See

discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/225341862

The two-stage constrained

equal awards and losses rules
for multi-issue allocation
situations

Article in Top · January 2005

Impact Factor: 0.83 · DOI: 10.1007/s11750-009-0073-8

CITATIONS READS

11 27

3 authors, including:

Silvia Lorenzo-Freire
University of A Coruña
20 PUBLICATIONS 129 CITATIONS

SEE PROFILE

Available from: Silvia Lorenzo-Freire

Retrieved on: 09 May 2016
Top
DOI 10.1007/s11750-009-0073-8
O R I G I N A L PA P E R

The two-stage constrained equal awards and losses

rules for multi-issue allocation situations

Silvia Lorenzo-Freire · Balbina Casas-Méndez ·

Ruud Hendrickx

Received: 19 August 2008 / Accepted: 30 December 2008

© Sociedad de Estadística e Investigación Operativa 2009

Abstract This paper considers two-stage solutions for multi-issue allocation situa-
tions, which are extensions of bankruptcy problems. Characterizations are provided
for the two-stage constrained equal awards and constrained equal losses rules, based
on the properties of composition up and composition down.

Keywords Bankruptcy problems · Multi-issue allocation situations · Constrained

equal awards · Constrained equal losses · Two-stage solutions

Mathematics Subject Classification (2000) 91A12 · 91B14

1 Introduction

In a bankruptcy problem, there is an estate to be divided among a number of

claimants, whose total claim exceeds the estate available. Calleja et al. (2005) intro-
duced multi-issue allocation situations to model bankruptcy-like problems in which
the estate is divided not on the basis of a single claim for each agent, but bearing
in mind several claims resulting from so-called issues. Many real situations can be
modeled by means of multi-issue allocation situations.

S. Lorenzo-Freire
Departamento de Matemáticas, Facultad de Informática, Universidad de Coruña, 15071 La Coruña,
Spain

B. Casas-Méndez ()
Departamento de Estadística e Investigación Operativa, Facultad de Matemáticas, Universidad
de Santiago de Compostela, 15782 Santiago de Compostela, Spain
e-mail: balbina.casas.mendez@usc.es

R. Hendrickx
Department of Econometrics and Operations Research, Faculty of Economics, Tilburg University,
Tilburg, The Netherlands
 S. Lorenzo-Freire et al.

We can consider, for example, the allocation of the budget of the European Union.
Each year the European Union must allocate money to different issues, such as agri-
culture and natural resources, freedom, security and justice, citizenship programmes,
external relation actions, etc. Moreover, the countries which are members of the Eu-
ropean Union have different necessities on these issues.
Another example may be a University with different libraries placed in several
faculties. The central services of the University decide how to allocate some money
(the estate) to these libraries (issues). Each department (agent) has a number of claims
on the estate. A department may not only be implied in one faculty but also in several
faculties, being possible to have claims in different libraries.
As a solution to such a multi-issue allocation situation, Calleja et al. (2005) present
two extensions of the run-to-the-bank rule for bankruptcy situations by O’Neill
(1982). These extensions reflect the significance of the issues for the division of the
estate in a sophisticated way, using compensation payments.
One drawback of their approach, however, is that although the compensations form
an elegant mechanism to deal with the issues, the rules are difficult to compute. More-
over, the characterizations of the two rules (using O’Neill-like consistency properties)
are rather complex.
González-Alcón et al. (2007) solve this complexity problem by constructing a dif-
ferent extension of the run-to-the-bank rule. The idea behind their two-stage approach
is simple: firstly, one should explicitly allocate money to the issues, and secondly,
the amount allocated to each issue should be divided among the players. In fact,
González-Alcón et al. (2007) do not simply apply the bankruptcy run-to-the-bank
rule twice, but rather the underlying run-to-the-bank (marginal) vectors and then take
the average. They provide a relatively easy characterization in terms of a property
called issue consistency.
This two-stage approach is also taken in Borm et al. (2005). They study bankruptcy
problems with a priori unions and use the same idea to first allocate money to the
unions and then redivide the money within each union.
In this paper, we apply the two-stage idea to the constrained equal awards and con-
strained equal losses rules. We characterize the two-stage constrained equal awards
rule using properties based on equal treatment (O’Neill 1982), composition up (the
property of composition was introduced in the context of taxation by Young (1987)
and Thomson (2001), who called it “composition up”), and invariance under claims
truncation (Curiel et al. 1987). Our characterization of the two-stage constrained
equal losses rule is based on duality and uses properties based on composition down
(the property of composition down is a property of path independence. Although the
properties of path independence are very common in the context of social choice, the
idea of path independence was first applied in the context of bankruptcy problems
by Moulin (2000). In Thomson (2001) this property of path independence is called
“composition down”) and composition from minimal rights (this property was first
introduced by Curiel et al. 1987 under the name of “minimal rights property.” Dagan
1996 refers to it as “v-separability,” and we use the name employed by Herrero and
Villar 1998, 2001).
A surprising feature of our characterizations is that the two-stage construction
does not follow straightforwardly from the properties used but is a consequence of
The two-stage constrained equal awards and losses rules

the strength of composition up and composition down. Even though our rules satisfy a
property of consistency in two stages that clearly reflects the two-stage scheme, con-
sistency is not used in the characterization result. We only need the weaker quotient
property together with composition up and a few other basic requirements to charac-
terize the two-stage constrained equal awards rule. Similarly, if we replace composi-
tion up by composition down, we obtain a characterization of the constrained equal
losses rule.
The outline of the paper is as follows. We present some necessary preliminaries
and introduce two-stage multi-issue allocation rules in Sect. 2. We characterize the
two-stage constrained equal awards rule in Sect. 3 and the constrained equal losses
rule in Sect. 4.

2 Two-stage rules for multi-issue allocation situations

2.1 Preliminaries

In this subsection, we give some basic definitions related to multi-issue allocation

situations that we use through our paper.
A bankruptcy problem (O’Neill 1982) is a triple (N, E, c) such that N is a finite
set of players. We denote n = |N |. E ≥ 0 represents the estate which is the available
amount to satisfy the players’ claims, and c ∈ RN + is the vector of claims, where ci
denotes the claim of player i. The main assumption of a bankruptcy  problem is that
the estate is not sufficient to satisfy all the claims, i.e., 0 ≤ E ≤ i∈N ci .
A bankruptcy rule is a function ψ which associates with every bankruptcy problem
(N, E, c) a vector ψ(N, E, c) ∈ RN such that:
• 0
≤ ψ(N, E, c) ≤ c (reasonability).
• i∈N ψi (N, E, c) = E (efficiency).

A multi-issue allocation (MIA) situation (Calleja et al. 2005) is a 4-tuple

(R, N, E, C), where R and N are finite sets of issues and players, respectively. We
denote r = |R| and n = |N|. E ≥ 0 is the estate to be divided, and C ∈ RR×N + is
the matrix of claims. An element cki represents the amount claimed by player i ∈ N
according to the issue k ∈ R. We assume that there is a positive
 claim in each issue
and for each player. Moreover, we assume that 0 ≤ E ≤ k∈R i∈N cki . Note that a
bankruptcy problem is an MIA situation with |R| = 1. For all k ∈ R, Ck = (cki )i∈N
denotes the vector of claims according to issue k ∈ R.
A multi-issue allocation (MIA) solution Ψ is a function that associates with every
MIA situation (R, N, E, C) a matrix Ψ (R, N, E, C) ∈ RR×N satisfying reasonabil-
ity and efficiency:
• 0≤ Ψ 
(R, N, E, C) ≤ C.
• k∈R i∈N Ψki (R, N, E, C) = E.

Note that Calleja et al. (2005) and González-Alcón et al. (2007) consider vector-
valued MIA solutions (in RN ) rather than matrix solutions. We need solutions in
RR×N in order to properly define composition up and composition down, both of
 S. Lorenzo-Freire et al.

which deal with reduced claims matrices. Of course, any matrix outcome gives rise
to a vector solution by just adding up all the amounts over the issues.
A bankruptcy situation with a priori unions (Borm et al. 2005) gives rise to an
MIA situation, where the issues correspond to the unions. In the associated claims
matrix, the issues are disjoint in the sense that each player has a claim on just one
issue.

2.2 Two-stage MIA rules

In this subsection, we define a natural two-stage procedure to define MIA solutions

from bankruptcy rules.

Definition 1 Let ψ be a bankruptcy rule, and let (R, N, E, C) be an MIA situation.

The two-stage solution Ψ ψ (R, N, E, C) is the MIA solution obtained from the fol-
lowing two-stage procedure.
First stage: Consider the so-called quotient bankruptcy problem (R, E, cR ), where
c = (c1R , . . . , crR ) ∈ RR denotes the vector of total claims in the issues, i.e., ckR =

R

i∈N cki for all k ∈ R. Divide the amount E among the issues using the bankruptcy
rule ψ. In this way, we obtain ψ(R, E, cR ) ∈ RR .
Second stage: For each k ∈ R, consider a new bankruptcy problem for the players
(N, ψk (R, E, cR ), Ck ) and apply the same bankruptcy rule ψ to this new bankruptcy
problem. So, we obtain ψ(N, ψk (R, E, cR ), Ck ) ∈ RN for each k ∈ R.

Figure 1 shows the whole procedure.

A rule Ψ satisfies consistency in two stages if for each MIA situation (R, N, E, C)
and for all k ∈ R, i ∈ N ,
   
Ψki (R, N, E, C) = Ψ1i {1}, N, Ψk1 R, {1}, E, cR , Ck ,

where (R, {1}, E, cR ) is the quotient MIA situation, which is an MIA situation with
only one player, where the claim of each issue equals the sum of all the players’
claims in that issue. From the definition, it is easy to deduce that each two-stage rule
Ψ ψ satisfies consistency in two stages.

⎛ ⎞
s11 ... sn1
Ψψ  Ψ ψ (N, R, E, C) = ⎜ .
⎝ ..
.. ⎟
(N, R, E, C) . ⎠
s1r ... snr

⎧ R
ψ 1⎫
 ψ ⎪
⎨ (N, ψ1 (R, E, c ), C1 )
s ⎪
⎬
..
(R, E, cR )  ψ(R, E, cR )  ..
. .⎪
⎪
⎩ ψ ⎭
(N, ψr (R, E, cR ), Cr )  sr
Fig. 1 Two-stage solutions: the whole procedure
The two-stage constrained equal awards and losses rules

Consistency in two stages is equivalent to the combination of two properties. The

first of these properties is the quotient property. A rule Ψ satisfies the quotient prop-
erty if for each MIA situation (R, N, E, C) and for all k ∈ R,
  
Ψki (R, N, E, C) = Ψk1 R, {1}, E, cR .
i∈N

This property means that the total quantity assigned to an issue in an MIA situation
is equal to the amount assigned to the same issue in the quotient MIA situation.
To motivate the quotient property we can think about an arbitror we must use who
has to solve a multi-issue allocation situation. The quotient property establishes that
the arbitror decides to assign to the issues the same quantity independently of the way
the demands in the issues are divided among the agents.
The second related property is consistency within the issues, which is a direct
translation of the (self-)consistency property introduced by Aumann and Maschler
(1985). An MIA rule Ψ satisfies consistency within the issues if for each MIA situa-
tion (R, N, E, C) and for all k ∈ R, i ∈ N ,
  
Ψki (R, N, E, C) = Ψ1i {1}, N, Ψkj (R, N, E, C), Ck .
j ∈N

This property says that redividing the total amount within an issue using the same
rule Ψ should yield the same outcome.

3 The two-stage constrained equal awards rule

In this section, we provide a characterization of the two-stage constrained equal

awards rule. The constrained equal awards rule for bankruptcy situations (CEA)
 defined by CEAi (N, E, c) = min{λ, ci } CEA
is for all i ∈ N , where λ is such that
i∈N min{λ, c i } = E. The two-stage rule Ψ is then given, for all (R, N, E, C)
and for all k ∈ R, i ∈ N , by

ΨkiCEA (R, N, E, C) = min{βk , cki },


where for all k ∈ R, βk is such that i∈N min{βk , cki } = min{λ, ckR }, and λ is such

that k∈R min{λ, ckR } = E.
Note that Ψ CEA differs from another natural extension of the CEA rule defined by

Ψki (R, N, E, C) = min{λ, cki } (1)

 
for every i ∈ N and k ∈ R with λ such that k∈R i∈N min{λ, cki } = E. A big
distinction between the two extensions is that the latter does not satisfy the quotient
property.
Below, we list some properties which we use to characterize the two-stage CEA
rule. The properties of composition up and invariance under claims truncation are
defined in MIA situations following the same spirit as in bankruptcy situations. Nev-
ertheless, the bankruptcy property of equal treatment of equals cannot be extended so
 S. Lorenzo-Freire et al.

easily. We have decided to extend it by applying the property twice: first among the
issues and secondly inside each issue.

Composition up A rule Ψ satisfies composition up if for all (R, N, E, C) and for

all 0 ≤ E  ≤ E, we have
 
Ψ (R, N, E, C) = Ψ (R, N, E  , C) + Ψ R, N, E − E  , C − Ψ (R, N, E  , C) .

According to this property, we can divide the total estate among the issues and the
players using two different procedures resulting in the same outcome. In the first
procedure, we divide the total estate directly using Ψ . In the other procedure, we first
divide a part E  of the estate and then divide the remainder E − E  on the basis of
the remaining claims, both times using Ψ . To motivate this property we take up again
the example of the European Union. Suppose that the total budget of the European
Union is divided in two different dates. It seems reasonable to require that the total
quantity obtained for each country does not depend on the way the budget is divided.

Monotonicity in the estate A rule Ψ satisfies monotonicity in the estate if for

all (R, N, E, C) and (R, N, E  , C) such that 0 ≤ E  ≤ E, Ψ (R, N, E  , C) ≤
Ψ (R, N, E, C).
Notice that monotonicity in the estate can be obtained as consequence of compo-
sition up.

Invariance under claims truncation A rule Ψ satisfies invariance under claims trun-
cation if for each MIA situation (R, N, E, C), we have
 
Ψ (R, N, E, C) = Ψ R, N, E, C E ,

where C E ∈ RR×N
+
E = min{c , E} for all k ∈ R and i ∈ N .
is such that cki ki
This property says that truncating each claim to the estate does not influence the
outcome. In other words, we consider the claims bigger than the estate as irrelevant
claims.

Equal treatment for the players within an issue A rule Ψ satisfies this property if for
each MIA situation (R, N, E, C) and for all k ∈ R and i, j ∈ N such that cki = ckj ,

Ψki (R, N, E, C) = Ψkj (R, N, E, C).

Equal treatment for the issues A rule Ψ satisfies equal treatment for the issues if
for all MIA situations (R, N, E, C) and for all k, k  ∈ R such that ckR = ckR ,
   
Ψk1 R, {1}, E, cR = Ψk  1 R, {1}, E, cR .

We have decided to extend the property of equal treatment of equals in the case
of bankruptcy problems using two properties: according to the first property, if the
total claim over two issues is the same, the total amount assigned to both issues must
coincide. In the case of the second property, if two players have the same claim in an
issue, then both agents must receive the same amount in this issue.
The two-stage constrained equal awards and losses rules

Proposition 1 The two-stage constrained equal awards rule, Ψ CEA , satisfies compo-
sition up, invariance under claims truncation, equal treatment for the players within
an issue, equal treatment for issues, and the quotient property.

Proof We only check that Ψ CEA satisfies the properties of composition up and in-
variance under claims truncation. The remaining properties follow immediately from
the definitions. Let (R, N, E, C) be an MIA situation.
(a) Composition up. As composition up is satisfied by CEA for bankruptcy prob-
lems, taking the MIA situation (R, {1}, E, cR ) we have that for all 0 ≤ E  ≤ E,
   
Ψ CEA R, {1}, E, cR = Ψ CEA R, {1}, E  , cR
  
+ Ψ CEA R, {1}, E − E  , cR − Ψ CEA R, {1}, E  , cR .

Let k ∈ R. By monotonicity, 0 ≤ ΨkCEA (R, {1}, E  , cR ) ≤ ΨkCEA (R, {1}, E, cR ). We

apply composition up taking into account these quantities:
   
Ψ CEA {k}, N, ΨkCEA R, {1}, E, cR , Ck
   
= Ψ CEA {k}, N, ΨkCEA R, {1}, E  , cR , Ck
     
+ Ψ CEA {k}, N, ΨkCEA R, {1}, E, cR − ΨkCEA R, {1}, E  , cR , Ck ,

where Ck = Ck − Ψ CEA ({1}, N, ΨkCEA (R, {1}, E  , cR ), Ck ).

Finally, using the definition of Ψ CEA and gathering the results of all issues in a
matrix, we obtain
Ψ CEA (R, N, E, C) = Ψ CEA (R, N, E  , C)
 
+ Ψ CEA R, N, E − E  , C − Ψ CEA (R, N, E  , C) .

(b) Invariance under claims truncation. Using the definition of Ψ CEA and the fact
that the CEA rule for bankruptcy problems satisfies invariance under claims trunca-
tion, we get for all k ∈ R and i ∈ N ,
 
ΨkiCEA R, N, E, C E
   R  E 
= Ψ1iCEA {1}, N, Ψk1 CEA
R, {1}, E, cE , Ck
   R E  
= Ψ1iCEA {1}, N, Ψk1 CEA
R, {1}, E, cE , CkE
   E  E 
= Ψ1iCEA {1}, N, Ψk1 CEA
R, {1}, E, cR , Ck
   
= Ψ1iCEA {1}, N, Ψk1 CEA
R, {1}, E, cR , CkE
    Ψ CEA (R,{1},E,cR ) 
= Ψ1iCEA {1}, N, Ψk1 CEA
R, {1}, E, cR , CkE k1
   Ψ CEA (R,{1},E,cR ) 
= Ψ1iCEA {1}, N, Ψk1
CEA
R, {1}, E, cR , Ck k1
   
= Ψ1iCEA {1}, N, Ψk1
CEA
R, {1}, E, cR , Ck
= ΨkiCEA (R, N, E, C). 
 S. Lorenzo-Freire et al.

Theorem 1 The only rule which satisfies composition up, invariance under claims
truncation, equal treatment for the players within an issue, equal treatment for issues,
and the quotient property is the two-stage constrained equal awards rule Ψ CEA .

Proof In view of Proposition 1, we only need to prove uniqueness. Let (R, N, E, C)

be an MIA situation, and let Ψ be a rule satisfying the five properties. The proof
follows the lines in Dagan (1996). To clarify the proof, we divide it in several parts.

Claim 1 The rule Ψ applied in the quotient problem coincides with CEA, i.e.,
   
Ψ R, {1}, E, cR = Ψ CEA R, {1}, E, cR . (2)

We omit the proof of this claim because it is an adaptation of the proof of Dagan
(1996) to the case of the bankruptcy problem (R, E, cR ).
Then, by Claim 1 and the quotient property,
  
Ψki (R, N, E, C) = Ψk1
CEA
R, {1}, E, cR (3)
i∈N

for all k ∈ R.
Define Nk = Nk (C) = {i ∈ N : cki > 0} and ρk = ρk (C) = |Nk |mini∈Nk cki for all
k ∈ R, and define ρ = ρ(C) = mink∈R ρk . Without loss of generality, assume that
claim c11 determines this minimum, i.e., ρ = |N1 |c11 .

Claim 2 If E ≤ rρ, then Ψ coincides with Ψ CEA , i.e.,

Ψ (R, N, E, C) = Ψ CEA (R, N, E, C). (4)

Proof of Claim 2 Take E 1 = mink∈R mini∈Nk cki (note that this minimum is not nec-
essarily attained by c11 ). Moreover, we define recursively for all t ∈ N, t ≥ 2,
 
E t−1
E =E
t t−1
+ min min cki − .
k∈R i∈Nk r|Nk |

Note that E t ≥ E t−1 for every t ≥ 2.

We will prove later that there exists s ∈ N such that for all s  ∈ N, s  ≥ s,
   
Es Es
min min cki − = c11 − . (5)
k∈R i∈Nk r|Nk | r|N1 |
This means that at some stage of the procedure, the minimal remaining claim must
shift to the (1, 1)-position and stay there.
Suppose that (5) is true. We may distinguish two possibilities:
• E ≤ Es .
If E ≤ E 1 , then (3), invariance under claims truncation, and equal treatment for
the players within an issue imply (4). In this way, for all k ∈ R, i ∈ N ,
E
Ψki (R, N, E, C) = = ΨkiCEA (R, N, E, C).
r|Nk |
The two-stage constrained equal awards and losses rules

If E 1 < E ≤ E 2 , then the first case, together with composition up, yields
 
Ψ (R, N, E, C) = Ψ R, N, E 1 , C
  
+ Ψ R, N, E − E 1 , C − Ψ R, N, E 1 , C
 
= Ψ CEA R, N, E 1 , C
  
+ Ψ R, N, E − E 1 , C − Ψ CEA R, N, E 1 , C ,

where the first equality follows from composition up, and the second is a direct con-
sequence of the case E ≤ E 1 . Besides,
 
E1
E − E 1 ≤ E − E 2 = min min cki −
k∈R i∈Nk r|Nk |
  
= min min cki − ΨkiCEA R, N, E 1 , C ,
k∈R i∈Nk

where the first equality follows from composition up, and the second is a direct con-
sequence of the case E ≤ E 1 .
Again, because of invariance under claims truncation, equal treatment of the is-
sues, and equal treatment for the players within an issue, we divide the amount
E − E 1 into identical parts, and hence
  
Ψ R, N, E − E 1 , C − Ψ CEA R, N, E 1 , C
  
= Ψ CEA R, N, E − E 1 , C − Ψ CEA R, N, E 1 , C

and, given that Ψ CEA satisfies composition up,

Ψ (R, N, E, C) = Ψ CEA (R, N, E, C).

We follow this procedure for all E t−1 < E ≤ E t with 2 ≤ t ≤ s to obtain that
Ψ (R, N, E, C) = Ψ CEA (R, N, E, C).
• E > Es .
s
Suppose that E s < E ≤ E s+1 = E s + (c11 − r|N
E
1|
), where the equality follows
s
from definition of E and (5). In this case,
 
Ψ (R, N, E, C) = Ψ R, N, E s , C
  
+ Ψ R, N, E − E s , C − Ψ CEA R, N, E s , C
 
= Ψ CEA R, N, E s , C
  
+ Ψ R, N, E − E s , C − Ψ CEA R, N, E s , C ,

where the first equality follows from composition up, and the second is a direct con-
sequence of the case E ≤ E s . Besides,
Es
E − E s ≤ E s+1 − E s = c11 −
r|N1 |
 S. Lorenzo-Freire et al.
 
Es
= min min cki −
k∈R i∈Nk r|Nk |
  
= min min cki − ΨkiCEA R, N, E s , C
k∈R i∈Nk
 
= c11 − Ψ11
CEA
R, N, E s , C .

Again, because of invariance under claims truncation, equal treatment for the is-
sues, and equal treatment for the players within an issue, we have
  
Ψ R, N, E − E s , C − Ψ CEA R, N, E s , C
  
= Ψ CEA R, N, E − E s , C − Ψ CEA R, N, E s , C

and, given that Ψ CEA satisfies composition up,

Ψ (R, N, E, C) = Ψ CEA (R, N, E, C).

We continue using this procedure for all E s+p−1 < E ≤ E s+p , p ≥ 2, with
 p−1 t
Es  1
E s+p = E s + c11 − 1− .
r|N1 | r|N1 |
t=0
∞
t=0 (1− r|N1 | ) = r|N1 | and thus Ψ(R, N, E, C) = Ψ
1 t CEA (R, N, E, C)
Note that
Es
for all E < E s + [c11 − r|N 1|
]r|N1 | = r|N1 |c11 = rρ.
By monotonicity and continuity in the estate, Ψ(R, N, E, C) = Ψ CEA (R, N, E, C)
also holds for E = rρ.
In order to finish the proof of Claim 2, we still have to show the validity of (5).
First, we show that there exist s ∈ N and k  ∈ R such that for all s  ∈ N, s  ≥ s,
   
Es Es
min min cki − = min ck i −
 .
k∈R i∈Nk r|Nk | i∈Nk  r|Nk  |
This means that at some stage of the procedure the minimal remaining claim must
shift to one position and stay there.
Let s̃ ∈ N and k̃ ∈ R be such that
E s̃−1
E s̃ = E s̃−1 + min ck̃i − .
i∈Nk̃ r|Nk̃ |

Then,
E s̃−1 E s̃−1
min ck̃i − ≤ min cki − (6)
i∈Nk̃ r|Nk̃ | i∈Nk r|Nk |
for any k ∈ R.
Let k ∈ R with |Nk | ≥ |Nk̃ |. Then
   
1 1 1 1
min ck̃i − min cki ≤ E s̃−1 − ≤ Et −
i∈Nk̃ i∈Nk r|Nk̃ | r|Nk | r|Nk̃ | r|Nk |
The two-stage constrained equal awards and losses rules

for any t ≥ s̃, where the first inequality follows from (6), and the second is a direct
consequence of E s̃−1 ≤ E t for t ≥ s̃ together with r|N1 | − r|N1 k | ≥ 0. Thus,
k̃

Et Et
E t + min ck̃i − ≤ E t + min cki − ,
i∈Nk̃ r|Nk̃ | i∈Nk r|Nk |

and E t+1 is achieved either in k̃ or in k ∈ R with |Nk | < |Nk̃ |. Since the number of
issues is finite, there exist s ∈ N and k  ∈ R such that the minimum is always achieved
in k  for every s  ≥ s.
Next, we only have to show that k  = 1. Since the minimum is always achieved in

k for every s + p with p ≥ 0, we have

E s+p E s+p
c11 − ≥ min ck  i − (7)
r|N1 | i∈Nk r|Nk  |

and
 
 p−1 t
Es 1
E s+p = E s + min ck  i − 1− . (8)
i∈Nk  r|Nk  | r|Nk  |
t=0

Replacing E s+p in (7) by the expression given in (8) and taking limits in p, we
get
r|Nk  |
c11 − min ck  i ≥ 0
r|N1 | i∈Nk
since
∞ 
 t
1
1− = r|Nk  |.
r|Nk  |
t=0

Therefore,
rρ = rρ1 = r|N1 |c11 ≥ r|Nk  | min ck  i = rρk  ,
i∈Nk 

where the first equality follows from our initial assumption. Since ρ = mink∈R ρk ,
it follows that ρ = ρk  . Then, we first rename the players in such a way that
ck  1 = mini∈Nk ck  i , and subsequently we rename the issues and call 1 to issue k  . 

Claim 3 If E > rρ, then Ψ coincides with Ψ CEA , i.e.,

Ψ (R, N, E, C) = Ψ CEA (R, N, E, C). (9)

Proof of Claim 3 Consider E 1 = rρ and C 1 = C. By composition up and Claim 2,

 
Ψ (R, N, E, C) = Ψ CEA R, N, E 1 , C
  
+ Ψ R, N, E − E 1 , C − Ψ CEA R, N, E 1 , C .
 S. Lorenzo-Freire et al.

Note that in the matrix of remaining claims, there is at least one more zero than
in the original claims matrix because c11 − Ψ11 CEA (R, N, E 1 , C) = 0. For these re-

maining claims, we redefine Nk and ρk for all k ∈ R and ρ in the same way, and we
reapply the procedure. The only thing that we have to take care of is that there might
be an issue with only zero claims remaining. In this case, this issue is not taken into
account when determining ρ since the players automatically receive zero because of
reasonability. In this way, we define for t ∈ N, t ≥ 2,
 
C t = C − Ψ CEA R, N, E t−1 , C ,
 
Nkt = i ∈ N : cki t
> 0 for each k ∈ R,
 
R t = k ∈ R : Nkt = ∅ with r t = |R t |,
 
ρkt = Nkt  min ckit
for each k ∈ R t ,
i∈Nkt

ρ t = min ρkt ,
k∈R t

E t = E t−1 + r t ρ t .
 
Consider s = min{t ∈ N : E t = k∈R i∈N cki }. Notice that s ≤ rn since, in each
step, at least one
remaining
 claim becomes zero.
Since E ≤ k∈R i∈N cki , we know that there exists t ≤ s such that E t−1 <
E ≤ Et .
Applying t times composition up and Claim 2, we obtain


t−1
   
Ψ (R, N, E, C) = Ψ R, N, E p , C p + Ψ R, N, E − E t−1 , C t
p=1


t−1
   
= Ψ CEA R, N, E p , C p + Ψ CEA R, N, E − E t−1 , C t
p=1

= Ψ CEA (R, N, E, C),

obtaining that (9) holds. 

Now, taking into account Claims 2 and 3, we conclude the proof of the theorem. 

Remark 1 A nice feature of the characterization in Theorem 1 is that although the

two-stage CEA rule satisfies consistency in two stages, we only need the quotient
property and not consistency within the issues. Apparently, composition up for MIA
solutions is quite a strong generalization of composition up for bankruptcy rules. We
have a similar feature in the context of classical bankruptcy problems where compo-
sition up joint with basic requirements of equal treatment and invariance under claims
truncation allow the characterization of the equal awards rule. The difference in the
case of multi-issue allocation situations is that the property of equal treatment splits
into two different properties that are equal treatment for issues and equal treatment
The two-stage constrained equal awards and losses rules

for the players within an issue. Moreover, we include the quotient property. A more
straightforward, but less interesting, characterization could be obtained by including
consistency within the issues and weakening composition up by replacing it with
composition up between the issues and composition up within the issues.

Proposition 2 The properties in Theorem 1 are independent.

Proof First, the rule defined by (1) satisfies all properties in the theorem except the
quotient property.
Both the two-stage constrained equal losses rule (which we will discuss in the next
section) and the two-stage proportional rule (O’Neill 1982) satisfy the properties of
composition up, equal treatment for the issues, equal treatment for the players within
an issue, and the quotient property. However, neither of these rules is invariant under
claims truncation.
On the other hand, the recursive completion rule (O’Neill 1982) in two stages,
the contested garment consistent rule (Aumann and Maschler 1985) in two stages,
and the adjusted proportional rule (Curiel et al. 1987) in two stages satisfy invariance
under claims truncation, equal treatment for the issues, equal treatment for the players
within an issue, and quotient problem but do not satisfy the property of composition
up.
Next, we introduce a rule that satisfies all the properties in Theorem 1 except equal
treatment for the players within an issue. In the first stage, the rule applies the CEA
bankruptcy rule to the quotient bankruptcy problem. In the second stage, it applies
the bankruptcy rule f 1 defined for all i ∈ N by

min{E, c1 } if i = 1,
fi1 (N, E, c) =
CEAi (N, E − min{E, c1 }, (0, c−1 )) otherwise,

where c−1 is the vector of claims obtained from c after eliminating its first compo-
nent c1 .
According to f 1 , the first agent has a priority, and he obtains his total claim. It is
well known that priority rules satisfy composition up (Moulin 2000). The rest of them
are paid according to the CEA bankruptcy rule over the remaining capital assigned to
the issue.
In this way, we have that the rule for the multi-issue allocation situation for all
k ∈ R and i ∈ N is given by

min{min{λ, ckR }, ck1 } if i = 1,
Ψk1 (R, N, E, C) =
min{βk , cki } otherwise,

where for all k ∈ R, i∈N \{1} min{βk , cki } = min{λ, ckR } − min{min{λ, ckR }, ck1 } and

where λ is such that k∈R min{λ, ckR } = E.
Finally, to define a rule which satisfies all the properties in Theorem 1 except
equal treatment for the issues, we apply, in the first stage, the bankruptcy rule f 1 to
the quotient problem and, in the second stage, the CEA rule. In this way, we have that
for all i ∈ N and all k ∈ R,

Ψki (R, N, E, C) = min{βk , cki },

 S. Lorenzo-Freire et al.


where β1 is such that i∈N min{β1 , c1i } = min{E, c1R }, and for all k ∈ R\{1}, βk is

such that i∈N min{βk , cki } = min{λ, ckR } with λ such that
    
min λ, ckR = E − min E, c1R .
k∈R\{1} 

4 The two-stage constrained equal losses rule

This section considers the constrained equal losses rule defined for a bankruptcy
 (N, E, c) by CELi (N, E, c) = max{0, ci − λ} for all i ∈ N , where λ is such
problem
that i∈N max{0, ci − λ} = E. The two-stage extension Ψ CEL is then given, for all
(R, N, E, C) and all k ∈ R and i ∈ N , by

ΨkiCEL (R, N, E, C) = max{0, cki − βk },


where for all k ∈ R, βk is such that i∈N max{0, cki − βk } = max{0, ckR − λ}, and λ

is such that k∈R max{0, ckR − λ} = E.
We now mention some properties for MIA solutions, which we use to characterize
the two-stage constrained equal losses rule.

Composition down An MIA solution Ψ satisfies composition down if for each MIA
situation (R, N, E, C) and for all E  ∈ R such that E  ≥ E,
 
Ψ (R, N, E, C) = Ψ R, N, E, Ψ (R, N, E  , C) .

If a rule Ψ satisfies composition down, we can divide the estate using two procedures
yielding the same result. The first procedure is to divide the estate directly using Ψ . In
the second procedure, we first divide a bigger estate E  ≥ E and then use the outcome
Ψ (N, R, E  , C) as claims matrix to divide the real estate E, both times using Ψ . In
this case, suppose that the central services of a University divide a budget among its
libraries based on a estimate of its value. Assume that after doing this division, it turns
out that the actual budget is shorter than the initial estimate. The property establishes
that it is the same to solve the actual problem or to apply the rule to the situation in
which the initial claims are changed by the unfeasible allocation initially proposed.

Composition from minimal rights An MIA solution Ψ satisfies composition from

minimal rights if for each MIA situation (R, N, E, C),

Ψ (R, N, E, C) = m(R, N, E, C)
  
+ Ψ R, N, E − mki (R, N, E, C), C − m(R, N, E, C) ,
k∈R i∈N

where the minimum right of player i in issue k is given by

  
mki (R, N, E, C) = max 0, E − cj .
∈R,j ∈N
(,j )=(k,i)
The two-stage constrained equal awards and losses rules

If a rule Ψ satisfies composition from minimal rights, we can divide the total
estate among the issues and the players using two different procedures. In the first
procedure, we divide the total estate directly using Ψ . In the second procedure, we
first give to each player i and each issue k the corresponding minimal right and then
divide the remainder estate on the basis of the updated claims using Ψ . This is a
variant of composition up.

Duality An MIA solution Ψ ∗ is the dual of another MIA solution Ψ if for each
MIA situation (R, N, E, C),
  
∗
Ψ (R, N, E, C) = C − Ψ R, N, cki − E, C .
k∈R i∈N

This property asserts that one rule is the dual of another one if it assigns to each
player in each issue what this player demands minus what this agent obtains dividing
the total losses according to the other rule. A rule Ψ is called self-dual if Ψ ∗ = Ψ .
The following lemma follows immediately from the observation that both CEA
and CEL are dual rules for bankruptcy situations.

Lemma 1 (Ψ CEA )∗ = Ψ CEL .

The property P ∗ is the dual property of P if for every MIA solution Ψ , we have
that Ψ satisfies property P if and only if the dual MIA solution Ψ ∗ satisfies prop-
erty P ∗ . A property is called self-dual when it is dual of itself. The next two lemmas
are extensions of results from Herrero and Villar (2001). The proofs are similar and
therefore omitted.

Lemma 2 If the MIA solution Ψ is characterized by means of independent proper-

ties, then the dual MIA solution Ψ ∗ is characterized by the dual properties. Moreover,
these dual properties are independent.

Lemma 3
• Composition up and composition down are dual properties.
• Invariance under claims truncation and composition from minimal rights are dual
properties.
• The quotient property, equal treatment for the players within an issue, and equal
treatment in the issues are self-dual.

Using the two previous lemmas and our characterization of the two-stage con-
strained equal awards rule, we characterize the constrained equal losses rule in the
following theorem.

Theorem 2 There is only one rule which satisfies the properties of composition down,
composition from minimal rights, equal treatment for the players within an issue,
equal treatment for issues, and the quotient property. This rule is the two-stage con-
strained equal losses rule Ψ CEL .
 S. Lorenzo-Freire et al.

Similar to the comment following Theorem 1, one can state that composition down
for MIA solutions is a strong generalization of composition down for bankruptcy
rules.

Acknowledgements The authors would like to thank Rodica Branzêi and four anonymous referees for
some useful comments. Financial support from Netherlands Organisation for Scientific Research (NWO)
is gratefully acknowledged by Ruud Hendrickx and from Spanish Ministry of Science and Technology and
FEDER through projects SEJ2005-07637-C02-01 and 02 and ECO2008-03484-C02-01 and 02 by Balbina
Casas and Silvia Lorenzo.

References

Aumann R, Maschler M (1985) Game theoretic analysis of a bankruptcy problem from the Talmud. J Econ
Theory 36:195–213
Borm P, Carpente L, Casas-Méndez B, Hendrickx R (2005) The constrained equal awards rule for bank-
ruptcy problems. Ann Oper Res 137:211–227
Calleja P, Borm P, Hendrickx R (2005) Multi-issue allocation situations. Eur J Oper Res 164:730–747
Curiel I, Maschler M, Tijs S (1987) Bankruptcy games. Z Oper Res (Now: Math Methods Oper Res)
31:143–159
Dagan N (1996) New characterizations of old bankruptcy rules. Soc Choice Welf 13:51–59
González-Alcón C, Borm P, Hendrickx R (2007) A composite run-to-the-bank rule for multi-issue alloca-
tion situations. Math Methods Oper Res 65:339–352
Herrero C, Villar A (1998) Preeminence and sustainability in bankruptcy problems. IVIE working papers
AD number 17
Herrero C, Villar A (2001) The three musketeers: four classical solutions to bankruptcy problems. Math
Soc Sci 42:307–328
Moulin H (2000) Priority rules and other asymmetric rationing methods. Econometrica 68:643–684
O’Neill B (1982) A problem of rights arbitration from the Talmud. Math Soc Sci 2:345–371
Thomson W (2001) On the axiomatic method and its recent applications to game theory and resource
allocation. Soc Choice Welf 18:327–387
Young HP (1987) On dividing an amount according to individual claims or liabilities. Math Oper Res
12:398–414