j of risk 2001
TRANSCRIPT
-
8/3/2019 J of Risk 2001
1/34
Coherent allocation of risk capital
Michel Denault*
The allocation problem stems from the diversification effect observed in risk
measurements of financial portfolios: the sum of the risks of many portfolios is
larger than the risk of the sum of the portfolios. The allocation problem is to
apportion this diversification advantage to the portfolios in a fair manner, yield-
ing, for each portfolio, a risk appraisal that takes diversification into account.
Our approach is axiomatic, in the sense that we first argue for the necessary
properties of an allocation principle and then consider principles that fulfill the
properties. Important results from the area of game theory find a direct applica-
tion. Our main result is to provide conceptual justification for a risk allocationprinciple that is based on the gradient of the risk measure. A numerical example
based on the rules of the Securities Exchange Commission for margin compu-
tations is provided.
1. Introduction
As an insurance against the uncertainty of the future net worths of its consti-
tuents, a firm could well, and would often be regulated to, hold an amount of
riskless investments. We will call this buffer the risk capital of the firm. From thefirms perspective, holding an amount of money dormant, ie, in extremely low
risk, low return instruments, is a burden. It is therefore natural to look for a fair
allocation of that burden between the constituents of the firm, especially when
the allocation provides a basis for performance comparisons of the constituents
between themselves. We will refer to the constituents as portfolios, but they
could just as well be business units.
The problem of allocation is interesting and non-trivial because the sum of the
risk capitals of each constituent is usually larger than the risk capital of the firm
taken as a whole, something called the diversification effect. This decrease
of total costs, or rebate, needs to be shared fairly between the constituents.
We stress fairness, as all constituents are from the same firm and none should
receive preferential treatment for the purpose of this allocation exercise.
1
The author expresses special thanks to F. Delbaen, who provided both the initial inspiration
for this work and generous subsequent advice, and to Ph. Artzner for drawing his attention to
Aubins literature on fuzzy games. Discussions with P. Embrechts, H.-J. Lthi, D. Straumann,
and S. Bernegger have been most fruitful. The author would also like to thank the Editor and
an anonymous referee for their useful comments. Finally, he gratefully acknowledges the
financial support of the S.S.H.R.C. (Canada), and of RiskLab (Switzerland), where most of
this research was carried out.
*Assistant Professor, Quantitative Methods; HEC Montral, 3000 Chemin de la Cte-Sainte-
Catherine, Montral, Canada, H3T 2A7.
Denault pages 23/11/01 4:18 pm Page 1
-
8/3/2019 J of Risk 2001
2/34
-
8/3/2019 J of Risk 2001
3/34
2. Risk measure and risk capital
In this paper we follow Artzner et al. (1999) in relating the risk of a firm to the
uncertainty of its future worth. The danger, inherent to the idea of risk, is that the
firms worth reaches such a low net value at a point in the future that it must stop
its activities. Risk is then defined as a random variable Xrepresenting a firms
net worth at a specified point in the future.
A risk measure quantifies the level of risk. Specifically, it is a mapping froma set of random variables (risks) to the real numbers: (X) is the amount of anumraire (eg, cash dollars) which, added to the firms assets, ensures that its
future worth is acceptable to the regulator, the chief risk officer or others. (For a
discussion of acceptable worths see Artzner et al., 1999.) Clearly, the heftier the
required safety net, the riskier the firm is perceived to be. We call (X) the risk
capital of the firm. The risk capital allocation problem is to allocate the amountof risk(X) between the portfolios of the firm.
We will assume that all random variables are from the space L(, A, ) ofbounded random variables over a fixed probability space. The reader who wishes
to do so can generalize the results along the lines of Delbaen (1998).
In their papers, Artzner et al. (1999) and (1997) have suggested a set of prop-
erties that risk measures should satisfy, thus defining the concept of coherent
measures of risk:2
DEFINITION 1 A risk measure :L is coherent if it satisfies the following
properties:
Subadditivity For all bounded random variables X and Y
(X+ Y) (X) + (Y)
Monotonicity For all bounded random variables X, Y such that3X Y
(X) (Y)
Positive homogeneity For all 0 and bounded random variable X
(X) = (X)
Translation invariance For all and bounded random variable X
(X+ B) = (X)
where B is the price, at some point in the future, of a reference, riskless invest-
ment whose price is 1 today.
For example, value-at-risk (VAR) is in general not a coherent risk measure,
although it is so for normally distributed random variables; expected shortfall is
a coherent risk measure.
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
3
Denault pages 23/11/01 4:18 pm Page 3
-
8/3/2019 J of Risk 2001
4/34
The properties that define coherent risk measures are to be understood as
necessary conditions for a risk measure to be reasonable. Let us briefly justify
them. Subadditivity reflects the diversification of portfolios, or that a mergerdoes not create extra risk (Artzner et al., 1999, p. 209). Monotonicity says that
if a portfolio Y is always worth more than X, then Ycannot be riskier than X.
Homogeneity is a limit case of subadditivity, representing what happens when
there is precisely no diversification effect. Translation invariance is a natural
requirement, given the meaning of the risk measure given above and its relation
to the numraire.
In this paper we require risk measures to be coherent; as a consequence,
VAR-based allocation principles, such as component value-at-risk, do not
qualify. We will not be concerned with specific risk measures until our example
in section 7.
3. Coherence of the allocation principle
An allocation principle provides a generic solution to the risk capital allocation
problem. We suggest in this section a set of axioms, which we argue are
necessary properties of a reasonable allocation principle. We will call coherent
an allocation principle that satisfies the set of axioms. The following definitions
are used:
N= {1, 2,, n} is the set of all portfolios of the firm.
Xi, i N, is a bounded random variable representing the net worth at time Tof the i th portfolio of a firm. We assume that the nth portfolio is a risklessinstrument with net worth at time Tequal toXn = B, whereB is the time Tprice of a riskless instrument with price 1 today.
X, the bounded random variable representing the firms net worth at some
point in the future T, is defined asX= ni=1Xi.
A is the set of risk capital allocation problems: pairs (N, ) composed of a setofn portfolios and a coherent risk measure .
K= (X) is the risk capital of the firm.
We can now define:
DEFINITION 2 An allocation principle is a function :A n that maps eachallocation problem (N, ) into a unique allocation:
The condition ensures that the risk capital is fully allocated. The Ki-notation is
used when the arguments are clear from the context.
: ( , )
( , )
( , )
( , )
( )N
N
N
N
K
K
K
K X
n n
i
i N
aM M
1
2
1
2
=
=such that
Michel Denault
Journal of Risk
4
Denault pages 23/11/01 4:18 pm Page 4
-
8/3/2019 J of Risk 2001
5/34
DEFINITION 3 An allocation principle is coherent if for every allocationproblem (N, ) the allocation (N, ) satisfies the three properties:
No undercut
Symmetry If by joining any subset MN\{i,j},4 portfolios i and j bothmake the same contribution to the risk capital, then Ki = Kj.
Riskless allocation
Kn = (B) =
Recall that the n thportfolio is a riskless instrument.
Furthermore, we call non-negative coherent allocation a coherent allocation
which satisfies Ki 0, i N.It is our proposition that the three axioms of Definition 3 are necessary con-
ditions of the fairness, and thus credibility, of allocation principles. In that sense,
coherence is a yardstick by which allocation principles can be evaluated.
The properties can be justified as follows. The no undercut property ensures
that no portfolio can undercut the proposed allocation: an undercut occurs when aportfolios allocation is higher than the amount of risk capital it would face as an
entity separate from the firm. Given subadditivity, the rationale is simple. Upon a
portfolio joining the firm (or any subset thereof), the total risk capital increases by
no more than the portfolios own risk capital: in all fairness, that portfolio cannot
justifiably be allocated more risk capital than it can possibly have brought to the
firm. The property also ensures that coalitions of portfolios cannot undercut, with
the same rationale. The symmetry property ensures that a portfolios allocation
depends only on its contribution to risk within the firm, and nothing else.
According to the riskless allocation axiom, a riskless portfolio should be allocated
exactly its risk measure, which, incidentally, will be negative. It also means that,
all other things being equal, a portfolio that increases its cash position should see
its allocated capital decrease by the same amount.
In the next section we introduce some game theory and see how it applies to
allocation problems.
4. Game theory and allocation to atomic players
Game theory is the study of situations where players adopt various strategies to
best attain their individual goals. Game theory definitions and theorems are
given in sections 4.1 and 4.2 and are then applied to risk capital allocation in
section 4.3.
M N K X i ii Mi M
,
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
5
Denault pages 23/11/01 4:18 pm Page 5
-
8/3/2019 J of Risk 2001
6/34
For now, players will be atomic, meaning that fractions of players are not
considered possible (players who are humans are naturally atomic, but when
players are portfolios fractions of players can be considered). Furthermore, wewill focus here on coalitional games.
DEFINITION 4 A coalitional game (N, c) consists of:
a finite set, N, of n players; and
a costfunction, c, that associates a real number, c(S), with each subset, S, of
N (called a coalition).
We denote by G the set of games with n players.
The goal of each player is to minimize the cost she incurs, and her strategies
consist of agreeing or not to take part in coalitions (including the coalition of all
players).
In the literature the cost function is usually assumed to be subadditive:
c(S T) c(S) + c(T) for all subsets S and TofNwith empty intersection; anassumption which we make as well.
One of the main questions tackled in coalitional games is the allocation of
the cost, c(N), between all players; this question is formalized by the concept of
value:
DEFINITION 5 A value is a function : G n that maps each game (N, c) intoa unique allocation:
Again, the Ki-notation can be used when the arguments are clear from the
context and when it is also clear whether we mean i(N, ) or i(N, c).
4.1 The core of a game
Given the subadditivity ofc, the players of a game have an incentive to form the
largest coalition N, since this brings an improvement of the total cost when
compared with the sum of their individual costs. They need only find a way to
allocate the cost, c(N), of the full coalition,N, between themselves; but in doing
so players still try to minimize their own share of the burden. Player i will even
threaten to leave the coalition Nif she is allocated a share, Ki , of the total cost
that is higher than her own individual cost c({i}). Similar threats may come from
coalitions S N: if i S Ki exceeds c(S), every player i in S could carry anallocated cost lower than his current K
iifS separated fromN.
: ( , )
( , )
( , )
( , )
( )N c
N c
N c
N c
K
K
K
K c N
n n
i
i N
aM M
1
2
1
2
=
=where
Michel Denault
Journal of Risk
6
Denault pages 23/11/01 4:18 pm Page 6
-
8/3/2019 J of Risk 2001
7/34
The set of allocations that do not allow such threat from any player or
coalition is called the core:
DEFINITION 6 The core of a coalitional game, (N, c), is the set of allocations
Knfor which i S Ki c(S)for all coalitions S N.
A condition for the core to be non-empty is the BondarevaShapley theorem.
Let C be the set of all coalitions ofN, and let us denote by 1S n
the
characteristic vector of the coalition S:
A balanced collection of weights is a collection of|C| numbers S in [0,1] suchthat S C S1S = 1N. An interpretation is that each player has one time unit todistribute between the coalitions, and each coalition S is active during a fraction
S of a unit of time, meaning that each player in S devotes her time solely to Sduring that period. Balancedness is then a feasibility condition on the players
distribution of her time.
A game is balanced if S C S c(S) c(N) for all balanced collections ofweights. Then:
THEOREM 1 (BondarevaShapley (Bondareva, 1963; Shapley, 1967)) A coali-
tional game has a non-empty core if and only if it is balanced.
PROOF See, for example, Osborne and Rubinstein (1994).
4.2 The Shapley value
The Shapley value was introduced by Lloyd Shapley (1953) and ever since has
received a considerable amount of interest (see Roth, 1988).
We use the abbreviation i(S) = c(S i) c(S) for any set S N, i S. Twoplayers i and j are interchangeable in (N, c) if either one makes the same
contribution to any coalition S it may join that contains neither i norj : i(S) =
j(S) for each S Nand i,j S. A player i is a dummy if it brings the contri-bution c(i) to any coalition S that does not contain it already: i(S) =c(i). Weneed to define the three properties:
Symmetry If players i andj are interchangeable, (N, c)i = (N, c)j . Dummy player For a dummy player, (N, c)i = c(i). Additivity over games For two games (N, c1) and (N, c2), (N, c1 + c2) =
(N, c1) + (N, c2), where the game (N, c1 + c2) is defined by (c1 + c2)(S) =c1(S) + c2(S) for all S N.
The rationale of these properties will be discussed in the next section. The
axiomatic definition of the Shapley value is then:
11
0S i
i S( ) =
if
otherwise
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
7
Denault pages 23/11/01 4:18 pm Page 7
-
8/3/2019 J of Risk 2001
8/34
DEFINITION 7 (Shapley, 1953) The Shapley value is the only value that satisfies
the properties of symmetry, dummy player, and additivity over games.
Let us now bring together the core and the Shapley value: when does the
Shapley value yield allocations that are in the core of the game? To our knowl-
edge the only relevant results are those of Shapley (1971) and Aubin (1979). The
former involves the property of strong subadditivity:
DEFINITION 8A coalitional game is strongly subadditive if it is based on a strongly
subadditive5 cost function:
c(S) +c(T) c(S T) + c(S T)
for all coalitions S Nand TN.
THEOREM 2 (Shapley, 1971) If a game (N, c) is strongly subadditive, its core
contains the Shapley value.
The second condition that ensures that the Shapley value is in the core is:
THEOREM 3 (Aubin, 1979) If for all coalitions S, | S | 2,
then the core contains the Shapley value.
The implications of these two results are discussed in the next section.
Let us end this section with the algebraic definition of the Shapley value,
which provides both an interpretation (see Shapley, 1953, or Roth, 1988) and an
explicit computational approach.
DEFINITION 9 The Shapley value, KSh, for the game (N, c) is defined as:
(1)
where s = | S |, andCi represents all coalitions of N that contain i.
Note that this requires the evaluation ofc for each of the 2n
possible coalitions,
unless the problem has some specific structure. Depending on what c is, this task
may become impossibly long, even for moderate n.
Ks n s
nc S c S i i N i
S i
Sh =
( ) ( )! ( )!
!( ) ( \ { }) ,
1
C
( ) ( )
1 0S T
T Sc T
Michel Denault
Journal of Risk
8
Denault pages 23/11/01 4:18 pm Page 8
-
8/3/2019 J of Risk 2001
9/34
4.3 Risk capital allocations modelled as coalitional games
Clearly, we intend to model risk capital allocation problems (N, ) as coalitionalgames (N, c). We can associate the portfolios of a firm with the players of a game
and the risk measure with the cost function c :
(2)
Allocation principles naturally become values.
Note that given (2), being coherent and thus subadditive in the sense(X+ Y) (X) + (Y) of Definition 1, implies that c is subadditive in the sensec(S T) c(S) +c(T) given above.
The core
Allocations satisfying the no undercut property lie in the core of the game,
and, if none does, the core is empty. There is only an interpretational distinction
between the two concepts: while a real player can threaten to leave the full
coalitionN, a portfolio cannot walk away from a bank. However, if the allocation
is to be fair, undercutting should be avoided. Again, this holds also for coalitions
of individual players/portfolios.
The non-emptiness of the core is therefore crucial to the existence of coherent
allocation principles. From Theorem 1, we have:
THEOREM 4 If a risk capital allocation problem is modelled as a coalitional game
whose cost function c is defined with a coherent risk measure through (2), thenits core is non-empty.
PROOF Let 0 S 1 for S C, and S C S1S = 1N. Then
By Theorem 1 the core of the game is non-empty. s
S
S S
S i
i S
S i
i SS
S i
S S ii N
c S X
X
X
c N
=
=
=
C C
C
C
( )
( )
,
c S X S N ii S
( ) =
for
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
9
Denault pages 23/11/01 4:18 pm Page 9
-
8/3/2019 J of Risk 2001
10/34
The Shapley value
With the allocation problem modelled as a game, the Shapley value yields a risk
capital allocation principle. Much more, it is a coherent allocation principle, butfor the no undercut axiom. Symmetry is satisfied by definition. The riskless
allocation axiom of Definition 3 is implied by the dummy player axiom: from
our definitions in section 3, the reference, riskless instrument (cash and equiva-
lents) is a dummy player.
Note that additivity over games is a property that the Shapley value possesses
but which is not required of a coherent allocation principle. As discussed in
section 5.3, additivity conflicts with the coherence of the risk measures.
The Shapley value as a coherent allocation principle
From the above, the Shapley value provides us with a coherent allocation principle
ifit maps games to elements of the core. This is the case when the conditions of
either Theorem 2 or 3 are satisfied. The case of Theorem 2 is perhaps disappoin-
ting as the strong subadditivity ofc implies an overly stringent condition on :
THEOREM 5 Let be a positively homogeneous risk measure such that(0) = 0. Let c be defined over the set of subsets of random variables in L
through
c(S) = (i SXi). Then, if c is strongly subadditive, is linear.
PROOF Consider any random variablesX, Y,ZinL
. The strong subadditivity of
c implies
(X+Z) + (Y+Z) (X+ Y+Z) + (Z)
but also
(X+Z) + (Y+Z) = (X+ (Y+Z) Y) + (Y+Z)
(X+ (Y+Z)) + ((Y+Z) Y)
= (X+ Y+Z) + (Z)
so that
(X+Z) + (Y+Z) = (X+ Y+Z) + (Z)
By takingZ= 0, we obtain the additivity of. Then, combining
(X) = (XX) (X) = (X)
with the positive homogeneity of, we obtain that is homogeneous, and thuslinear. s
Michel Denault
Journal of Risk
10
Denault pages 23/11/01 4:18 pm Page 10
-
8/3/2019 J of Risk 2001
11/34
That risks be plainly additive is difficult to accept since it eliminates all possi-
bility of diversification effects.
Unfortunately, the condition of Theorem 3 is also a strong one, at least in noway implied by the coherence of the risk measure . We thus fall short of aconvincing proof of the existence of coherent allocations. However, we consider
next another type of coalitional game where a slightly different definition of
coherence yields much stronger existence results.
5. Allocation to fractional players
In the previous section portfolios were modelled as players of a game, each of
them indivisible. This indivisibility assumption is not a natural one as we could
consider fractions of portfolios as well as coalitions involving fractions of port-
folios. The purpose of this section is to examine a variant of the allocation gamewhich allows divisible players.
This time around, we dispense with the initial separation of risk capital allo-
cation problems and games and introduce the two simultaneously. As before,
players and cost functions are used to model, respectively, portfolios and risk
measures, and values give us allocation principles.
5.1 Games with fractional players
The theory of coalitional games has been extended to continuous players who
need neither be in nor out of a coalition but who have a scalable presence.
This point of view seems much less incongruous if the players in questionrepresent portfolios: a coalition could consist of 60% of portfolio A, and 50% of
portfolio B. Of course, this meansx% ofeach instrument in the portfolio.
Aumann and Shapleys book Values of Non-Atomic Games (1974) was the
seminal work on the game concepts discussed in this section. There, the interval
[0, 1] represents the set of all players, and coalitions are measurable sub-
intervals (in fact, elements of a -algebra). Any sub-interval contains one ofsmaller measure, so there are no atoms ie, smallest entities that could be called
players; hence the name non-atomic games.
Some of the non-atomic game theory was later recast in a more intuitive set-
ting: an n-dimensional vector
n
+represents the level of presence of each
of the n players in a coalition. The original papers on the topic are Aubin (1979)
and (1981), Billera and Raanan(1981), Billera and Heath (1982) and Mirman
and Tauman (1982). Aubin called such games fuzzy; we call them (coalitional)
games with fractional players:
DEFINITION 10 A coalitional game with fractional players (N, , r) consists of
a finite set N of players, with |N| = n; a positive vector n+ , each component representing for one of the n
players his full involvement; and
a real-valued cost function r:
n
, r: a
r() such that r(0) = 0.
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
11
Denault pages 23/11/01 4:18 pm Page 11
-
8/3/2019 J of Risk 2001
12/34
Players are portfolios; the vector represents, for each portfolio, the size ofthe portfolio, in a reference unit. (The could also represent the business
volumes of the business units). The ratio ii then denotes a presence oractivity level, for player/portfolio i , so that a vector n+ can be used torepresent a coalition of parts of players. We still denote by Xi the random
variable of the net worth of portfolio i at a future time T, and Xn keeps its
riskless instrument definition of section 3, with time Tnet worth equal to B,with some constant. Then the cost function r can be identified with a riskmeasure through
so that r() = (N). By extension, we also call r() a risk measure. The expres-sionXii is theper-unitfuture net worth of portfolio i.
The definition of coherent risk measure (Definition 1) is adapted to the frac-
tional players context as:
DEFINITION 11 A risk measure, r, is coherent if it satisfies the four properties:
Subadditivity6 For all * and** in n,
r(* + **) r(*) + r(**)
Monotonicity For all * and** in n,
where the left-hand side inequality is again understood as in footnote 3.
Degree-one homogeneity For all n and for all +,
r( ) = r()
Translation invariance For all n,
One can check that ris coherent if and only if is.
r r
n
n
n
( )
=
1
2
1
0
M
i
ii
i N
i
ii
i N
X X r r
* *** **( ) ( )
r Xi
ii
i N
( )
=
Michel Denault
Journal of Risk
12
Denault pages 23/11/01 4:18 pm Page 12
-
8/3/2019 J of Risk 2001
13/34
5.2 Coherent cost allocation to fractional players
The portfolio sizes given by allow us to treat allocations on a per-unitbasis.We thus introduce a vector kn, each component of which represents theper-unit allocation of risk capital to each portfolio. The capital allocated to each
portfolio is obtained by a simple Hadamard (ie, component-wise) product
(3) .* k = K
Let us also define, in a manner equivalent to the concepts of section 4:
DEFINITION 12 A fuzzy value7 is a mapping assigning to each coalitional game
with fractional players (N, , r) a unique per-unit allocation vector
with
tk = r() (4)
Again, we use the k-notation when the arguments are clear from the context.Clearly, a fuzzy value provides us with an allocation principle if we generalize
the latter to the context of divisible portfolios.
We can now define the coherence of fuzzy values:
DEFINITION 13 Let r be a coherent risk measure. A fuzzy value
: (N, , r) a kn
is coherent if it satisfies the properties defined below and if k is an element of the
fuzzy core:
Aggregation invariance Suppose the risk measures r and r satisfy r() =r()for some m n matrix and all such that0 , then
(5)(N, , r) = t(N, , r)
Continuity The mapping is continuous over the normed vector space Mn ofcontinuously differentiable functions r:
n+ that vanish at the origin.
Non-negativity underr non-decreasing8 If r is non-decreasing, in the sense
that r() r(*) whenever0 * , then
(6)(N, , r) 0
: ( , , )
( , , )
( , , )
( , , )
N r
N r
N r
N r
k
k
kn n
aM M
1
2
1
2
=
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
13
Denault pages 23/11/01 4:18 pm Page 13
-
8/3/2019 J of Risk 2001
14/34
Dummy player allocation If i is a dummy player in the sense that
whenever0 and* = except in the ith component, then
(7)
Fuzzy core The allocation (N, , r) belongs to the fuzzy core of the game(N, , r) if for all such that0
(8)t(N, , r) r()
as well as t(N, , r) r().
The properties required of a coherent fuzzy value can be justified essentially
in the same manner as was done in Definition 3. Aggregation invariance is akin
to the symmetry property: equivalent risks should receive equivalent alloca-
tions. Continuity is desirable to ensure that similar risk measures yield similar
allocations. Non-negativity under non-decreasing risk measures is a natural
requirement to enforce that more risk implies more allocation". The dummy
player property is the equivalent of the riskless allocation of Definition 3 and is
necessary to give risk capital the sense we gave it in section 2: an amount of
riskless instrument necessary to make a portfolio acceptable, riskwise. Finally,
note that the fuzzy core is a simple extension of the concept of core: allocationsobtained from the fuzzy core through (3) allow no undercut from any player,
coalition of players or coalition with fractional players. Such allocations are
fair in the same sense that core elements were considered fair in section 4.3.
Much less is known about this allocation problem than is known about the
similar problem described in section 4. On the other hand, one solution concept
has been well investigated: the AumannShapley pricing principle.
5.3 The AumannShapley value
Aumann and Shapley extended the concept of Shapley value to non-atomic
games in their original book (1974). The result was called the AumannShapley value, and later it was recast in the context of fractional-player games,
where it is defined as:
(9)
for player i ofN. The per-unit cost kiAS is thus an average of the marginal costs of
the ith portfolio, as the level of activity or volume increases uniformly for all port-
folios from 0 to . The value has a simpler expression, given our assumed coher-ence of the risk measure r; indeed, consider the result from standard calculus:
i ii
N r k rAS AS
d( , , ) ( ) = =
0
1
kX
ii
i
= ( )
r r
X
i i
i
i( ) ( )
( )* *
= ( )
Michel Denault
Journal of Risk
14
Denault pages 23/11/01 4:18 pm Page 14
-
8/3/2019 J of Risk 2001
15/34
LEMMA 1 If f is a k-homogeneous function, ie, f(x) = kf(x), then f(x) xi is(k 1)-homogeneous.
As a result, since ris 1-homogeneous,
(10)
and the per-unit allocation vector is the gradient of the mapping revaluated at
the full-presence level :
(11)AS(N, , r) = kAS = r()
We call this gradient AumannShapley per-unit allocation, or simply Aumann
Shapley prices. The amount of risk capital allocated to each portfolio is thengiven by the components of the vector
(12)KAS = kAS .*
Note that the gradient (11) is called marginal VAR in the value-at-risk literature.
However, marginal VAR is based on VAR, a not-coherent risk measure, and the
allocation principle derived from it (called component VAR) is not a coherent
one.
Axiomatic characterizations of the AumannShapley value
As in the Shapley value case, a characterization consists of a set of properties
that uniquely define the AumannShapley value. Many characterizations exist
(see Tauman, 1988); we concentrate here on those of Aubin (1979) and (1981)
and Billera and Heath (1982). Both characterizations are for values of games
with fractional players as defined above, only their assumptions on rdiffer from
ours: their cost functions are taken to vanish at zero and to be continuously
differentiable but are not assumed coherent. Aubin also implicitly assumes rto
be homogeneous of degree one. Let us define:
A fuzzy value is linear if for any two games (N, , r1) and (N, , r2) andscalars 1 and 2 it is additive and 1-homogeneous in the risk measure:
(N, , 1r1 + 2 r2) = 1(N, , r1) + 2 (N, , r2)
Then, the following properties of a fuzzy value are sufficient to uniquely define
the AumannShapley value (9):
Aubins Billera and Heaths
linearity linearity
aggregation invariance aggregation invariance
continuity non-negativity under rnon-decreasing
i i i
N r k rAS AS
( , , )( )
= =
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
15
Denault pages 23/11/01 4:18 pm Page 15
-
8/3/2019 J of Risk 2001
16/34
In fact, both Aubin, and Billera and Heath prove that the AumannShapley value
satisfies all four properties in the table above.
So, is the AumannShapley value a coherent fuzzy value when ris a coherentrisk measure? Note first that the coherence ofrimplies its homogeneity, as well
as r(0) = 0. Being continuously differentiable is not automatic however; let us
assume for now that r does have continuous derivatives. The eventual non-
differentiability will be discussed later.
Clearly, two properties are missing from the set above for to qualify ascoherent: the dummy player property and the fuzzy core property. The former
causes no problem: given (10), the very meaning of a dummy player in Defini-
tion 13 implies:
LEMMA 2 When the allocation process is based on a coherent risk measure r, the
AumannShapley prices (10) satisfy the dummy player property.
Concerning the fuzzy core property, one very interesting result of Aubin is the
following:
THEOREM 6 (Aubin, 1979) The fuzzy core (8) of a fuzzy game (N, r, ) with posi-tively homogeneous r is equal to the sub-differential r() of r at .
As Aubin noted, the theorem has two very important consequences:
THEOREM
7 (Aubin, 1979) If the cost function r is convex (as well as positivelyhomogeneous), the fuzzy core is non-empty, convex and compact.
If furthermore r is differentiable at, the core consists of a single vector, thegradientr().
The direct consequence of this is that the AumannShapley value is indeed a
coherent fuzzy value, given that it exists:
COROLLARY 1 If(N, r, ) is a game with fractional players, with r a coherent cost function that is differentiable at , then the AumannShapley value (11) is acoherent fuzzy value.
PROOF The corollary follows directly from (10), Theorem 7, and the fact that
under positive homogeneity ris subadditive if and only if it is convex.
This corollary is our most useful result from a practical point of view. It says that
if we use a coherent but also differentiable risk measure, and if we deem impor-
tant the properties of allocation given in Definition 13, the allocation r() .* is a right way to go.
As a final note, let us mention that the condition of non-increasing marginal
costs, given in Billera and Heath (1982) for the the membership ofAS(N, , r) inthe fuzzy core, in fact implies that ris linear whenever it is homogeneous.
Michel Denault
Journal of Risk
16
Denault pages 23/11/01 4:18 pm Page 16
-
8/3/2019 J of Risk 2001
17/34
On linear values and uniqueness
From the results given above, the AumannShapley value is the only linear
coherent allocation principle when the cost function is adequately differen-tiable. However, linearity over risk measures is not required of a coherent
allocation principle: while 1-homogeneity is quite acceptable, the additivity
part (N, , r1 + r2) = (N, , r1) + (N, , r2) causes the following problem.Because of the riskless condition, a coherent risk measure cannot be the sum of
two other coherent risk measures as it leads to the contradiction (written in the
less cluttered but equivalent notation)
(X) = (X+ B) = 1(X+ B) + 2(X+ B)
= 1(X) +
2(X) 2
= (X) 2
Therefore, the very definition of additivity would imply that we consider non-
coherent risk measures. On the other hand, convex combinations of coherent risk
measures are coherent (see Delbaen, 2000), so we could make the following
condition part of the definition of coherent allocations.
DEFINITION 14 A fuzzy value satisfies the convex combination property if forany two games (N, , r1) and(N, , r2) and any scalar [0,1]
(N, , r1 + (1 )r2) = (N, , r1) + (1 ) (N, , r2)
That condition implies linearity when combined with the 1-homogeneity with
respect to r of AS(N, , r) (which holds given the aggregation invarianceproperty). This would make the AumannShapley value the unique coherent
allocation principle. However, we see no compelling, intuitive reason to include
linearity (under one form or another) in the definition of coherent fuzzy
allocation, allowing for the existence of non-linear coherent fuzzy allocation
principles, a topic that is left for further investigation.
The same remarks on uniqueness and linearity apply to the Shapley value andallocation in the non-divisible players context. Note that the debate on the perti-
nence of linearity is far from new: as early as 1957 Luce and Raiffa observed
that [additivity] strikes us as a flaw in the concept of [Shapley] value.
On the differentiability requirement
Concerning the differentiability of the risk measures/cost functions, recent
results are encouraging. Tasche (1999) and Scaillet (2000) give conditions under
which a coherent risk measure, the expected shortfall, is differentiable. The
conditions are relatively mild, especially in comparison with the temerarious
assumptions common in the area of risk management. Explicit first derivatives
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
17
Denault pages 23/11/01 4:18 pm Page 17
-
8/3/2019 J of Risk 2001
18/34
are provided, which have the following interpretation: they are expectations of
the risk factors, conditioned on the portfolio value being below a certain quantile
of its distribution. This is very interesting: it shows that when the AumannShapley value is used with a shortfall risk measure, the resulting (coherent)
allocation is again of a shortfall type:
where q is a quantile of the distribution ofiXi.Even when r is not differentiable something can often be saved. Indeed,
suppose that r is not differentiable at but is the supremum of a set of para-meterized functions that are themselves convex, positively homogeneous and
differentiable at :
(13)r() = supp P
w(,p)
where P is a compact set of parameters of the functions w, and w(,p) is uppersemi-continuous inp. Then Aubin (1979) proved:
The fuzzy core is the closed convex hull of all the values AS(N, , w(,p)) of thefunctions w that are active at, ie, that are equal to r().
Thus, should (13) arise (which is not unlikely think of Lagrangian relaxation
when ris defined by an optimization problem), the above result provides a setofcoherent values to choose from.
Alternative paths to the AumannShapley value
It is very interesting that the recent report by Tasche (1999) comes, fundamen-
tally, to the same result that is obtained in this section namely, that given some
differentiability conditions on the risk measure , the correct way of allocatingrisk capital is through the AumannShapley prices (10). Tasches justification of
this contention is, however, completely different; he defines as suitable capital
allocations such that if the risk-adjusted return of a portfolio is above average,
then, at least locally, increasing the share of this portfolio improves the overallreturn of the firm. Note that the work of Schmock and Straumann (1999) points,
again, to the same conclusion. In the approach of Tasche (1999) and Schmock
and Straumann (1999) the AumannShapley prices are in fact the unique satis-
factory allocation principle.
Other important results on the topic include those of Artzner and Ostroy (1983),
who, working in a non-atomic measure setting, provide alternative characteriza-
tions of differentiability and sub-differentiability, with the goal of establishing
the existence of allocations through, basically, Eulers theorem. See also the
forthcoming Delbaen (2000).
On Eulers theorem,9 note also that condition (4) of the allocation vector
K E X X qi i ii
=
Michel Denault
Journal of Risk
18
Denault pages 23/11/01 4:18 pm Page 18
-
8/3/2019 J of Risk 2001
19/34
follows directly from it, and that out of consideration for this some authors have
called the allocation principle (10) the Euler principle. See, for example, the
attachment to the report of Patrik, Bernegger and Regg (1999), which providessome properties of this principle.
We shall end this section by drawing the attention of the reader to the
importance of the coherence of the risk measure (and the rderived from it) forthe allocation.
The subadditivity of the risk measure is a necessary condition for the exis-
tence of an allocation with no undercut in both the atomic and the fractional
players contexts.
The homogeneity of the risk measure ensures the simple form (10) of the
AumannShapley prices.
Both subadditivity and homogeneity are used to prove that the core is non-empty (Theorem 4) in the atomic game setting. In the fuzzy game setting, the
two properties are used to show that the AumannShapley value is in the
fuzzy core (under differentiability). They are also used in the non-negativity
proof in Appendix A.
The riskless property is central to the definition of the riskless allocation
(dummy player) property.
6. The non-negativity of the allocation
Given our definition of risk measure, a portfolio may well have a negative riskmeasure, with the interpretation that the portfolio is then safer than deemed
necessary.
Similarly, the allocation of a negative amount to a portfolio does not pose a
conceptual problem; the interpretation is that this portfolio is a hedge within the
firms global portfolio, ie, a constituent that reduces the level of risk.
Unfortunately, in some applications we may consider, negativity poses a
problem. If the amount is to be used in a RAPM-type quotientReturnAllocated
capital, negativity has a rather nasty drawback in that a portfolio with an allo-
cated capital slightly below zero ends up with a negative risk-adjusted measure
of large magnitude, whose interpretation is less than obvious. A negative alloca-
tion is therefore not so much a concern with the allocation itself as with the use
we would like to make of it.
A crossed-fingers, and perhaps most pragmatic approach, is to assume that
the coherent allocation is inherently non-negative. In fact, one could reasonably
expect non-negative allocations to be the norm in real-life situations. For
example, provided that no portfolio of the firm ever reduces the risk measure
when added to any subset of portfolios of the firm:
c (S {i}) c (S) S N, i N\ S
then the Shapley value is necessarily non-negative. The equivalent condition for
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
19
Denault pages 23/11/01 4:18 pm Page 19
-
8/3/2019 J of Risk 2001
20/34
the AumannShapley prices is the property of non-negativity under non-decreas-
ing r(Equation (6)): if the antecedent always holds, the per-unit allocations are
non-negative.Another approach would be to enforce non-negativity by requiring more of
the risk measure. For example, the core and the non-negativity of the Ki s form
a set of linear inequalities (and one linear equality), so that the existence of a
non-negative core solution is equivalent to the existence of a solution to a linear
system. Specifically, a hyperplane separation argument proves that such a
solution will exist if the following condition on holds:
(14)
The proof is given in Appendix A. The condition could be interpreted as follows.
First, assume that (iNXi) > 0, which is reasonable if we are indeed to allo-cate some risk capital. Then (14) says that there is no positive linear combination
of (each and every) portfolios that runs no risk. In other words, a perfectly
hedged portfolio cannot be attained by simply reweighting the portfolios if all
portfolios are to have a positive weight. However, unless one is willing to
impose such a condition on the risk measure, the fact is that the issue of the non-
negativity remains unsatisfactorily resolved for the moment.
7. Allocation with a SEC-like risk measure
In this section we provide some examples of applications of the Shapley and
AumannShapley concepts to a problem of margin (ie, risk capital) allocation.
The risk measure we use is derived from the Securities and Exchange
Commission (SEC) rules for margin requirements (Regulation T), as described
in the National Association of Securities Dealers (NASD) document (1996).
These rules are used by stock exchanges to establish the margins required of
their members as a guarantee against the risk that the members portfolios
involve (the Chicago Board of Options Exchange is one such exchange). The
rules themselves are not constructive in that they do not specify how the margin
should be computed; this computation is left to each member of the exchange,
who must find the smallest margin complying with the rules. Rudd andSchroeder (1982) proved that a linear optimization problem (LP) modelled the
rules adequately and was sufficient to establish the minimum margin of a port-
folio that is, to evaluate its risk measure. It is worth mentioning that given this
LP-based risk measure, the corresponding coalitional game has been called the
linear production game by Owen (1975); see also Billera and Raanan (1981).
For the purpose of the article we restrict the risk measure to simplistic
portfolios of calls on the same underlying stock and with the same expiration
date. This restriction of the SEC rules is taken from Artzner et al. (1999), who
use it as an example of a non-coherent risk measure. In the case of a portfolio of
calls the margin is calculated through a representation of the calls by a set of
{ }
+
n ii N
i Ni i i
i N
X X, min
Michel Denault
Journal of Risk
20
Denault pages 23/11/01 4:18 pm Page 20
-
8/3/2019 J of Risk 2001
21/34
spread options, each of which carries a fixed margin. To obtain a coherent
measure of risk, we prove later that it is sufficient to represent the calls by a set
of spreads andbutterfly options. Note that such a change to the margin rules wasproposed by the NASD and recently accepted by the SEC (NASD, 2000).
7.1 Coherent, SEC-like margin calculation
We consider a portfolio consisting ofCP calls at strike price P, where P belongs
to a set of strike prices P= {Pmin, Pmin + 10,, Pmax 10, Pmax}. This assump-
tion about the format of the strike prices set P, including the intervals of 10,
makes the notation more palatable without loss of generality. We also make the
simplifying assumption that there are as many long calls as short calls in the
portfolio, ie, PP CP = 0. Both assumptions remain valid throughout section 7.
We will denote by CP the vector of the CP parameters, P P. While CP fullydescribes the portfolio, it certainly does not describe the future value of the port-
folio, which depends on the price of the underlying stock at a future date.
Although risk measures were defined as a mapping on random variables, we
nevertheless write (CP) since the considered here can be defined by usingonly CP. On the other hand, there is a simple linear relationship between CP and
the future worths (under an appropriate discretization of the stock price space),
so that an expression such as (C*P + CP**) is also justified. Only later, in the caseof the proof of monotonicity of, need we treat with more care the distinctionbetween number of calls and future worth.
We can now define our SEC-like margin requirement. To evaluate the margin(or risk measure) of the portfolio CP, we first replicate its calls with spreadsand butterflies, defined as in Table 1. The variables shall represent the number of
each specific instrument. AllHand Kare understood to be in P for the spreads,
withH K; allHare in P \{Pmin, Pmax} for the butterflies.As in the SEC rules, fixed margins are attributed to the instruments used for the
replicating portfolio, ie, spreads and butterflies in our case. Spreads carry a mar-
gin of (H K)+ = max(0,H K), which effectively means that bull spreads carry
no margin, while bear spreads do. Short butterflies are given a margin of 10, while
long butterflies require no margin. In simple language, each instrument requires a
margin equal to the worst potential loss, or negative payoff, it could yield.
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
21
TABLE 1 Call equivalents of spreads and butterflies
Variable Instrument Calls equivalent
SH,K Spread, long in H, short in K
BH
long Long butterfly, centred at H
BH
short Short butterfly, centred atH
One long call at price H, one short call at
strike K
One long call at H 10, two short
calls at H, one long call at H + 10
One short call at H 10, two long calls at
H, one short call at H + 10
Denault pages 23/11/01 4:18 pm Page 21
-
8/3/2019 J of Risk 2001
22/34
By definition, the margin of a portfolio of spreads and butterflies is the sum of
the margins of the instruments taken separately.
It is easy to extend the results of Rudd and Schroeder (1982) to take intoaccount our introduction of fixed margins on butterflies; the margin (CP) of theportfolio can be evaluated with the linear optimization problem (SECLP):
minimize ftY
(SECLP)
subject to AY= CPY 0
The column of variables Ystands for
where S is a column vector of all spread variables considered (appropriately
ordered: bull spreads, then bear spreads) and Blong and Bshort are appropriately
ordered column vectors of butterfly variables. The objective function ftY is
shorthand notation for
andA is
The matrix is read as follows. Its first column indicates that the first component
of Y, the variable associated with the bull spread SPmin, Pmin+10, contributes, per
spread, one long call with strike Pmin and one short call with strike Pmin + 10 to
the portfolio of calls CP. Similarly, from the first column of the third part ofA,
the component B longPmin+10 of Ycontributes, per butterfly, one long call with each
strike Pmin
and Pmin
+ 20 and two short calls with strike Pmin
+ 10.
A =
1 1 0 1 1 0 1 0 0 1 0 0
1 0 0 1 0 0 2 1 0 2 1 0
0 1 0 0 1 0 1 2 0 1 2 0
0 0 0 0 0 0 0 1 0 0 1 0
0 0 0 0 0 0 0 0 1 0 0 1
0 0 1 0 0 1 0 0
L L L L
L L L L
L L L L
L L L L
M M M M M M M M M M M M
L L L L
L L L
2 0 0 2
0 0 1 0 0 1 0 0 1 0 0 1
L
L L L L
f H K S Bt
H K
H K
H
H P P
Y= ++
( ) ,
, \{ , }min maxP P
10short
Y
S
B
B
=
long
short
Michel Denault
Journal of Risk
22
Denault pages 23/11/01 4:18 pm Page 22
-
8/3/2019 J of Risk 2001
23/34
The objective function represents the margin; the equality constraints ensure
that the portfolio is exactly replicated. The risk measure thus defined is coherent;
the proof is given in Appendix B.
7.2 Computation of the allocations
Given this risk measure as a linear optimization problem, the Shapley value is
easy to compute when the firm is divided into a small number of portfolios (the
firm is defined as the sum of the portfolios). First, the margin of every possible
coalition of portfolios is calculated. Then the margin allocated to each portfolio
is computed using the formula of the Shapley value given in Definition 9.
The computation of the AumannShapley value is even simpler. Note that by
working in the fractional players framework we implicitly assume that fractions
of portfolios are sensible instruments. We choose the vector of full presence ofall players, , to be the vector of ones denoted by e. Recall that the AumannShapleyper-unitmargin allocated to the ith portfolio is
(15)
where r() is the margin required of the sum of all the portfolios i, each scaledby a scalar i, so that r(e) =(CP). In vector notation, k
AS = r(e).Let us first define the linear operator L which maps the level of presence of
the portfolios to numbers of calls in the firm. If there are |P | different calls(equivalently here, |P | strike prices) and n portfolios, thenL is a |P | n matrix,such thatLe = CP. Examples ofL
tare the three-by-five matrices at the top of the
tables given in section 7.3.
Now, the optimal dual solution, *, of the linear program (SECLP), obtainedautomatically when computing the margin of the firm, provides the rates of change
of the margin when the presence ofeach specific call varies. However, using the
complementarity condition satisfied at the optimal solution pair (Y*, *), we canwrite
(16)
(17)
so that the components ofLt* give the marginal rates of change of the objective
value of (SECLP), as a function of portfolio presence, evaluated at the point of
full presence of all portfolios.
Put in one sentence, the most interesting result of this section is that the
AumannShapley allocation is only a matrix product away from the lone evalu-
ation of the margin for the firm.
Finally, concerning the uniqueness of the allocation and the differentiability
of the risk measure, we can only say that they depend directly on the uniqueness
f
L e
t t
t
Y CP* =
= ( )
( )
( )
*
*
kr e
ii
AS =
( )
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
23
Denault pages 23/11/01 4:18 pm Page 23
-
8/3/2019 J of Risk 2001
24/34
of the optimal solution of the dual problem of (SECLP). Although there is no
special reason for multiple optimal dual solutions to occur here, it can well
happen in which case we have obtained one of many acceptable allocations, asnoted in the differentiality discussion of section 5.3.
7.3 Numerical examples of coherent allocation
We can obtain a somewhat more practical feeling of Shapley and Aumann
Shapley allocations by looking at examples.
For all allocation examples below, the reference firm is the same; its values
ofCP, P {10, 20, 30, 40, 50}, are
meaning one short call at strike 10, two short calls at strike 20, eight long calls
at strike 30, etc. The linear programming problem (SECLP) is solved with any
optimization software (we used MATLAB), and yields a margin (CP) = 40,obtained with the solution
and all other variables equal to 0.10 The vector of dual variables, also obtained
directly from the software, is (*)t= [ 20 10 0 0 0 ].In each of the tables below, we give the division of the firm in three portfolios
such that CP1+ CP2
+ CP3= CP, followed by the Shapley allocation and the
AumannShapley allocation whose computations were explained in the previous
section. The tables are followed by some observations. Consider first the division:
In this example, coalitions of portfolios incur margins as follows:
With these margins, the Shapley allocation can be computed using the formula (1)
C C C C C C
C C C
P P P P P P
P P P
1 2 1 3 2 3
1 2 3
+( ) = +( ) = +( ) =
( ) = ( ) = ( ) =40 20 30
20 20 10
C C C C C 10 20 30 40 50
1 0 6 6 1
0 2 2 0 0
0 0 0 1 1
1 2 8 7 2
Shapley AumannShapley
15 20
20 20
5 0
Total 40 40
C
C
C
P
P
P
1
2
3
S S B B
S S
S
20 10 30 40 30 40 50 20 30 40
30 10 30 50
30 20
0 125 0 625 2 5 1 375
0 875 0 500
0 750
, , , , , ,
, ,
,
. . . .
. .
.
= = = == ==
short long
C C C C C 10 20 30 40 50
1 2 8 7 2Total
Michel Denault
Journal of Risk
24
Denault pages 23/11/01 4:18 pm Page 24
-
8/3/2019 J of Risk 2001
25/34
of section 4.2. The AumannShapley allocation is given by Lt*, as explained
in the previous section:
Consider a second example:
Here, coalitions of portfolios incur the margins
Finally, the third example is
where the coalitions of portfolios incur
On these examples, we note that:
The Shapley and AumannShapley allocations do not agree. An equal alloca-
tion in this setting would have been fortuitous.
Null allocations do occur. Null (or negative) allocations cannot be ruled out
and do happen here, both for the Shapley and AumannShapley principles.
For use in a risk-adjusted return calculation (return divided by allocated
capital), these would indeed be problematic.
C C C C C C
C C C
P P P P P P
P P P
1 2 1 3 2 3
1 2 3
+( ) = +( ) = +( ) =( ) = ( ) = ( ) =
40 30 10
30 10 20
C C C C C 10 20 30 40 50
1 1 4 2 0
0 1 4 3 0
0 0 0 2 2
1 2 8 7 2
Shapley AumannShapley
26.66 30
6.66 10
6.66 0
Total 40 40
C
C
C
P
P
P
1
2
3
C C C C C C
C C C
P P P P P P
P P P
1 2 1 3 2 3
1 2 3
+( ) = +( ) = +( ) =
( ) = ( ) = ( ) =30 50 20
20 10 30
C C C C C 10 20 30 40 50
1 0 2 2 1
0 1 6 5 0
0 1 0 0 1
1 2 8 7 2
Shapley AumannShapley
20 20
0 10
20 10
Total 40 40
CC
C
P
P
P
1
2
3
=
=
Lt*1 0 6 6 1
0 2 2 0 0
0 0 0 1 1
20
10
0
0
0
20
20
0
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
25
Denault pages 23/11/01 4:18 pm Page 25
-
8/3/2019 J of Risk 2001
26/34
The AumannShapley allocation is a marginal rate of risk capital. It can be
checked that, for example in the first case, the risk capital of the whole firm
does not change when the presence of the third portfolio varies slightlyaround unity.
The Shapley allocation may not be in the core. Again, this was to be expected.
The Shapley allocation of the third example is not part of the core, as can be
checked manually.
As a final note, let us mention that we could have verified whether the Shapley
allocation is or is not part of the fuzzy core (whenever it is part of the core). One
way of doing this is through a bi-level linear program, which, unfortunately,
is known to be computationally demanding for large instances; see Marcotte
(1997) or Marcotte and Savard (2000). Further details on this issue are reserved
for later.
8. Conclusion
In this article we have discussed the allocation of risk capital from an axiomatic
perspective, defining in the process what we call coherent allocation principles.
Our original goal was to establish a framework within which financial risk
allocation principles could be compared as meaningfully as possible. Our stand
is that this can be achieved by binding the concept of coherent risk measures to
the existing game theory results on allocation.
We suggest two sets of desirable properties of allocation principles, each set
defining the coherence of risk capital allocation in a specific setting: either theconstituents of the firm are considered indivisible entities (in the coalitional
game setting) or, to the contrary, they are considered to be divisible (in the context
of games with fractional players). In the former case, we find that the Shapley
value is a coherent allocation principle, though only under rather restrictive con-
ditions on the risk measure used.
In the latter, fractional players setting, the AumannShapley value is a coher-
ent allocation principle under a much laxer differentiability condition on the risk
measure; under linearity it is also the unique coherent principle. In fact, given
that the allocation process starts with a coherent risk measure, this coherent allo-
cation simply corresponds to the gradient of the risk measure with respect to thepresence level of the constituents of the firm.
Our main message is that the gradient-of-the-risk-measure, AumannShapley
principle has beyond the computational advantage of being as easy to evaluate
as the risk measure itself a strong conceptual justification.
Michel Denault
Journal of Risk
26
Denault pages 23/11/01 4:18 pm Page 26
-
8/3/2019 J of Risk 2001
27/34
Appendix A: Proof of the non-negativity condition
The following result from section 6 is proved here.
THEOREM 8 A sufficient condition for a non-negative, no undercut allocation
to exist is
PROOF Let us recall that we denoted by 1S n
the characteristic vector of the
coalition S:
A non-negative, no undercut allocation Kexists when
(18)
Using Farkas lemma, this is equivalent to
(19)
which in turn is equivalent to
(20)
y y y S N i N
X y X y
N S
S i
S
i
i SS N
S i
i N
N
, , ,0 and
1 1 0
0
N N S S
S N
N S
i
i N
N i
i SS N
S
y y y y S N
X y X y
+
+
+
, , ,
=
K
K X S N
K X
K
n
tS i
i S
tN i
i N
such that
1
1
0
11
0S i
i S( ) =
if
otherwise
{ }
+
n ii N
i Ni i i
i N
X X, min
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
27
Denault pages 23/11/01 4:18 pm Page 27
-
8/3/2019 J of Risk 2001
28/34
Now, using the homogeneity and the subadditivity of,
Therefore, a sufficient condition for (18) (or (19) or (20)) to hold is
Finally, using the definition i = SiyS, we can write the sufficient condition
for (18)
Note that in the last step, we also used (i NXi) 0, a necessary conditionfor the existence of a non-negative, no undercut allocation; checking yN= 1,
yS = 0 in (19) shows this point. s
Appendix B: Proof of the coherence of the measure
We prove here that the risk measure obtained through (SECLP) is coherentinthe sense of Definition 1. We prove each of the four properties in turn below.
Subadditivity
For any two portfolios C*P and CP**
(C*P + CP**) (C*P) + (CP**)
PROOF If solving (SECLP) with C*P as the right-hand side of the equality
constraints yields a solution Y* and solving with CP** yields a solution Y**, then
Y* + Y** is a feasible solution for the (SECLP) with C*P + CP** as the right-hand
side. Subadditivity follows directly, given the linearity of the objective function.
S N
( ) i i i
i N
i
i N i Nii N X X 0, , min
y y y S N i N
X y X y
N S
S i
S
i S
S ii N
i
i N
N
, , , ,
( )
0
X y y X
y X
X y
ii SS N
S S ii SS N
S i
i SS N
i S
S ii N
=
=
Michel Denault
Journal of Risk
28
Denault pages 23/11/01 4:18 pm Page 28
-
8/3/2019 J of Risk 2001
29/34
Degree-one homogeneity
For any 0 and any portfolio CP,
(CP) = (CP)
PROOF This is again a direct consequence of the linear optimization nature ofas S is a solution of the (SECLP) with CP as the right-hand side of theconstraints when S is a solution of the (SECLP) with CP as the right-hand side.
Of course, the very definition of homogeneity implies that we allow fractions of
calls to be sold and bought.
Translation invariance11
Adding to any portfolio of calls CP an amount of riskless instrument worth today reduces the margin ofCP by .
PROOF There is little to prove here; rather we need to define the behaviour of
in the presence of a riskless instrument, and naturally we choose the transla-tion invariance property to do so. This property simply anchors the meaning of
margin.
Monotonicity
For any two portfolios C*
P and CP**
such that the future worth ofC*
P is always lessthan or equal to that ofCP**,
(C*P) (CP**)
Before proving monotonicity, let us first introduce the values VP, for
P {Pmin + 10, , Pmax, Pmax + 10}, which represent the future payoffs, or worths,of the portfolio for the future prices, P, of the underlying.12 Again, we write VPto denote the vector of all VPs. The components ofVP are completely determined
by the number of calls in the portfolio:
which is alternatively written VP =MCP, with the square, invertible matrixM:
M=
10 0 0
20 10 0
30 20 10
L
L
L
M M M O
V C P p P P P P PP pp P
P= + + +{ }
=
( ) , , , ,min
min min max max
1010 20 10
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
29
Denault pages 23/11/01 4:18 pm Page 29
-
8/3/2019 J of Risk 2001
30/34
The antecedent of the monotonicity property is, of course, the componentwise
V*P VP**.
We will also use the following lemma.
LEMMA 3 Under subadditivity, two equivalent formulations of monotonicity are,
for any three portfolios CP, C*
P and CP**,
and
PROOF The upper condition is sufficient, as it implies
and (0) = 0 from the very structure of (SECLP). The upper condition isnecessary as
s
PROOF OF MONOTONICITY As a consequence of the above lemma, it is sufficient
to prove that if a portfolio of calls always has non-negative future payoff, its
associated margin is non-positive.
A look at (SECLP) shows that the margin assigned to the portfolio will be
non-positive (in fact, zero) if and only if a non-negative, feasible solution of(SECLP) exists in which all spread variables SH,K with H> K and all short
butterfly variablesBshort have value zero. This means that there exists a solution
to the linear system
(21)A~Y = CP
(22)Y 0
where A~
is made of the first and third parts of the A that was defined for
(SECLP):
M M
M M
M
( ) = + ( )( )( )
( ) + ( )( )
( )
1 1
1 1
1
V V V V
V V V
V
P**
P*
P**
P*
P*
P**
P*
P*
( )0 ( )M 1 VP
0 01 ( ) V VP P M
V V V V P*
P**
P*
P** ( ) ( ) M M1 1
Michel Denault
Journal of Risk
30
Denault pages 23/11/01 4:18 pm Page 30
-
8/3/2019 J of Risk 2001
31/34
and Y is the appropriate vector of spread and butterfly variables. We obtain a
new, equivalent system of equations MA~
Y=MCP
= VP
by pre-multiplying by
the invertible matrix M introduced above. Recall now that we have made the
assumption that the portfolio contains as many short calls as long calls, ie,
PP CP = 0. Thus, we need only prove that there exists a non-negative solution
to the system
MA~
Y= VP whenever VP 0 and e tM1VP = 0
(e t is a row vector of 1s). A simple observation ofMA~
shows that its columns
span the same subspace as the set of columns
Observing furthermore that
so that any VP satisfying etM1VP = 0 has identical last two components, the
right-hand side ofMA~
Y= VP can always be expressed as a non-negative linear
combination of the columns ofMA~. s
e Mt
t
=
1
00
0
1
1
M
1
0
0
0
0
0
0
1
0
0
0
0
0
0
0
1
0
0
0
0
0
0
1
1
M M M M
, , , ,
A =
1 1 0 1 0 0
1 0 0 2 1 0
0 1 0 1 2 0
0 0 0 0 1 0
0 0 0 0 0 1
0 0 1 0 0 2
0 0 1 0 0 1
L L
L L
L L
L L
M M M M M M
L L
L L
L L
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
31
Denault pages 23/11/01 4:18 pm Page 31
-
8/3/2019 J of Risk 2001
32/34
1 At a more general level the interested reader may consult a game theory reference such as
the book by Osborne and Rubinstein (1994), the book edited by Roth (1988) (which
includes the survey by Tauman, 1988), or the survey article by Young (1994), which con-
tain legions of references on the subject.
2 On this topic see also Artzner (1999) and Delbaen (1998) and (2000).
3 The relationX Ybetween two random variables is taken to mean X() Y() for almostall in the probability space (, A, ).
4 Here \ denotes the set relative complement:A\B contains all the elements ofA less those of
the intersection ofA andB.
5 By definition, a strongly subadditive set function is subadditive. We follow Shapley (1971)
in our terminology; note that he calls convex a function satisfying the reverse relation of
strong subadditivity.
6 Note that under degree-one homogeneity, subadditivity is equivalent to convexity:r( * + (1 ) **) r(*) + (1 ) r(**)
7 Like fuzzy game, the term seems to have been coined by Aubin (1979).
8 Called monotonicity by some authors.
9 Which states that ifFis a real, n-variable, homogeneous function of degree k,
10 Note that our extension of Rudd and Schroeders result given in section 7.1 to account
for the butterflies destroys the transportation problem structure that ensured integer
solutions. Consequently, non-integer solutions (eg, 1.125 contracts of a certain spread)
are to be expected, as is the case in our example. For portfolios with a large number of
instruments, rounding the solution would be an adequate approach. Alternatively, integ-
rality constraints could be imposed in the optimization.
11 Note that, in the interest of a tidier notation, we departed from our previous usage and did
not use the last component ofCP to denote the riskless instrument.
12 Obviously, the latter set of prices may be too coarse a representation of possible future
prices and is used to keep the notation compact; starting with a finer P relieves this
problem; similarly, Pmin and Pmax can be set as low and high as desired.
REFERENCES
Artzner, P. (1999). Application of coherent risk measures to capital requirements in insurance.
North American Actuarial Journal 3(2), April, pp. 1125.
Artzner, P., Delbaen, F., Eber, J.-M., and Heath, D. (1997). Thinking coherently.Risk10(11),
pp. 6871.
Artzner, P., Delbaen, F., Eber, J.-M., and Heath, D. (1999). Coherent measures of risk. Mathe-
matical Finance 9(3), pp. 20328.
Artzner, P., and Ostroy, J. M. (1983). Gradients, subgradients and economic equilibria.
Advances in Applied Mathematics 4, pp. 24559.
xF x
xx
F x
xx
F x
xk F xn
n1
12
2
+
+ +
=
( ) ( ) ( )( )L
Michel Denault
Journal of Risk
32
Denault pages 23/11/01 4:18 pm Page 32
-
8/3/2019 J of Risk 2001
33/34
Aubin, J.-P. (1979). Mathematical Methods of Game and Economic Theory (Amsterdam:
North-Holland).
Aubin, J.-P. (1981). Cooperative fuzzy games. Mathematics of Operations Research 6(1),pp. 113.
Aumann, R., and Shapley, L. (1974). Values of Non-Atomic Games (Princeton, NJ: Princeton
University Press).
Billera, L. J., Heath, D. C., and Raanan, J. (1978). Internal telephone billing rates a novel
application of non-atomic game theory. Operations Research 26(6), pp. 95665.
Billera, L. J., and Heath, D. C. (1982). Allocation of shared costs: a set of axioms yielding a
unique procedure.Mathematics of Operations Research 7(1), pp. 329.
Billera, L. J., and Raanan, J. (1981). Cores of nonatomic linear production games. Mathe-
matics of Operations Research 6(3), pp. 4203.
Bondareva, O. N. (1963). Some applications of the methods of linear programming to the
theory of cooperative games. Problemy Kibernetiki 10, pp. 11939. (In Russian)
Delbaen, F. (1998). Coherent risk measures on general probability spaces. Research paper,
Department of Mathematics, ETH-Zrich, 1998.
Delbaen, F. (2000). Notes for the tutorial at the Scuola Normale Superiore, Pisa. (Forthcoming).
Gouriroux, C., Laurent, J.-P., and Scaillet, O. (1999). Sensitivity analysis of values at risk.
Forthcoming inJournal of Empirical Finance, Special Issue on Risk Management.
Jorion, P. (2000). Value at Risk: The New Benchmark for Managing Financial Risk. Second
Edition (Blacklick, OH: McGraw-Hill).
Luce, R. D., and Raiffa, H. (1957). Games and Decisions. (New York: John Wiley and Sons).
Marcotte, P. (1997). Inquations variationnelles: motivation, algorithmes de rsolution et
quelques applications. Publication CRT-97-02, Centre de Recherche sur les Transports.
Marcotte, P., and Savard, G. (2000). Bilevel programming. In The Encyclopedia of Optimization.
C. A. Floudas and P. M. Pardalos (eds). (Dordrecht: Kluwer Academic). (Forthcoming)
Mirman, L. J., and Tauman, Y. (1982). Demand compatible equitable cost sharing prices.
Mathematics of Operations Research 7(1), pp. 4056.
National Association of Securities Dealers (NASD) (1996). NASD Manual. (Chicago: C.C.R.,
July).
National Association of Securities Dealers (NASD) (2000). Notice to members No. 01-11,
November. Available at www.nasdr.com.
Osborne, M., and Rubinstein, A. (1994). A Course in Game Theory. (Cambridge, MA: MIT
Press).
Owen, G. (1975). On the core of linear production games. Mathematics of Operations
Research 9, pp. 35870.
Patrik, G. S., Bernegger, S., and Regg, M. B. (1999). The use of risk adjusted capital to
support business decision-making. In the Casualty Actuarial Society Forum, Spring 1999
Edition, pp. 243334.
Roth, A. (ed.) (1988). The Shapley Value. Essays in Honor of Lloyd S. Shapley. (Cambridge:
Cambridge University Press).
Coherent allocation of risk capital
Volume 4/Number 1, Fall 2001
33
Denault pages 23/11/01 4:18 pm Page 33
-
8/3/2019 J of Risk 2001
34/34
Rudd, A., and Schroeder, M. (1982). The calculation of minimum margin. Management
Science, 28(12), pp. 136879.
Scaillet, O. (2000). Nonparametric estimation and sensitivity analysis of expected shortfall.Available at: www.econ.ucl.ac.be/IRES/CSSSP/home_pa_pers/scaillet/scaill.htm.
Schmock, U., and Straumann, D. (1999). Private communication.
Shapley, L. (1953). A value for n-person games. In Contributions to the Theory of Games, Vol-
ume II. Annals of Mathematics Studies, 28, A. W. Tucker and R. D. Luce (eds). (Princeton,
NJ: Princeton University Press).
Shapley, L. (1967). On balanced sets and cores. Naval Research Logistics Quarterly 14,
pp. 45360.
Shapley, L. (1971). Cores of convex games.International Journal of Game Theory 1, Issue 1
(197172), pp. 1126.
Tasche, D. (1999). Risk contributions and performance measurement. Report of the Lehrstuhl
fr mathematische Statistik, T.U. Mnchen. Available at: www-m4.mathematik.tu-
muenchen.de/m4/Papers/Tasche/riskcon.pdf.
Tauman, Y. (1988). The AumannShapley prices: a survey. Chapter 18 of The Shapley Value.
Essays in honor of Lloyd S. Shapley. A. Roth (ed.). (Cambridge: Cambridge University
Press).
Young, H. P. (1994). Cost allocation. Chapter 34 of Handbook of Game Theory, Volume 2.
R. J. Aumann and S. Hart (eds). (Elsevier Science).
Michel Denault34
Denault pages 23/11/01 4:18 pm Page 34