-
las. It fulfills the four most desirable properties that a multivariate statistical model should
satisfy. In particular, the bivariate margins belong to a simple but flexible one-parameter
family of bivariate copulas, called linear Spearman copula, which is similar but not identical
to the convex family of Fréchet. It is shown that the distribution and stop-loss transform
of dependent sums from this multivariate family can be evaluated using explicit integral
formulas, and that these dependent sums are bounded in convex order between the cor-
responding independent and comonotone sums. The model is applied to the evaluation of
the economic risk capital for a portfolio of risks using conditional value-at-risk measures.
A multivariate conditional value-at-risk vector measure is considered. Its components co-
incide for the constructed multivariate copula with the conditional value-at-risk measures
of the risk components of the portfolio. This yields a “fair” risk allocation in the sense that
each risk component becomes allocated to its coherent conditional value-at-risk.

2000 Mathematics Subject Classification: 62E15, 62H20, 62P05, 91B30.

1. Introduction. A natural framework for the construction of multivariate nonnor-

mal distributions is the method of copulas, justified by the theorem of Sklar [48]. It
permits a separate study and modeling of the marginal distributions and the depen-
dence structure. According to Joe [30, Section 4.1], a parametric family of distributions
should satisfy four desirable properties.
(a) There should exist an interpretation like a mixture or other stochastic represen-
tation.
(b) The margins, at least the univariate and bivariate ones, should belong to the same
parametric family and numerical evaluation should be possible.
(c) The bivariate dependence between the margins should be described by a param-
eter and cover a wide range of dependence.
(d) The multivariate distribution and density should preferably have a closed-form
representation; at least numerical evaluation should be possible.
In general, these desirable properties cannot be fulfilled simultaneously. For exam-
ple, multivariate normal distributions satisfy properties (a), (b), and (c) but not (d).
The method of copulas satisfies property (c) but implies only partial closedness under
the taking of margins, and can lead to computational complexity as the dimension in-
creases. In fact, it is an open problem to find parametric families of copulas that satisfy
all of the desirable properties. In the present paper, such a parametric family, called
346 WERNER HÜRLIMANN

multivariate linear Spearman copula, is constructed (formula (4.9)). It is based on the

method of mixtures of independent conditional distributions.
A growing need for and interest in suitable multivariate nonnormal distributions
stem from applications in actuarial science and finance, especially in risk manage-
ment. Given a risk or portfolio of risks, represented by a random variable X or random
vector X = (X1 , . . . , Xn ) with distribution Fx (x), one looks for risk measures suitable to

n
model the economic risk capital of the risk X or aggregate risk i=1 Xi . Two simple
measures are the value-at-risk and the conditional value-at-risk. Given a random vari-
able X, one considers the value-at-risk (VaR) to the confidence level α, defined as the
lower α-quantile:
 
VaRα [X] = QX (α) = inf x : FX (x) ≥ α , (1.1)

and the upper conditional value-at-risk (CVaR+ ) to the confidence level α, defined by
 
CVaR+
α [X] = E X | X > VaRα [X] . (1.2)

The VaR quantity represents the maximum possible loss, which is not exceeded with
the probability α. The CVaR+ quantity is the conditional expected loss given that the
loss strictly exceeds its value-at-risk. Next, consider the α-tail transform X a of X with
distribution


0, x < VaRα [X],
FX α (x) = FX (x) − α (1.3)

 , x ≥ VaRα [X].
1−α

Rockafellar and Uryasev [44] define conditional value-at-risk (CVaR) to the confidence
level α as the expected value of the α-tail transform, that is, by

CVaRα [X] = E[X α ]. (1.4)

The obtained measure is a coherent risk measure in the sense of Artzner et al. [4, 5] and
coincides with CVaR+ in the case of continuous distributions. It is well known that the
VaR measure is not coherent. For simplicity, we restrict throughout the attention to the
case of continuous distributions and identify CVaR with CVaR+ . For portfolios of risks,
we define a multivariate conditional value-at-risk vector measure, whose components
coincide for the multivariate linear Spearman copula with the CVaR measures of the
risk components of the portfolio (Theorem 6.1). This yields a “fair” risk allocation in the
sense that each risk component becomes allocated to its coherent univariate conditional
value-at-risk measure.
A more detailed outline of the content follows. Based on the method of copulas
summarized in Section 2.1, we recall in Section 2.2 the construction of parametric fam-
ilies of multivariate copulas using mixtures of independent conditional distributions.
Following this approach, it is first necessary to focus on a simple but sufficiently flex-
ible one-parameter family of bivariate copulas, called linear Spearman copula, which
is similar but not identical to the convex family of Fréchet [23] and is introduced in
Section 3.1. The analytical evaluation of the distribution and stop-loss transform of
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 347

bivariate sums following a linear Spearman copula, required in conditional value-at-

risk calculations, is presented in Section 3.2. Section 4 is devoted to the construction
of the new multivariate family of copulas that satisfies the four desirable properties
(a), (b), (c), and (d). Two important features of the multivariate linear Spearman copula
are presented in Section 5. First, we show that the distribution and stop-loss transform
of dependent sums following a multivariate linear Spearman copula can be evaluated
using explicit integral formulas (Theorem 5.1). Then, we establish that these depen-
dent sums are bounded in convex order between the corresponding independent and
comonotone sums (Theorem 5.2). Finally, Section 6 presents our application to condi-
tional value-at-risk.

2. Multivariate models with arbitrary marginals. Our view of multivariate statisti-

cal modeling is that of Joe [30, Section 1.7]: “Models should try to capture important
characteristics, such as the appropriate density shapes for the univariate margins and
the appropriate dependence structure, and otherwise be as simple as possible.” To ful-
fill this, a parametric family of multivariate distributions should satisfy the desirable
properties (a), (b), (c), and (d) mentioned in Section 1. It is an open problem to find para-
metric families of copulas that satisfy all these desirable properties (Joe [30, Section
4.13, page 138]). In the present paper, such a parametric family is constructed. It is
based on the method of mixtures of independent conditional distributions, discussed
in Section 2.2.

2.1. The method of copulas. Though copulas have been introduced since Sklar [48],
their use in insurance and finance is more recent. Textbooks treating copulas include
those by Hutchinson and Lai [29], Joe [30], Nelsen [41], and Drouet Mari and Kotz [17].
Recall that the copula representation of a continuous multivariate distribution al-
lows for a separate modeling of the univariate margins and the dependence struc-
ture. Denote by Mn := Mn (F1 , . . . , Fn ) the class of all continuous multivariate random
variables (X1 , . . . , Xn ) with given marginals Fi of Xi . If F denotes the multivariate dis-
tribution of (X1 , . . . , Xn ), then the copula associated with F is a distribution function
C : [0, 1]n → [0, 1] that satisfies

F (x) = C F1 x1 , . . . , Fn xn , x = x1 , . . . , x n ∈ R n . (2.1)

Reciprocally, if F ∈ Mn and Fi−1 are quantile functions of the margins, then

C(u) = F F1−1 u1 , . . . , Fn−1 un , u = u1 , . . . , un ∈ [0, 1]n , (2.2)

is the unique copula satisfying (2.1) (theorem of Sklar [48]).

Copulas are especially useful for the modeling and measurement of bivariate de-
pendence. For an axiomatic definition, one needs the important notion of concordance
ordering. A copula C1 (u, v) is said to be smaller than a copula C2 (u, v) in concordance
order, written C1 ≺ C2 , if one has

C1 (u, v) ≤ C2 (u, v), (u, v) ∈ [0, 1]2 . (2.3)

348 WERNER HÜRLIMANN

Definition 2.1 (Scarsini [46]). A numeric measure κ, written κX,Y or κC , of asso-

ciation between two continuous random variables X and Y with copula C(u, v) is a
measure of concordance if it satisfies the following properties:
(C1) κX,Y is defined for every couple (X, Y ) of continuous random variables;
(C2) −1 ≤ κX,Y ≤ 1, and κX,−X = −1, κX,X = 1;
(C3) κX,Y = κY ,X ;
(C4) if X and Y are independent, then κX,Y = 0;
(C5) κ−X,Y = κX,−Y = −κX,Y ;
(C6) if C1 ≺ C2 , then κC1 ≤ κC2 ;
(C7) if {(Xn , Yn )} is a sequence of continuous random variables with copulas Cn and
if {Cn } converges pointwise to C, then limn→∞ κCn = κC .
Two famous measures of concordance are Kendall’s tau,
1 1
∂ ∂
τ = 1−4· C(u, v) · C(u, v)du dv, (2.4)
0 0 ∂u ∂v

and Spearman’s rho,

1 1 
ρS = 12 · C(u, v) − uv du dv. (2.5)
0 0

The latter parameter will completely describe the bivariate dependence in our construc-
tion. When extreme values are involved, tail dependence should also be measured.

Definition 2.2. The coefficient of (upper) tail dependence of a couple (X, Y ) of con-
tinuous random variables is defined by

λ = λX,Y = lim Pr Y > FY−1 (u) | X > FX−1 (u) , (2.6)

u→1−

provided a limit λ ∈ [0, 1] exists. If λ ∈ (0, 1], this defines the asymptotic dependence
(in the upper tail), while if λ = 0, this defines the asymptotic independence.

Tail dependence is an asymptotic property. Its calculation follows easily from the
relation

1 − 2u + C(u, u)
λ = λX,Y = lim . (2.7)
u→1− 1−u

2.2. Mixtures of independent conditional distributions. Our goal is the construc-

tion of a parametric family of n-dimensional copulas that satisfies the desirable proper-
ties (a), (b), (c), and (d). It uses a simple variant of the method of mixtures of conditional
distributions described by Joe [30, Section 4.5]. To satisfy property (b), we focus on the
n Fréchet classes F Ci := F Ci (Fij , j = i), i = 1, . . . , n, of n-variate distributions for which
the bivariate margins Fij (xi , xj ) = F(Xi ,Xj ) (xi , xj ) = Cij Fi (xi ), Fj (xj ) , j = i, belong to
a given parametric family of copulas Cij ui , uj . Assume that the conditional distribu-
tions

∂Cij  
Fj|i xj xi = F i xi , F j xj (2.8)
∂ui
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 349

are well defined. The n-variate distribution such that the random variables Xj , j = i,
are conditionally independent, given Xi , is contained in F Ci and is defined by
 
xi 
F (i) (x) = Fj|i xj t · dFi (t). (2.9)
−∞
j=i

Choosing appropriately the bivariate copulas Cij ui , uj , it is possible to construct

n-variate copulas C (i) (u1 , . . . , un ), i = 1, . . . , n, such that F (i) belongs to C (i) and the
bivariate margins Fij , j = i, belong to Cij . Moreover, any convex combination of the
C (i) ’s, that is,


n 
n
C u1 , . . . , u n = λi C (i) u1 , . . . , un , 0 ≤ λi ≤ 1, λi = 1, (2.10)
i=1 i=1

is again an n-variate copula, which, by appropriate choice, may satisfy the desirable
properties.

3. A bivariate model with arbitrary marginals. Our aim is the construction of a

parametric family of n-variate copulas satisfying the four desirable properties in Section
2. Following the approach through mixtures of independent conditional distributions
described in Section 2.2, it is first necessary to focus on a simple but flexible one-
parameter family of bivariate copulas, called linear Spearman copula, which is intro-
duced in Section 3.1. The analytical evaluation of the distribution and stop-loss trans-
form of bivariate sums following a linear Spearman copula, often required in actuarial
and financial calculations, is presented in Section 3.2.
The dependence parameter of the linear Spearman copula is Spearman’s grade cor-
relation coefficient. In practice, however, often only Pearson’s linear correlation coeffi-
cient is available. Stochastic relationships between these two parameters, which allow
parameter estimation from each other, have been derived by Hürlimann [25].

3.1. The linear Spearman copula. We consider a one-parameter family of copu-

las Cθ (u, v), which is able to model continuously a whole range of dependence be-
tween the lower Fréchet bound C−1 (u, v) = max(u + v − 1, 0), the independent copula
C0 (u, v) = uv, and the upper Fréchet bound C1 (u, v) = min(u, v). Such families are
called inclusive or comprehensive (Devroye [14, page 581]). A number of inclusive fami-
lies of copulas are well known, namely, those by Fréchet [23], Plackett [42], Mardia [39],
Clayton [8], and Frank [21]. Another one, which is similar but not identical to the convex
family of Fréchet [23], is the linear Spearman copula defined by

Cθ (u, v) = 1 − |θ| · C0 (u, v) + |θ| · Csgn(θ) (u, v). (3.1)

For θ ∈ [0, 1], this copula is family B11 in Joe [30, page 148]. It represents a mixture
of perfect dependence and independence. If X and Y are uniform (0, 1), Y = X with
probability θ, and Y is independent of X with probability 1 − θ, then (X, Y ) has the
linear Spearman copula. This distribution has been first considered by Konijn [34] and
motivated by Cohen [9] along Cohen’s kappa statistic (see Hutchinson and Lai [29,
Section 10.9]). For the extended copula, the chosen nomenclature linear refers to the
350 WERNER HÜRLIMANN

piecewise linear sections of this copula, and Spearman refers to the fact that the grade
correlation coefficient ρS by Spearman [49] coincides with the parameter θ. This follows
from the calculation

1 1 
ρS = 12 · Cθ (u, v) − uv du dv = θ. (3.2)
0 0

The linear Spearman copula, which leads to the linear Spearman bivariate distribution,
has a singular component, which, according to Joe, should limit its field of applicabil-
ity. Despite this, it has many interesting and important properties, and is suitable for
computation. Moreover, it is a good competitor in fitting bivariate cumulative returns,
as shown by Hürlimann [28].
For the reader’s convenience, we describe first two extremal properties. Kendall’s tau
for this copula is defined as follows:

1 1
∂ ∂
τ = 1−4· Cθ (u, v) · Cθ (u, v)du dv
0 0 ∂u ∂v
(3.3)
1  
= ρS · 2 + sgn ρS ρS .
3

Invert this to get

 √
−1 + 1 + 3τ, τ ≥ 0,
ρS =
1 − √1 − 3τ,
(3.4)
τ ≤ 0.

Relate this to the convex two-parameter copula by Fréchet [23] defined by

Cα,β (u, v) = β · C−1 (u, v) + (1 − α − β) · C0 (u, v)

(3.5)
+ α · C1 (u, v), α, β ≥ 0, α + β ≤ 1.

Since ρS = α − β and τ = ((α − β)/3)(2 + α + β) for this copula, one has the inequalities


τ ≤ ρS ≤ −1 + 1 + 3τ, τ ≥ 0,
 (3.6)
1 − 1 − 3τ ≤ ρS ≤ τ, τ ≤ 0.

The linear Spearman copula satisfies the following extremal property. For τ ≥ 0, the
upper bound for ρS in Fréchet’s copula is attained by the linear Spearman copula, and
for τ ≤ 0, it is the lower bound, which is attained.
In case τ ≥ 0, a second more important extremal property holds, which is related to a
conjectural statement. Recall that Y is stochastically increasing on X, written SI(Y |X), if
Pr(Y > y | X = x) is a nondecreasing function of x for all y. Similarly, X is stochastically
increasing on Y , written SI(X|Y ), if Pr(X > x | Y = y) is a nondecreasing function of
y for all x. (Note that Lehmann [36] speaks instead of positive regression dependence.)
If X and Y are continuous random variables with copula C(u, v), then one has the
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 351

equivalences (Nelsen [41, Theorem 5.2.10])

∂
SI(Y |X) ⇐⇒ C(u, v) is nonincreasing in u for all v,
∂u
(3.7)
∂
SI(X|Y ) ⇐⇒ C(u, v) is nonincreasing in v for all u.
∂v

The Hutchinson-Lai conjecture consists of the following statement. If (X, Y ) satisfies

the properties (3.7), then ρS satisfies the inequalities

  
3
−1 + 1 + 3τ ≤ ρS ≤ min τ, 2τ − τ 2 . (3.8)
2

The upper bound 2τ −τ 2 is attained for the one-parameter copula introduced by Kimel-
dorf and Sampson [33] (see also Hutchinson and Lai [29, Section 13.7]). The lower bound
is attained by the linear Spearman copula, as shown already by Konijn [34, page 277].
Alternatively, if the conjecture holds, the maximum value of Kendall’s tau given by ρS
is attained for the linear Spearman copula. Note that the upper bound ρS ≤ (3/2)τ
has been disproved recently by Nelsen [41, Exercise 5.36]. The remaining conjecture
√
−1 + 1 + 3τ ≤ ρS ≤ 2τ − τ 2 is still unsettled (however, see Hürlimann [27] for the case
of bivariate extreme value copulas).
As an important modeling characteristic, we show that the linear Spearman copula
leads to a simple tail dependence structure. Using (2.7), one obtains

1 − 2u + Cθ (u, u)
λ(X, Y ) = lim = lim (1 − u + θu) = θ. (3.9)
u→1− 1−u u→1−

Therefore, unless X and Y are independent, a linear Spearman couple is always asymp-
totically dependent. This is a desirable property in insurance and financial modeling,
where data tend to be dependent in their extreme values. In contrast to this, the ubiq-
uitous Gaussian copula always yields asymptotic independence, unless perfect correla-
tion holds (Sibuya [47], Resnick [43, Chapter 5], and Embrechts et al. [20, Section 4.4]).

3.2. Distribution and stop-loss transform of bivariate sums. For several purposes
in actuarial science and finance, it is of interest to have analytical expressions for the
distribution and stop-loss transform of dependent sums S = X + Y , denoted respec-
tively by FS (x) = Pr(S ≤ x) and πS (x) = E[(S − x)+ ]. If (X, Y ) follows a linear Spearman
bivariate distribution, we show in Theorem 3.4 that the evaluation of these quantities
depends on the knowledge of the quantiles and stop-loss transform of the independent
sum of X and Y , denoted by S ⊥ = X ⊥ + Y ⊥ , where (X ⊥ , Y ⊥ ) represents an independent
version of (X, Y ) such that X ⊥ and Y ⊥ are independent and X ⊥ and Y ⊥ are identi-
cally distributed as X and Y . Similarly, if (X + , Y + ) is a comonotone version of (X, Y )
with bivariate distribution F(X + ,Y + ) (x, y) = min{FX (x), FY (y)}, the sum is denoted by
S + = X + + Y + , while if (X − , Y − ) is a countercomonotone version such that (X − , −Y − )
is a comonotone couple, the sum is denoted by S − = X − + Y − . We assume throughout
that the margins have continuous and strictly increasing distribution functions, hence
the quantile functions are uniquely defined. A linear Spearman random couple (X, Y )
with Spearman coefficient θ is denoted by LSθ (X, Y ).
352 WERNER HÜRLIMANN

Lemma 3.1. For each LSθ (X, Y ), θ ∈ [−1, 1], the distribution and stop-loss transform
of the sum S = X + Y satisfy the relationships

FS (x) = 1 − |θ| · FS ⊥ (x) + |θ| · FS sgn(θ) (x),

(3.10)
πS (x) = 1 − |θ| · πS ⊥ (x) + |θ| · πS sgn(θ) (x).

Proof. This follows without difficulty from the representation (3.1).

Lemma 3.2. Suppose (X + , Y + ), respectively (X − , −Y − ), is a comonotone couple with

continuous and strictly increasing marginal distributions. Then, for all u ∈ (0, 1), one
has the additive relations

FS−1 −1 −1
+ (u) = FX (u) + FY (u), FS−1 −1 −1
− (u) = FX (u) + FY (1 − u), (3.11)
 −1   −1   −1 
πS + FS + (u) = πX FX (u) + πY FY (u) , (3.12)
 −1     
πS − FS − (u) = πX FX−1 (u) + E[Y ] − FY−1 (1 − u) − πY FY−1 (1 − u) . (3.13)

Proof. If (X + , Y + ) is a comonotone couple, it belongs to the copula C(u, v) =

min(u, v). Inserting the expression for the conditional distribution FY |X=x (y) =
(∂C/∂u)[FX (x), FY (y)] = 1{x≤QX [FY (y)]} into the formula for the distribution of a sum

∞
FX+Y (s) = FY |X=x (s − x)dFX (x) (3.14)
−∞

and making the change of variable FX (x) = u, one obtains

us
FX+Y (s) = du = us , (3.15)
0

where us solves the equation FX−1 (us ) + FY−1 (us ) = s. Therefore, (3.15) is equivalent to
−1
FX+Y (us ) = FX−1 (us ) + FY−1 (us ), and since s is arbitrary, the first part of (3.11) is shown.
The second part of (3.11) follows similarly using the copula C(u, v) = max(u+v −1, 0).
To show (3.13), consider the “spread” function of a random variable X defined by

  ∞
TX (u) := πX FX−1 (u) = x − FX−1 (u) dFX (x)
−1 (u)
FX
1
(3.16)
= FX−1 (t) − FX−1 (u) dt.
u

Using (3.11), one immediately obtains from (3.16) that

 
TS + (u) = πS + FS−1
+ (u)

1 1
= FX−1 (t) − FX−1 (u) dt + FY−1 (t) − FY−1 (u) dt (3.17)
u u
   
= πX FX−1 (u) + πY FY−1 (u) ,
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 353

which shows the first part of (3.13). For the second part of (3.13), one similarly obtains

 
TS − (u) = πS − FS−1
− (u)

1 1
= FX−1 (t) − FX−1 (u) dt + FY−1 (1 − t) − FY−1 (1 − u) dt
u u
  1
= πX FX−1 (u) + FY−1 (z) − FY−1 (1 − u) dz (3.18)
0
1
− FY−1 (z) − FY−1 (1 − u) dz
1−u
   
= πX FX−1 (u) + E[Y ] − FY−1 (1 − u) − πY FY−1 (1 − u) .

The Lemma is shown.

Remark 3.3. In case of continuous and strictly increasing margins, the first additive
relations in (3.11) and (3.13) extend easily to n-variate sums S + = X1+ + · · · + Xn+ of
mutually comonotonic random variables:


n
  n
 
FS−1
+ (u) = FX−1
i
(u), πS + FS−1
+ (u) = πXi FX−1
i
(u) . (3.19)
i=1 i=1

For the quantile, this is already found by Landsberger and Meilijson [35]. Both relations
are given by Dhaene et al. [16], Kaas et al. [32], and Hürlimann [26]. Our elementary
approach has the advantage to yield the additional result for S − . These relations are of
great importance in economic risk capital evaluations using the value-at-risk and con-
ditional value-at-risk measures. They imply that the maximum CVaR for the aggregate
loss L = L1 +· · ·+Ln of a portfolio L = (L1 , . . . , Ln ) with fixed marginal losses is attained
at the portfolio with mutually comonotone components, and it is equal to the sum of
the CVaR of its components (Hürlimann [26, Theorems 2.2 and 2.3]):

     n
 
max CVaRα [L] = CVaRα L+ = CVaRα Li . (3.20)
i=1

In contrast to this, the maximum VaR of a portfolio with fixed marginal losses is not
attained at the portfolio with mutual components. This assertion is related to Kol-
mogorov’s problem treated by Makarov [38], Rüschendorf [45], Frank et al. [22], Denuit
et al. [13], Durrleman et al. [18], Luciano and Marena [37], Cossette et al. [10], and Em-
brechts et al. [19]. In the comonotonic situation, one has with (3.19) only the additive
relation

   n
 
VaRα L+ = VaRα Li . (3.21)
i=1

Theorem 3.4. For each LSθ (X, Y ), θ ∈ [−1, 1], the distribution and stop-loss trans-
form of the sum S = X + Y are determined as follows. For each u ∈ [0, 1], one has with
354 WERNER HÜRLIMANN

uθ = (1/2)[1 − sgn(θ)] + sgn(θ)u the formulas

   
FS FX−1 (u) + FY−1 uθ = 1 − |θ| · FS ⊥ FX−1 (u) + FY−1 uθ + |θ| · u,
   
πS FX−1 (u) + FY−1 uθ = 1 − |θ| · πS ⊥ FX−1 (u) + FY−1 uθ

    (3.22)
+ |θ| · πX FX−1 (u) + sgn(θ) · πY FY−1 uθ

1   
+ 1 − sgn(θ) · E[Y ] − FY−1 uθ .
2

Proof. Apply Lemmas 3.1 and 3.2.

Though not always of simple form, analytical expressions for one of density, distri-
bution, and stop-loss transform of the independent sum S ⊥ = X ⊥ +Y ⊥ from parametric
families of margins often exist. A numerical evaluation using computer algebra systems
is then easy to implement. For example, this is possible for the often encountered mar-
gins from the normal, gamma, and lognormal families of distributions (see Johnson et
al. [31] and Hürlimann [25]).

4. A multivariate generalization. We restrict our attention to the construction of

n-variate distributions F (x1 , . . . , xn ) whose positive dependent bivariate margins
Fr s (xr , xs ) belong to linear Spearman copulas with general Spearman coefficients ρrSs ∈
[0, 1]. The more complicated case ρrSs ∈ [−1, 1] has been illustrated for trivariate dis-
tributions by Hürlimann [25, Section 9].
For each i ∈ {1, . . . , n}, the n-variate distribution F (i) (x1 , . . . , xn ) belongs to the n-
variate copula C (i) (u1 , . . . , un ) and has bivariate margins Fij (xi , xj ), j = i, which belong
to the linear Spearman copula

Cij ui , uj = 1 − θij ui uj + θij min ui , uj , (4.1)

where θij ∈ [0, 1], and by symmetry, θji = θij . Applying the method of mixtures of
independent conditional distributions, one considers the conditional distributions

∂Cij  
Fj|i xj |xi = F i xi , F j xj
∂ui (4.2)
= 1 − θij · Fj xj + θij · 1{xi ≤F −1 [Fj (xj )]} .
i

Denote by θ (i) = (θij , j = i) the vector of the (1/2)n(n − 1) dependence parameters,

and let ∆(i) be the set of the 2n−1 vectors δ(i) = (δij , j = i), where δij ∈ {0, 1}. Then the
n-variate mixture F (i) of independent conditional distributions (4.2), defined in (2.9),
belongs to the n-variate copula
  
  1−δij δij j=i (1−δij )
(i)
C u1 , . . . , u n = 1 − θij θij · ui
δ(i) ∈∆(i) j=i
     (4.3)
 1−δij 1−δij 1− j=i (1−δij )
· uj · min uj , ui .
j=i
j=i
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 355

This representation shows that each C (i) is a convex combination of the n different
elementary copulas
 

n
r
EC u1 , . . . , un = min uj · ui , r = 0, 2, 3, . . . , n. (4.4)
1≤j≤r
i=r +1

A distribution with copula EC r is the convolution of a distribution with r comonotone

components and a distribution with n−r independent components. This observation is
useful for the analytical evaluation of the distribution and stop-loss transform of depen-
dent sums from convex combinations of these elementary copulas (see Theorem 5.1).
(i)
To obtain the bivariate copula Cr s (ur , us ), which belongs to the bivariate margin Fr s
of F (i) , one sets uk = 1 for all k = r , s in (4.3) to get the bivariate linear Spearman copula

 1 − θr s ur us + θr s min ur , us , i = r or i = s,
Cr(i) ur , us = (4.5)
s  1 − θ θ u u + θ θ min u , u , i = r , s.
ir is r s ir is r s

(i)
It follows that the Spearman correlation coefficient of Cr s is equal to

 θr s , i = r or i = s,
(i)
ρS r s = (4.6)
θ θ , i = r , s.
ir is

Therefore, for r = i or s = i, the distribution F (i) has the desired linear Spearman
bivariate margins Fr s with Spearman’s rho θr s . Unfortunately, for the other indices
r , s = i, the bivariate margin Fr s has the Spearman correlation coefficient θir θis , which
in general differs from the parameter θr s . To construct an n-variate distribution F ,
whose linear Spearman bivariate margins Fr s may have more general Spearman’s rho
ρrSs ∈ [0, 1], we consider the convex combination of the copulas C (i) , i ∈ {1, . . . , n},
defined for all θ = (θij ), θij = 1, by
 
1 
n  1
C(u1 , . . . , un ) = · · C (i) u1 , . . . , un ,
cn (θ) i=1 j=i 1 − θij
(4.7)

n 
1
cn (θ) = .
i=1 j=i
1 − θij

If θij = 1 for all i, j, one sets C(u1 , . . . , un ) = min1≤j≤n (uj ), which is the copula of n
comonotone random variables. Using (4.6), one sees that the linear Spearman bivariate
margins Fr s have Spearman’s rho determined by
 
1 
n  1  
ρrSs = · · θr s εir + εis + θir θis 1 − εir 1 − εis , (4.8)
cn (θ) i=1 j=i 1 − θij

j j j
where εi is a Kronecker symbol such that εi = 1 if j = i and εi = 0 if j = i. Though it
S
has not been shown that the functions (4.8), which map θ = (θij ), θij = 1, to ρ S = (ρij ),
S
ρij = 1, are one-to-one, the constructed copula (4.7) is sufficiently general and simple to
yield tractable positive dependent n-variate distributions with bivariate margins equal
356 WERNER HÜRLIMANN

or at least close to given linear Spearman bivariate margins. By appropriate choice of the
univariate margins, say gamma or lognormal margins, the obtained parametric family
of n-variate copulas satisfies the four desirable properties in Section 2.
To obtain expressions which can be implemented, insert (4.3) into (4.7) and rearrange
terms to get the formula

1
C u1 , . . . , u n =
cn (θ)
  

n
· n· ui
(4.9)
i=1
   

n  
r
θi1 ij 
+ · min uij · uk .
r =2 i1 =···=ir j=2
1 − θi1 ij 1≤j≤r
k∉{i1 ,...,ir }

In particular, this shows that the constructed n-variate copula is a convex combination
of elementary copulas of the type defined in (4.4). Regrouping these terms further, one
obtains simpler expressions. For example, if n = 3, one has

    
 θij 
C u1 , u2 , u3 = c3 (θ)−1 · 3u1 u2 u3 + 2 min ui , uj uk
i<j
1 − θij k=i,j
   (4.10)
 θij θik
+ min u1 , u2 , u3 .
k=j=i
1 − θij 1 − θik

5. Properties of the multivariate linear Spearman copula. In the present section,

some interesting and useful properties of the multivariate Spearman copula will be de-
rived. We begin with the analytical exact evaluation of the distribution and stop-loss
transform of dependent sums from an n-variate distribution with copula (4.9). In gen-
eral, suppose an n-dimensional copula is a convex combination of other copulas, say


C = λj C j . Then the distribution FS (s) and stop-loss transform πS (s) of dependent

n j
sums S = i=1 Xi from the multivariate model with copula C are the convex combina-
tions of the distributions FS j (s) and stop-loss transform πS j (s) of the dependent sums

n j

S j = i=1 Xi from the multivariate models with copulas C j , that is, FS (s) = λj FS j (s)


and πS (s) = λj πS j (s). Since this result applies to the n-variate copula (4.9), it suf-
fices, up to permutations of variables, to discuss the evaluation of the distribution and
stop-loss transform of sums from an elementary copula of the type EC r in (4.4).
A multivariate distribution with copula EC 0 belongs to a random vector (X1 , . . . , Xn )
with independent components, while a distribution with copula EC n belongs to a ran-
dom vector with comonotone components. For EC 0 , the distribution and stop-loss
transform of sums are obtained using convolution formulas, while for EC n , they are ob-
tained through the addition of the same quantities from the individual components as
stated in Remark 3.3. For example, the case of gamma marginals has been thoroughly
discussed by Hürlimann [26]. There remains the derivation of summation formulas for
the other n−2 copulas. We restrict the attention to nonnegative random variables with
continuous and strictly increasing distributions whose densities exist.
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 357

Given random variables Xi , 1 ≤ i ≤ n, with fixed marginal distributions Fi (x), sup-

pose that the distribution of the random vector (X1+ , . . . , Xr+ , Xr⊥+1 , . . . , Xn⊥ ) belongs to the
copula EC r , 2 ≤ r ≤ n − 1. More precisely, X1+ , . . . , Xr+ represent the comonotonic ver-
sion of X1 , . . . , Xr , Xr⊥+1 , . . . , Xn⊥ represent the independent version of Xr +1 , . . . , Xn , and
(X1+ , . . . , Xr+ ) is independent from Xi , r + 1 ≤ i ≤ n.

Theorem 5.1. Suppose (X1+ , . . . , Xr+ , Xr⊥+1 , . . . , Xn⊥ ) is a random vector whose distri-
bution belongs to the copula EC r , 2 ≤ r ≤ n − 1. Assume that the continuous and
strictly increasing marginal distributions Fi (x) with support [0, ∞) have densities fi (x),

n
1 ≤ i ≤ n, and set X = i=r +1 Xi⊥ . Then the distribution and stop-loss transform of the

r
sum S = X + i=1 Xi+ are determined by the formulas

   
us 
r 
r
 −1
FS (s) = ufX s − Fi−1 (u) · fi Fi−1 (u) du, (5.1)
0
i=1 i=1
   
us 
r 
r
 −1 −1
πS (s) = E[S] − s + uFX s − Fi−1 (u) · fi Fi (u) du, (5.2)
0
i=1 i=1

where us solves the equation


r
Fi−1 us = s. (5.3)
i=1

r
Proof. Set Y = i=1 Xi+ and use Dhaene and Goovaerts [15, Lemma 2] to obtain the
formulas πS (s) = E[S] − s + I(s) and FS (s) = 1 + (d/ds)πS (s) = I  (s), with

s
I(s) = F(X,Y ) (x, s − x)dx. (5.4)
0

By assumption, X is independent from Y , hence F(X,Y ) (x, w) = FX (x)·FY (w). Inserting

in (6.4) and making the change of variable FY (t) = u, one successively obtains

s FY (s)  
I(s) = FX (s − t)FY (t)dt = uFX s − QY (u) · QY (u)du
0 0
   r  (5.5)
us r   
= uFX s − Fi−1 (u) · fi Fi−1 (u) du,
0
i=1 i=1

r
where the last equality follows from the fact that Y = i=1 Xi+ is a comonotone sum,
and the definition of us in (5.3). The formula (5.2) is shown. Formula (5.1) follows from

s
FS (s) = I  (s) = FX (0) · FY (s) + fX (s − t)FY (t)dt
0
s (5.6)
= fX (s − t)FY (t)dt
0

making the same change of variable FY (t) = u.

358 WERNER HÜRLIMANN

Next, taking pattern from the recent contributions by Denuit et al. [12, Theorem
3.1] and Hürlimann [26, Remark 2.1], it is important to know if the constructed n-
variate “positive dependent” distributions associated to random vectors (X1 , . . . , Xn ) are

n
such that the dependent sums S = i=1 Xi are always bounded in convex order by the

n
n
corresponding independent sum S ⊥ = i=1 Xi⊥ and the comonotone sum S + = i=1 Xi+ .

Theorem 5.2. Suppose (X1 , . . . , Xn ) is a random vector whose distribution belongs to

the copula (4.9). Then one has the stochastic inequalities S ⊥ ≤sl ≤ S ≤sl S + .

Proof. Since the copula (4.9) is a convex combination of elementary copulas of

the type (4.4) and the operation of building dependent sums from random vector with
such copulas is preserved under stop-loss order, it suffices to show the assertion for
the elementary ECnr in (4.4) (the lower dimension is added for distinction). One applies
induction on n. For n = 2, the result is trivial because EC20 yields S ⊥ and EC22 yields S + .
Assume that the result holds for the dimension n and show it for n + 1. One has the
product representation


EC r u1 , . . . , un · un+1 , r ∈ {0, 2, 3, . . . , n},
r n
ECn+1 u1 , . . . , un+1 = (5.7)
min u , . . . , u · u r = n + 1,
1 n n+1 ,

n
which shows that Xn+1 is independent of (X1 , . . . , Xn ), hence also of Sn = i=1 Xi . Since
the stop-loss order is preserved under convolutions, it follows from the induction as-
⊥ ⊥ ⊥ +
sumption Sn⊥ ≤sl ≤ Sn ≤sl Sn+ that Sn+1 = Sn⊥ + Xn+1 ≤sl ≤ Sn + Xn+1 = Sn+1 ≤sl Sn+1 .

6. Multivariate conditional value-at-risk and risk allocation. Given a random vari-

able X with survival function SX (x) = Pr(X > x), consider the univariate stop-loss trans-
∞
form defined by πX (x) = E[(X − x)+ ] = x SX (t)dt. It is related to the mean excess
function mX (x) = E[X − x | X > x] through πX (x) = SX (x) · mX (x). The extension of
these notions to a multivariate setting is straightforward.
Let X = (X1 , . . . , Xn ) be a random vector with n-variate survival function SX (x) =
Pr(X1 > x1 , . . . , Xn > xn ). Then the ith component of the stop-loss transform vector
∞
(πX1 (x), . . . , πXn (x)) is defined by πXi (x) = xi SX (x1 , . . . , xi−1 , t, xi+1 , . . . , xn )dt, i = 1, . . . , n.
1 n i
It is related to the mean excess vector (mX (x), . . . , mX (x)), mX (x) = EXi − xi | Xj >
i i
xj , j = 1, . . . , n , through the relationships πX (x) = SX (x) · mX (x), i = 1, . . . , n.
In the univariate case, the conditional value-at-risk of X to the confidence level α
satisfies the relations

 
CVaRα [X] = E X | X > VaRα [X]
 
= VaRα [X] + mX VaRα [X]
(6.1)
1  
= VaRα [X] + πX VaRα [X] .
1−α

In the multivariate situation, the conditional value-at-risk vector to the confidence level
α, denoted by CVaRα [X] = (CVaR1α [X], . . . , CVaRn
α [X]), satisfies similarly the defining
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 359

relations

   
CVaRiα [X] = E Xi | Xj > VaRα Xj , j = 1, . . . , n
i
 
= VaRα [Xi ] + mX VaRα [X]
  (6.2)
  π i VaRα [X]
= VaRα Xi + X  , i = 1, . . . , n,
SX VaRα [X]

where VaRα [X] = (VaRα [X1 ], . . . , VaRα [Xn ]) defines a value-at-risk vector.
We point out the usefulness of this multivariate extension to the evaluation of the
economic risk capital of a portfolio of risks X = (X1 , . . . , Xn ). It is natural to define the

n
economic risk capital of the aggregate risk S = i=1 Xi as a multivariate conditional
value-at-risk vector to the confidence level α by setting

     
CVaRα S|X = E S | Xj > VaRα Xj , j = 1, . . . , n . (6.3)

Allocating to the risk Xi the ith component of the conditional value-at-risk vector (6.2)
defines the multivariate conditional value-at-risk allocation principle, which by (6.3)
turns out to be an additive allocation principle, which satisfies the identity

 

n 
n
 
CVaRα Xi |X = CVaRiα Xi . (6.4)
i=1 i=1

The proposed risk allocation rule yields a simple solution to the difficult risk allocation
problem (e.g., Tasche [50] and Denault [11]).
For continuous univariate marginals, the univariate conditional value-at-risk mea-

n
sures CVaRα [ i=1 Xi ] and CVaRα [Xi ], i = 1, . . . , n, are coherent risk measures in the
sense of Artzner et al. [4, 5] (e.g., Acerbi [1] and Acerbi and Tasche [2, 3]). In view of
this fact, it seems important to derive connections between these risk measures and
their multivariate counterparts. The following main result yields an interesting and
surprising result.

Theorem 6.1. Suppose X = (X1 , . . . , Xn ) is a random vector with strictly increasing

continuous margins whose distribution belongs to the multivariate linear Spearman cop-
ula (4.9). Then one has CVaRiα [X] = CVaRα [Xi ], i = 1, . . . , n, that is, each component of
the conditional value-at-risk vector coincides with the conditional value-at-risk of the
corresponding risk component.

Proof. In a first step, we show the result for X with elementary copula (4.4). For
simplicity, set xi,α = VaRα [Xi ], i = 1, . . . , n, xα = (x1,α , . . . , xn,α ), and ε = 1−α. From the
form of the survival copula (4.4), one obtains the survival function

 n 
  
SX (x) = min SXi xi S Xi xi . (6.5)
1≤j≤r
i=r +1
360 WERNER HÜRLIMANN

Using that SXi (xi ) = ε and the definition of the stop-loss transform components, one
obtains

SX xα = εn−r +1 , πXi xα = πXi xi,α εn−r . (6.6)

Inserted in (6.2) one gets, CVaRiα [X] = xi,α + (1/ε)πXi (xi,α ) = CVaRα [Xi ], which shows
the result in this special case. The copula (4.9) is a convex combination of elemen-
tary copulas EC r (up(1) , . . . , up(n) ) with appropriate permutations (p(1), . . . , p(n)) of
(1, . . . , n), for which (6.6) still holds. Since multivariate survival functions and stop-
loss transforms are preserved under convex combinations, one sees that ε · πXi (xα ) =
πXi (xi,α ) · SX (xα ) for all X with copula (4.9), which implies the desired result.

This result has a remarkable economic implication. In a world of multivariate lin-

ear Spearman distributed risks with copula (4.9), the evaluation of the economic risk
capital of the aggregate risk using (6.3) automatically yields a “fair” risk allocation
in the sense that each risk component becomes allocated to its coherent univariate
conditional value-at-risk. Of course, this rule is only applicable in situations where
additivity is a desirable property. Whenever a discount for aggregation is prevalent
and a subadditive risk allocation is preferred, the economic risk capital is better evalu-

n
ated using CVaRα [ i=1 Xi ]. In this situation, a nonnegative diversification effect, defined

n
n
by CVaRα [ i=1 Xi ] − i=1 CVaRα [Xi ], which can be evaluated using Theorem 5.1, will
occur.
As a topic for future research, it is interesting to study the largest class of distribu-
tions or copulas for which the property of Theorem 6.1 holds. As the following example
shows, this class is bigger than the convex hull of all permuted elementary copulas (4.4).

Example 6.2 (a bivariate distribution with the property of Theorem 6.1). Consider
a random couple (X, Y ) with survival function S(X,Y ) (x, y) = (1/2)e−x−y + e− max(x,y) ,
1 −y
x, y ≥ 0. One has π(X,Y ) (x, y) = S(X,Y ) (x, y) + (1/2)(y − x)+ e , SX (x) = πX (x) =
(1/2)e−x , and SY (y) = πY (y) = (1/2)e−y . An elementary calculation shows that

   
CVaR1α (X, Y ) = CVaRα [X] = CVaR2α (X, Y )
  (6.7)
= CVaRα [Y ] = 1 − ln 2(1 − α) .

For a more comprehensive understanding, it will be necessary to discuss in the future

the impact of the dependence structure of various multivariate distributions on the
calculation of the multivariate conditional value-at-risk vector, which in general differs
from the vector of the univariate conditional value-at-risk measures. We illustrate this
issue at some simple parametric families of bivariate distributions with exponential
margins.
Let (X, Y ) be a random couple with exponential survival margins SX (x) = exp(−x/µX )
and SY (y) = exp(−y/µY ). For comparison, we use the Marshall-Olkin [40] bivariate ex-
ponential, which exhibits positive dependence, the negatively dependent exponential
Gumbel [24], and the bivariate linear Spearman, which allows both positive and negative
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 361

Table 6.1. Conditional value-at-risk comparisons.

Univariate CVaR Bivariate CVaR

ρ = 0.5 Marshall-Olkin Linear Spearman Marshall-Olkin Linear Spearman
X 4.80445 4.80445 4.80445 4.80445
Y 4.20389 4.20389 4.47718 4.20389
S 7.56412 8.03684 9.28162 9.00835
ρ = −0.3 Gumbel Linear Spearman Gumbel Linear Spearman
X 4.80445 4.80445 4.18356 4.80445
Y 4.20389 4.20389 3.66058 4.20389
S 5.37364 5.82982 7.84414 9.00835

dependence. The conditional value-at-risk measures of the margins are CVaRα [X] =
(1 − ln ε)µX , CVaRα [Y ] = (1 − ln ε)µY , with ε = 1 − α. Formulas for the other risk mea-
sures are derived in a straightforward way. For the dependent sum S = X +Y , we use the
formula CVaRα [S] = FS−1 (α) + (1/ε)πS [FS−1 (α)], which requires expressions for FS (s)
and πS (s). Pearson’s rho is denoted by ρ. A typical numerical example is summarized
in Table 6.1, where µX = 0.85715 and µY = 0.75.

Marshall-Olkin bivariate exponential [40]. The survival distribution of this

model and the required formulas to evaluate Table 6.1 are summarized by the following
formulas:

S(x, y) = exp − αx − βy − γ max(x, y) , α, β > 0, γ ≥ 0,

1 1 γ
µX = , µY = , ρ= ≥ 0,
α+γ β+γ α+β+γ

  1 γ
CVaR1α (X, Y ) = 1− εα(µY −µX )+ − ln ε · µX ,
α α+γ

  1 γ
CVaR2α (X, Y ) = 1− εα(µX −µY )+ − ln ε · µY ,
β β+γ
  (6.8)
1 1 1
FS (x) = 1 − (α + β + γ) + e−(1/2)(α+β+γ)x
2 α+γ −β β+γ −α
β α
+ e−(α+γ)x + e−(β+γ)x ,
α+γ −β β+γ −α
 
1 1
πS (x) = + e−(1/2)(α+β+γ)x
α+γ −β β+γ −α
   
1 1 1 1
− − e−(α+γ)x − − e−(β+γ)x .
α+γ −β α+γ β+γ −α β+γ

Bivariate exponential Gumbel [24]. The survival distribution of this model and
the required formulas to evaluate Table 6.1 are summarized by the following formulas:

S(x, y) = exp(−αx − βy − θxy), α, β > 0, 0 ≤ θ ≤ αβ,

∞
1 1 e−αt dt
µX = , µY = , ρ = αβ − 1 ≤ 0,
α β 0 β + θt
362 WERNER HÜRLIMANN


  1
CVaR1α (X, Y ) = − ln ε µX ,
1 − θµX µY ln ε

  1
CVaR2α (X, Y ) = − ln ε µY
1 − θµX µY ln ε
FS (x) = 1 − e−αx − e−βx
        
θ α−β θ α−β 2
+ g  (x) − β + x+ g(x) exp −βx − x+ ,
4 θ 4 θ
   
−αx −βx θ α−β 2
πS (x) = µX e + µY e + g(x) exp −βx − x+ ,
4 θ
∞
    
 θk 1 α − β 2k+1 α − β 2k+1
2k+1
g(x) = x − − (−1) x + .
k=0
k! (2k + 1)22k+1 θ θ
(6.9)

Linear Spearman bivariate exponential. The survival distribution of this

model and the required formulas to evaluate Table 6.1 are summarized by the following
formulas:


(1 − θ)e−αx−βy + θ min e−αx , e−βy , θ ∈ [0, 1],
S(x, y) =

(1 + θ)e−αx−βy − θ max e−αx + e−βy − 1, 0 , θ ∈ [−1, 0],

1 1
µX = , µY = ,
α β (6.10)
 
∞
1  1
ρ = θ 1+ 1 − sgn(θ) 2 y ln(1 − e−βy )e−βy dy ,
2 µY 0
   
CVaR1α (X, Y ) = CVaRα [X], CVaR2α (X, Y ) = CVaRα [Y ] (Theorem 6.1).

The quantities FS (x) and πS (x) are calculated using Theorem 3.4.
Some comments and recommendations are in order. By positive dependence, the
multivariate conditional value-at-risk for the Marshall-Olkin is greater than for the lin-
ear Spearman, while by negative dependence, it is smaller for the Gumbel than for the
linear Spearman. If the risk measure should be invariant with respect to the depen-
dence structure, or if the diversification effect should vanish (additive risk allocation),
we recommend the use of the multivariate conditional value-at-risk for the linear Spear-
man. A discrimination of the risk measure with respect to the dependence structure is
obtained by using either CVaR[S] or CVaR[S|X] for copulas different from the linear
Spearman one. A maximal diversification effect is obtained by using CVaR[S] with the
Marshall-Olkin by positive dependence and with the Gumbel by negative dependence. A
more stable diversification effect is obtained with the linear Spearman. Whether these
observations generalize to other bivariate distributions and extend to the multivariate
situation is left to future investigations.
Finally, we want to mention that the approach of the present paper has a lot of alter-
native significant applications including very recent ones like Cherubini and Luciano
[6, 7].
 MULTIVARIATE FRÉCHET COPULAS AND CONDITIONAL VALUE-AT-RISK 363

References
[1] C. Acerbi, Spectral measures of risk: a coherent representation of subjective risk aversion,
J. Banking and Finance 26 (2002), no. 7, 1505–1518.
[2] C. Acerbi and D. Tasche, Expected shortfall: a natural coherent alternative to value-at-risk,
Econom. Notes 31 (2002), no. 2, 379–388.
[3] , On the coherence of expected shortfall, J. Banking and Finance 26 (2002), no. 7,
1487–1503.
[4] P. Artzner, F. Delbaen, J.-M. Eber, and D. Heath, Thinking coherently, RISK 10 (1997), no. 11,
68–71.
[5] , Coherent measures of risk, Math. Finance 9 (1999), no. 3, 203–228.
[6] U. Cherubini and E. Luciano, Pricing and hedging vulnerable credit derivatives with copulas,
Econom. Notes 32 (2003), no. 2, 219–242.
[7] , Pricing vulnerable options with copulas, J. Risk Finance 5 (2003), no. 1, 27–39.
[8] D. G. Clayton, A model for association in bivariate life tables and its application in epi-
demiological studies of familial tendency in chronic disease incidence, Biometrika 65
(1978), no. 1, 141–151.
[9] J. Cohen, A coefficient of agreement for nominal scales, Educ. Psychol. Meas. 20 (1960),
37–46.
[10] H. Cossette, M. Denuit, J. Dhaene, and É. Marceau, Stochastic approximations of present
value functions, Schweiz. Aktuarver. Mitt. (2001), no. 1, 15–28.
[11] M. Denault, Coherent allocation of risk capital, The Journal of Risk 4 (2001), no. 1, 1–34.
[12] M. Denuit, J. Dhaene, and C. Ribas, Does positive dependence between individual risks in-
crease stop-loss premiums? Insurance Math. Econom. 28 (2001), no. 3, 305–308.
[13] M. Denuit, C. Genest, and É. Marceau, Stochastic bounds on sums of dependent risks, Insur-
ance Math. Econom. 25 (1999), no. 1, 85–104.
[14] L. Devroye, Nonuniform Random Variate Generation, Springer-Verlag, New York, 1986.
[15] J. Dhaene and M. J. Goovaerts, Dependency of risks and stop-loss order, Astin Bull. 26 (1996),
no. 2, 201–212.
[16] J. Dhaene, S. Wang, V. Young, and M. J. Goovaerts, Comonotonicity and maximal stop-loss
premiums, Schweiz. Aktuarver. Mitt. (2000), no. 2, 99–113.
[17] D. Drouet Mari and S. Kotz, Correlation and Dependence, Imperial College Press, London,
2001.
[18] V. Durrleman, A. Nikeghbali, and T. Roncalli, How to get bounds for distribution convolu-
tions? A simulation study and an application to risk management. Working paper,
2000, http://gro.creditlyonnais.fr/content/fr/ home_mc.htm.
[19] P. Embrechts, A. Höing, and A. Juri, Using copulae to bound the value-at-risk for functions
of dependent risks, Finance Stoch. 7 (2003), no. 2, 145–167.
[20] P. Embrechts, A. J. McNeil, and D. Straumann, Correlation and dependence in risk man-
agement: properties and pitfalls, Risk Management: Value at Risk and Beyond (Cam-
bridge, 1998) (M. A. H. Dempster, ed.), Cambridge University Press, Cambridge, 2002,
pp. 176–223.
[21] M. J. Frank, On the simultaneous associativity of F (x, y) and x + y − F (x, y), Aequationes
Math. 19 (1979), no. 2-3, 194–226.
[22] M. J. Frank, R. B. Nelsen, and B. Schweizer, Best-possible bounds for the distribution of
a sum—a problem of Kolmogorov, Probab. Theory Related Fields 74 (1987), no. 2,
199–211.
[23] M. Fréchet, Remarques au sujet de la note précédente, C. R. Acad. Sci. Paris Sér. I Math. 246
(1958), 2719–2720 (French).
[24] E. J. Gumbel, Bivariate exponential distributions, J. Amer. Statist. Assoc. 55 (1960), 698–
707.
[25] W. Hürlimann, Analytical evaluation of economic risk capital and diversification using
linear Spearman copulas, Working paper, 2001, http://www.gloriamundi.org/ and
http://www.mathpreprints.com/math/Preprint/werner.huerlimann/20011125.1/
1/.
 364 WERNER HÜRLIMANN

[26] , Analytical evaluation of economic risk capital for portfolios of gamma risks, Astin
Bull. 31 (2001), no. 1, 107–122.
[27] , Hutchinson-Lai’s conjecture for bivariate extreme value copulas, Statist. Probab.
Lett. 61 (2003), no. 2, 191–198.
[28] , Fitting bivariate cumulative returns with copulas, Comput. Statist. Data Anal. 45
(2004), no. 2, 355–372.
[29] T. P. Hutchinson and C. D. Lai, Continuous Bivariate Distributions, Emphasising Applica-
tions, Rumsby Scientific Publishing, Adelaide, 1990.
[30] H. Joe, Multivariate Models and Dependence Concepts, Monographs on Statistics and Ap-
plied Probability, vol. 73, Chapman & Hall, London, 1997.
[31] N. L. Johnson, S. Kotz, and N. Balakrishnan, Continuous Univariate Distributions, 2nd ed.,
Wiley Series in Probability and Mathematical Statistics: Applied Probability and Sta-
tistics, vol. 1, John Wiley & Sons, New York, 1994.
[32] R. Kaas, J. Dhaene, and M. J. Goovaerts, Upper and lower bounds for sums of random
variables, Insurance Math. Econom. 27 (2000), no. 2, 151–168.
[33] G. Kimeldorf and A. Sampson, One-parameter families of bivariate distributions with fixed
marginals, Commun. in Statist. 4 (1975), 293–301.
[34] H. S. Konijn, Positive and negative dependence of two random variables, Sankhyā 21 (1959),
269–280.
[35] M. Landsberger and I. Meilijson, Co-monotone allocations, Bickel-Lehmann dispersion and
the Arrow-Pratt measure of risk aversion, Ann. Oper. Res. 52 (1994), 97–106.
[36] E. L. Lehmann, Some concepts of dependence, Ann. Math. Statist. 37 (1966), 1137–1153.
[37] E. Luciano and M. Marena, Value at risk bounds for portfolios of non-normal returns, New
Trends in Banking Management (C. Zopoudinis, ed.), Physica Verlag, Heidelberg,
2003, pp. 207–222.
[38] G. D. Makarov, Estimates for the distribution function of the sum of two random variables
with given marginal distributions, Teor. Veroyatnost. i Primenen. 26 (1981), no. 4,
815–817 (Russian), translated in Theory Probab. Appl. 26(1982), 803–806.
[39] K. V. Mardia, Families of Bivariate Distributions, Griffin’s Statistical Monographs and
Courses, Hafner Publishing, Darien, Connecticut, 1970.
[40] A. W. Marshall and I. Olkin, A generalized bivariate exponential distribution, J. Appl. Prob-
ability 4 (1967), 291–302.
[41] R. B. Nelsen, An Introduction to Copulas, Lecture Notes in Statistics, vol. 139, Springer-
Verlag, New York, 1999.
[42] R. L. Plackett, A class of bivariate distributions, J. Amer. Statist. Assoc. 60 (1965), 516–522.
[43] S. I. Resnick, Extreme Values, Regular Variation, and Point Processes, Applied Probability.
A Series of the Applied Probability Trust, vol. 4, Springer-Verlag, New York, 1987.
[44] R. T. Rockafellar and S. Uryasev, Conditional value-at-risk for general loss distributions, J.
Banking and Finance 26 (2002), no. 7, 1443–1471.
[45] L. Rüschendorf, Random variables with maximum sums, Adv. in Appl. Probab. 14 (1982),
no. 3, 623–632.
[46] M. Scarsini, On measures of concordance, Stochastica 8 (1984), no. 3, 201–218.
[47] M. Sibuya, Bivariate extreme statistics. I, Ann. Inst. Statist. Math. Tokyo 11 (1960), 195–210.
[48] M. Sklar, Fonctions de répartition à n dimensions et leurs marges, Publ. Inst. Statist. Univ.
Paris 8 (1959), 229–231.
[49] C. Spearman, The proof and measurement of association between two things, Amer. J. Psych.
15 (1904), 72–101.
[50] D. Tasche, Risk contributions and performance measurement, Techn. Univ., preprint, 1999,
http://www-m4.mathematik.tu-muenchen.de/m4/pers/tasche/

Werner Hürlimann: Aon Re and IRMG (Switzerland) Ltd., Sternengasse 21, CH-4010 Basel
E-mail address: [email protected]
URL: www.geocities.com/hurlimann53

View publication stats