Upload
mmpanero
View
215
Download
0
Embed Size (px)
Citation preview
7/27/2019 dp06-04
1/28
1
Characterizations of Proportional
Rules in Claims Problems
Hirofumi YAMAMURA
Discussion Paper No. 2006-04
April 14, 2006
7/27/2019 dp06-04
2/28
2
Characterizations of Proportional
Rules in Claims Problems*
Hirofumi YAMAMURA+
Tokyo Institute of Technology
April 14, 2006
Abstract
When resources are divided among agents, resources are in many cases divided
proportionally to their claims. In this paper, we provide new axiomatizations of
generalized proportional rules based on the axiom decentralizability due to Moulin
[13,14,15]. Decentralizability requires that no reallocation by any coalition can affect
the awards of the agents outside the coalition. Our results can explain the results of
Chun [7] and J u=Miyagawa=Sakai [10] based on non-manipulability requirements as
corollaries.
We moreover introduce a coalitional form game called a claims reallocation game
which describes a claims problem under any established division rule. We require the
core of a claims reallocation game be always nonempty for claimants to make an
agreement on how to divide. We show that under a certain condition, the core of a claims
reallocation game is always nonempty if and only if the division rule is a generalized
proportional rule.
J EL Classification: C71, D63, D70.
Key words: Bankruptcy problem; Proportional Rule; No Advantageous Reallocation;
Decentralizability; Claims Reallocation Game; Core.
* The author thanks Professor Takehiko Yamato for his helpful comments.+Department of Social Engineering, Graduate School of Decision Science and Technology,
2-12-1 Oh-okayama, Meguro-ku, Tokyo, 152-8552, J apan; yamamura-h@soc. titech.ac.jp
7/27/2019 dp06-04
3/28
3
1. Introduction
When a firm goes bankrupt, how should its liquidation value be divided among
creditors? When cooperation in a group yields a surplus, how its surplus be divided
among the members of the group? Or how should a government redistribute its citizens
incomes among them? There are many kinds of division problems that should be solved
according to a rule.
Such problems are usually solved according to each agents property, which we call
an agents claim. For example, in a bankruptcy problem, the liquidation value is
divided among creditors according to their credits. A claims problem we are going to
study in this paper is as follows. There are a set of agents and a divisible resource to be
divided among them. Each agent has a claim to this resource. A claims problem is how
this divisible resource should be divided among agents according to their claims.
We should note that claims are transferable in many cases. Then, a division rule to
solve claims problems should be established in consideration for the transferability of
claims. That is, a division rule should be made so as not to cause troubles by any
reallocation of claims. In this paper, we study the class of division rules which are not
troubled by reallocations of claims by the following two approaches: an axiomatic
approach and a game theoretic approach.
In axiomatic studies on claims problems, non-manipulability condition, which
demands that there be no agents who benefit by any reallocation of their claims, is the
most studied property to be satisfied in consideration for the transferability of claims.
This property is studied by ONeill [16] in bankruptcy problems and by Banker [5] in
cost allocation problems. And then, Moulin [12] defined this requirement as an axiom
called No Advantageous Reallocation (in short NAR) and Chun [7] introduced NAR in
claims problems. After them, de Frutos [8], J u [9] and J u=Miyagawa=Sakai [10] have
studied non-manipulability. (See Thomson [18] for a survey.)
A typical example of a non-manipulable division rule is the proportional rule, which is
widely used in modern laws. Chun [7] showed that the proportional rule is the only rule
satisfying NAR and supplementary axioms. J u=Miyagawa=Sakai [10] defined
generalized proportional rules and characterize them based on NAR.
J u=Miyagawa=Sakai [10] also axiomatize the proportional rule based on a weaker
version of NAR, they call pairwise reallocation proofness, which takes only pairwise
reallocations into consideration.
In the axiomatic part of this paper, we introduce other weaker versions of NAR which
7/27/2019 dp06-04
4/28
4
take only k-person reallocations into consideration and study to what extent we can
weaken axioms required to characterize generalized proportional rules. To study this
problem, we reconsider the axiom Monlin [13,14,15] defined asdecentralizability
and
introduce its descendent axioms. These axioms help us to order the relationships among
axioms and to study the axiomatization of generalized proportional rules. Our results
can explain the results of Chun [7] and J u=Miyagawa=Sakai [10] based on
non-manipulability requirements as corollaries.
In the game theoretic part of this paper, we investigate a new coalitional game
theoretic approach that rationalizes generalized proportional rules. In the previous
game theoretic approaches, like in ONeill [16] and Aumann=Maschler [3], a problem
under the nature state, where there is no rule to solve claims problems, has been
described as a coalitional form game. Game theoretic studies have provided the
rationales for several division rules as solutions of this game, such as the random
arrival rule (ONeill [16]) and the Talmudic rule (Aumann= Maschler [3]). However,
they have not given a satisfactory answer to the question why the proportional rule is
actually used. The reason why they have not rationalized the proportional rule may
have something to do with Atsumi [2]s criticism about previous game theoretic
approaches.
Atsumi [2]s criticism is summarized as follows. Game theoretic studies have
analyzed a coalitional form game where the worth of a coalition is defined as what it can
get without going to court. However, the establishment of a division rule causes a
different situation from what they have analyzed, because claimants act with this
division rule taking into consideration. So, the division rules they have conducted
through the analysis of this games may not be conclusive.
In this paper, following Atsumi [2]s criticism, we investigate a new coalitional form
game which describes a bankruptcy problem after a rule is established as an act. We
assume claims are transferable among creditors as in modern laws. Then, each
subgroup of claimants may manipulate the sum of their awards by reallocating their
claims. So, the game we consider in this paper might be called a claims reallocation
game.
We provide the rationale for proportional division rules through the analysis of claims
reallocation games. We require that the core of a claims reallocation game be always
non-empty in order for claimants to make an agreement on how to divide the liquidation
value. We show only proportional rules satisfy this requirement under a certain
condition.
The remainder of this paper is as follows. In section 2, we define a claims problem and
7/27/2019 dp06-04
5/28
5
generalized proportional rules. In section 3, we introduce axioms and consider the
relationships among axioms. In section 4, we axiomatize generalized proportional rules
by axioms introduced in section 3 to consider to what extent we can weaken the axiomsrequired to characterize generalized proportional rules. In section 5, we investigate a
new coalitional form game called a Claims Reallocation Game and characterize
generalized proportional rules through the analysis ofClaims Reallocation Games. In
section 6, we state some concluding remarks.
2. The Model
There is a finite set nN ,,2,1 L of claimants. For each iN, i has a claim
kKkiki
Rcc where kK ,,2,1 L , characterized by a finite dimensional vector.
A claims profile c is a list of individual claims knNii
Rcc . For each SN, let
Sc Si
ic . RE is a divisible value to be divided among N. Then, an n-person claims
problem is formalized by the pair RREc
kn
),( such that
k
N Rc .n
D denotes the set of feasible n-person claims problems. That is,
kNknn RcRREcD |),( .
Letnn DD be the domain of n-person claims problems. The domain nD is rich if
nDEc ),( , then for any nDEc ),( such that NN cc and EE ,nDEc ),( .
The richness of the domain is necessary for reallocations of claims to be taken into
consideration. In this paper, we consider only rich domains.
Given a domainn
D , a division rule over a domainn
D is a function nnn RDf :
which associates with each problemnDEc ),( an award profile n
Ni
n
i REcf ),( .
In this paper, we focus on proportional division rules and its generalizations. Then, we
define what J u=Miyagawa=Sakai [10] calls a generalized proportional rule.
Definition 2.1. A rule nnn RDf : is a generalized proportional rule if there existmappings
nk
RRRA : ,kk
RRRW : such that,
7/27/2019 dp06-04
6/28
6
Kk
Nk
Nk
ikNi
n
i EcWc
cEcAEcf ),(),(),( ,
for each iN and each
n
DEc ),( .
What is called the proportional rule is a special case of a generalized proportional
rule.
Example 2.2. Let k=1. A rule nnn RDf : is the proportional rule if it is ageneralized proportional rule which satisfies NiEcA Ni ,0),( and EEcW N ),( .
On the other hand, the egalitarian rule is one of a generalized proportional rule.
Example 2.3. Let k=1. Define the egalitarian rule nnn RDf : such thatE
nEcfni
1),( , for each iN andeach nDEc ),( .
The egalitarian rule is one of a generalized proportional rule such that EEcW N ),(
and Nin
EEcA Ni ,),( .
3. Axioms
In this chapter, we define a number of axioms which a rulenf should satisfy. First
of all, we introduce one of our main axioms, which is interpreted as a condition for
non-manipulability, called no advantageous reallocation.
No Advantageous Reallocation (NAR)1.For any ),( Ec , nDEc ),( , if E=Eand thereexists NS such that SS cc and SNjcc jj \, , then
Si Sin
i
n
i EcfEcf ),(),( . In other words, for any NS , Sin
if )( is
dependent only on Nc , E, and SNjjc / .
1 J u=Miyagawa=Sakai[10] calls this axiomReallocation Proofness.
7/27/2019 dp06-04
7/28
7
This axiom NAR is introduced by Moulin [12] in quasi-linear bargaining problems. In
the context of claims problem, Chun [7] introduced NAR. NAR requires that no coalition
can benefit by reallocating their claims. Since claims are in principle transferable inmodern law, we should regard NAR as a necessary axiom to avoid manipulation.
NAR deals with any feasible coalition. Then a way to weaken NAR is to deal only with
particular coalitions. Actually, J u=Miyagawa=Sakai [10] introduced an axiom which
takes only two parsons coalitions into consideration2. In this paper, we consider more
generally weaker versions of NAR which take only k-person coalitions into
consideration.
No Advantageous Reallocation for k-person Coalitions (NAR-k). For any ),( Ec ,nDEc ),( , if E=Eand there exists NS , | S| =k such that SS cc and jj cc ,
SNj \ , then Si Sin
i
n
i EcfEcf ),(),( . In other words, for any NS ,
| S| =k, Sin
if )( is dependent only on Nc , E, and SNjjc / .
Next, we introduce another one of our main axiom Moulin [13,14,15] defined as
Decentralizability.
Decentralizabil ity (DEC).For any ),( Ec , nDEc ),( , if NN cc , E=Eand ii cc ,
then ),(),( EcfEcf nin
i . In other words, for any Ni , )(nif is dependent only on
Nc , E, and ic .
There are some reasons we consider DEC again, which seems similar with NAR. First,
we have to note that DEC itself has a property a rulenf should satisfy. Referred by
Moulin[15], only Nc , E and ic are required to compute i s award. Therefore, we can
interpret DEC as a requirement to simplify calculating each agents award.
Moreover, when we consider the relationship between NAR and DEC, we notice that
DEC contrasts strikingly with NAR. NAR demands that no reallocation by any coalition
can benefit the members of this coalition. On the other hand, DEC demands that no
2
J u=Miyagawa=Sakai[10] calls this axiomPairwise Reallocation Proofness,
which isequivalent to NAR-2 defined as follows.
7/27/2019 dp06-04
8/28
8
reallocation by any coalition can affect the awards of agents outside this coalition. I t is
mainly because no coalitional reallocation can damage agents outside this coalition why
we require NAE. Therefore, we can regard DEC as a more essential axiom.However, these are not all of reasons why we reintroduce DEC. We reconsider DEC
because DEC and its descendant axioms help us to order the relationships among
axioms. Moreover, DEC and its descendant axioms are useful to study the possibility (or
impossibility) of the axiomatizations of generalized proportional rules. To study these
problems, the following axiom, we call constantness, is a key axiom.
Constantness (CON).For any nDEcEc ),(),,( , if NN cc and EE , then
Nin
iNi
n
i EcfEcf ),(),( . In other words,
Ni
n
if )( is dependent only on Nc
and E.
The axiom Pareto Optimality, which is usually requested, is a special case of
constantness.
Pareto Optimality (PO).For anyn
DEc ),( , EEcfNin
i ),( .
The axiom PO is usually considered to be satisfied, but PO may not be satisfied when
it costs agents to solve claims problems. For example, mediation fees or trial costs are
required. Then, we can justify CON as Moulin [15] justifies DEC. By CON, only Nc
and E are required to compute the sum of awards to all agents. So, CON can make the
computation more simple, which may make a cost to solve problems cheaper.
Next, we consider descendant axioms of DEC. One way to introduce descendant
axioms of DEC is to introduce coalitional versions of DEC, which deal only with
particular coalitions, as we introduce descendant axioms of NAR. To introduce
coalitional decentralizability, we can consider the following two axioms.
Decentralizabili ty for k-person coalitions (DEC-k).For any NS , | S| =k, and ),( Ec ,nDEc ),( , if NN cc , E=Eand ii cc , Si , then
Si
n
i
Si
n
i EcfEcf ),(),( .
In other words, for any NS , | S| =k,
Si
n
if )( is dependent only on Nc , E, and
7/27/2019 dp06-04
9/28
9
Sii
c .
Strong Decentralizability for k-person coalitions (SDEC-k).For any NS , | S| =k,and ),( Ec , nDEc ),( if NN cc , E=E, and SS cc , then
Si
n
i
Si
n
i EcfEcf ),(),( . In other words, for any NS , | S| =k,
Si
n
if )( is
dependent only on Nc , E, and Sc .
DEC-krequires that for any k-person coalition the sum of their awards be never
affected by reallocation by agents outside this coalition. I f they make a coalition, their
awards are dependent of any reallocation outside them. SDEC-kadditionally requests
that any reallocation among any k-person coalition can benefit the members of this
coalition.
Another way to introduce a descendant axiom of DEC is to deal only with agents with
a particular property. In this paper, we consider a weaker version of DEC, which deals
only nulls, those whose claims are zero.
Decentralizability for Nulls (DECN). For any ),( Ec andn
DEc ),( , if NN cc ,
E=E, and 0 ii cc , then ),(),( EcfEcfn
i
n
i . In other words, for any Ni ,
)(nif is dependent only on Nc , E i f 0ic .
The axiom No Awards for Nulls, which is usually considered, is a special case of
DENC.
No Awards for Nulls (NAN).For any ),( Ec nD , if 0ic , then 0),( Ecfni .
To solve a huge bankruptcy problem, taxes are sometimes used. Then, those who have
no claims must accept negative awards through taxes to them. Or, in the context of
income redistribution problems, citizens are usually assured of a living standard with a
certain level, regardless of their incomes. Then, even if a citizens income is zero, he
must redistribute a positive award. These examples are showing that a null agent may
7/27/2019 dp06-04
10/28
10
not get a zero award in a certain rule. This axiom DECN requires that awards for nulls
be dependent only on Nc and E even if No Award for Null is not satisfied.
These are all descendant axioms of NAR and DEC we study as follows. Next, we studythe relationships among these axioms. First of all, we consider the relationship between
NAR and DEC. The following claim 3.1. and example 3.2. show that DEC is strictly
weaker than NAR. That is, NAR implies DEC but DEC does not always imply NAR.
Claim 3.1. NAR implies CON and DEC.
Proof. First, we are going to proof NAR implies CON. For any ),( Ec , nDEc ),( , if
NN cc and E=E , then by NAR,
Ni
n
i
Ni
n
i EcfEcf ),(),( .
So,nf satisfies CON. Second let us proof NAR implies DEC. For any ),( Ec ,
nDEc ),( , if NN cc , E=E , ii cc and, then
),(),(),(),(),(),(
\\EcfEcfEcfEcfEcfEcf jiNj
n
jNj
n
jiNj
n
jNj
n
j
n
i
by NAR. So, for any Ni , )(
n
if is dependent only on C, E, and ic .
Example 3.2. Let k=1 and consider a rule nf such that NiEc
cEcf
N
in
i
,),(
2
satisfies DEC, but this solution does not satisfy NAR.
However, assuming constrantness, we can easily show that NAR becomes equivalent
to DEC3. This fact also indicates that constantness has an important role to associate
NAR with DEC.
Claim 3.3. Assume CON. Then, DEC is equivalent to NAR.
Proof. By claim 3.1, it is sufficient to prove that NAR implies DEC under CON. For any
),( Ec , nDEc ),( , if NN cc , E=E and there exists NS such that SS cc and
3 Moulin [13] shows this fact under PO.
7/27/2019 dp06-04
11/28
11
jj cc , SNj / , then Ni Nin
i
n
i EcfEcf ),(),( by DEC and
),(),( EcfEcf nin
i , SNj / by DEC. Therefore,
Si SNjn
jNi
n
i
n
i EcfEcfEcf \ ),(),(),(
Ni Sin
iSNj
n
j
n
i EcfEcfEcf ),(),(),( /
Next we consider the relationships among descendant axioms of DEC and NAR. The
following claim3.4, 3.5, and 3.6 shows the relationships.
Claim 3,4. For k
7/27/2019 dp06-04
12/28
12
Si
n
i
SNi
n
i
Ni
n
i
SNi
n
i
Ni
n
i
Si
n
i EcfEcfEcfEcfEcfEcf ),(),(),(),(),(),(//
since | N/S| =n-k. So,
n
f satisfies DEC-k.
Claim 3.6. Assume CON, then for k
7/27/2019 dp06-04
13/28
13
4. Main Results
In this chapter, we characterize generalized proportional rules based on DEC or NAR
or their descendent axioms.
Theorem 4.1. Let n3. A rule satisfies DEC, CON and OB if and only if it is ageneralized proportional rule.
Proof. I t is clear that a generalized proportional rule satisfies DEC, CON and OB. Then,
we prove the converse. Letnf be a rule satisfying DEC, CON and OB, then define a
mappingnk RRRA : and
nknn RRRW : such that,
)0,,(),( EcfEcA Nn
N , ),()0,,(),( EcAEcfEcW NNnn .
By DEC ofnf , A is well defined, and by DEC and CON, OB of nf , nW also
satisfies DEC, CON and OB. Then we replace ),( EcW ni by ),,( iNn
i cEcW for each i
N.
Step 1. I fnf satisfies DEC and CON, then nW is symmetric. That is, for any
nDEc ),( , if ji cc , then ),,(),,( jNn
jiN
n
i cEcWcEcW .
Proof of Step 1. Take any jiNk ,\ , and consider ),( Ec , ),( Ec nD such that
EEE , ccc ji , cccc Nkk , ikNhch ,\,0 , and 0hc ,
jkNh ,\ . Since )(nW satisfies DEC and CON, and 0)0,,( EcW Nn
i , iN,
),,(),,(),,(),,( hNn
h
Ni
iN
n
iiNN
n
iiN
n
i cEcWcEcWcEcWcEcW
),,(),,(),,(),,( jNn
jjN
n
jhN
n
h
Ni
iN
n
i cEcWcEcWcEcWcEcW
Step 2. I fnf satisfies DEC and CON, for each Ni ),,( EcW N
n
i is additive with
7/27/2019 dp06-04
14/28
14
regard to ic . That is, if iii ccc , then
),,(),,(),,( iNn
iiN
n
iiN
n
i cEcWcEcWcEcW .
Proof of Step 2. Take any iNkj \, and consider ),( Ec and nDEc ),( such
that
.,,/,0
)(
0,
,
kjiNhc
ccccc
ccc
ccccc
EE
h
Nkk
jj
ii
Then, ),(),(),,(),,( EcWEcWcEcWcEcW njn
iNn
jNn
i
))(,,(),( cccEcWEcW NNn
k
Ni
n
i
),,(),( ccEcWEcW Nn
i
n
i .
SincenW is symmetric by step 1, ),,(),,( cEcWcEcW NiNj . Therefore,
),,(),,(),,( ccEcWcEcWcEcW Nn
iN
n
iN
n
i .
For each kK, let us define a mapping RRRRW knk : such that,
)0,,0,,0,,0,,(),,( 1 LL kNn
kN
n
k cEcWcEcW .
Then, by step 2,
Kk
kN
n
kN
n cEcWcEcW ),,(),,( 111 .
and for any 1/Ni , by step 1,
Kk
ikN
n
kiN
n
iN
n
i cEcWcEcWcEcW ),,(),,(),,( 1 .
Then, for each kK, let us define a mapping RRREcW kNn
k :),( such that,
),,(),( NkNn
kN
n
k cEcWEcW .
Since, for each kK , ),,( EcW Nn
k is additive with regards to kc and satisfies OB,
7/27/2019 dp06-04
15/28
15
then
),(),,( EcW
c
ccEcW Nk
Nk
ikikN
n
k
by J ensens equation 4 (See Aczel=Dhombres [1].) Therefore, only generalized
proportional rules satisfy DEC, CON and OB.
We moreover investigate the axiomatizations of generalized proportional rules based
on weaker versions of NAR or DEC. The following theorem shows to what extent axioms
required to characterize generalized proportional can be weaken without any other
constraints. To proof this theorem, the idea of coalitional decentralizability is useful.
Theorem 4.2. Let 2kn-3. A rule satisfies DEC-k ( or NAR-(n-k) ), CON and OB if andonly if it is a generalized proportional rule.
Proof. We shownf satisfies DEC if nf satisfies DEC-k, 2kn-3. Then, by theorem
4.1,nf must be a generalized proportional rule.
FornDEc ),( , consider a problem nDEc ),( such that E=E , 11 cc , 02 c ,
323 ccc , ),,4(, nicc ii L . Assuming DEC-(n-k), 3kn-1,nf must
satisfy DEC-(n-3) and DEC-(n-2) by Claim 3.4. Since, 23,2\ nN and
33,2,1\ nN , then by DCE-(n-3) ,DEC-(n-2),
).,(),(),(),(),(),( 13,2,1\
1
3,2\3,2,1\3,2\
1 EcfEcfEcfEcfEcfEcfn
Ni
n
Ni
n
i
Ni
n
i
Ni
n
i
n
Taking the similar steps, fornDEc ),( such that E=E , 11 cc , 0ic ,
)1,,2( ni L and 1ccc Nn , we can say ),(),( 11 EcfEcfnn
. Therefore, for
any ),( Ec , nDEc ),( , if NN cc , E=E and 11 cc , then ),(),( 11 EcfEcfnn .
For any iN, 1i , we can similarly show )(nif is dependent only on Nc and E and
ic .
4 Let X be an interval of R which contains 0 and consider a mapping RXf : whichsatisfies for any x,yX, if x+yX then f(x)+f(y)=f(x+y) and one sided boundedness.
Then, there exists cR such that f(x)=cx, for any xX.
7/27/2019 dp06-04
16/28
16
However, we cannot characterize generalized proportional rule by NAR-2 (or DEC-
(n-2) ) only. The following example shows a rule which satisfies NAR-2 (or DEC-(n-2) ),
CON and OB but is not a generalized proportional rule.
Example 4.3. Let k=1 and n4. Consider a rule nf such thatNieachforE
c
cccEcf
N
iiin
i
,3
),( 11 ,
where n+1=1. This rulenf satisfies CON and NAR-2 ( or DEC-(n-2) ), because
EEc
c
EcfN
Ni
i
Ni
n
i
)113(
),( and
Ec
ccc
Ec
ccccEcfEcf
N
ji
jiNk
k
N
jjiin
j
n
i
11
).(\11
2)(2
),(),(
, if j=i+1,
Ec
ccccccEcfEcf
N
jjjiiin
j
n
i
1111 33),(),(
E
c
ccccc
N
jiNk
kjjii
,\1111 3
, otherwise.
Needless to say, nf satisfies OB. However, this rule nf is not a generalized
proportional rule.
In order to axiomatize generalized proportional rules based on NAR-2 (or DEC-(n-2) ),
an additional axiom is required. The following theorem 4.4. is an axiomatization of
generalized proportional rules based on NAR-2 (or DEC-(n-2) ) with an additional axiom
DECN.
Theorem 4.4. Let n3. A rule satisfies DEC-(n-2) ( or NAR-2 ), DECN, CON and OB ifand only if it is a generalized proportional rule.5
Proof. For SN, define SNjcDEcSD jnn /,0|),()( and a mapping
),( EcA N as in theorem 4.1. By DECN, ),( EcA N is well-defined.
5 This theorem is a generalized rule version of J u=Miyagawa=Sakai [10 Theorem 4].
7/27/2019 dp06-04
17/28
17
Step 1. For any SN, | S| =3, there exists a mappingkkn
S RRRW : such that,
Kk Nkn
S
Nk
ikNi
ni EcW
ccEcAEcf ),(),(),( , for each iN and each )(),( SDEc n .
Proof of Step 1. We can prove this lemma similarly with the proof of Theorem 4.1.
Step 2. For any S,TN, such that | S| =| T| =3,n
T
n
S WW .
Proof of Step 2. When 2 TS , take any kK, i,jST, ij, and nDEc ),( such
that )0,,,0( LL Nki cc , iNj ccc and 0kc , k N/ i,j . Since
)(),( SDEc n and )(),( TDEc n ,
),(),(),(),( EcWEcWEcAEcf Nkn
TNk
n
SNi
n
i .
Therefore,n
T
n
S WW .
When 2 TS , we can take S,TN, | S| =| T| =3, such that SS TS
2 TT . So, nTm
T
n
S
n
S WWWW .
Step 3. If for any UmSS
n SDEc
:
)(),(
Kk
Nk
Nk
ikNi
n
i EcWc
cEcAEcf ),(),(),( , then
for any U1:
)(),(
mSS
n SDEc
Kk
Nk
Nk
ikNi
n
i EcWc
cEcAEcf ),(),(),( .
Proof of step 3. Let (c,E) )(SD n , | S| =k+1, then for any jN, ),(),( EcAEcf Njn
j ,
and for any i,i S,
Kk
Nk
Nk
kiikii
n
i
n
i EcWc
ccAAEcfEcf ),(),(),( by NAR-2.
By these equations, we can show
Kk
Nk
Nk
iki
n
i EcWc
cAEcf ),(),( for any iS.
By step 2 and step 3, for any (c,E) )(NDD
nn
,
7/27/2019 dp06-04
18/28
18
Kk
Nk
Nk
iki
n
i EcWc
cAEcf ),(),( for each iN which accords with a generalized
proportional rule.
Next, we axiomatize generalized proportional rules based on stronger versions of
coalitional decentralizability. Based on SDEC, we can axiomatize generalized
proportional rules without constantness.
Theorem 4.5.(1) Let n5. A rule satisfies SDEC-2 and OB if and only if it is a generalized
proportional rule.
(2) Let n5 and n-2k3. A rule satisfies SDEC-k and OB if and only if it is a
generalized proportional rule.
Proof. We provenf satisfies CON and DEC-k (kn-3) under the condition (1) or (2).
Then,nf must be a generalized proportional rule by theorem 4.2.
Step 1. Let k2 and nk+2. Ifnf satisfies SDCE-k, then nf satisfies CON.
Proof of Step 1. For nDEc ),( , consider a problem nDEc ),( such that E=E ,
01 c , 212 ccc , 2,1\, Nicc ii . Then, by SDEC-k,
n
i
n
i
kSnS Si
n
i
knkS
nS Si
n
i
kn
n
i
n
i EcfEcfk
n
CEcf
k
n
CEcf
3,,32,,323
),(),(21
),(21
),(LL
and,
n
i
n
i
kSS knSi
n
i
nn Ecf
k
n
C
Ecf
kSSNS
EcfEcf32,1 22
21 ),(
2
21),(
,2,1|
1),(),(
).,()(),(
2
21),(
,2,1|
121
32,1 22
EcfEcfEcfk
n
CEcf
kSSNS
nnn
i
n
i
kSS knSi
n
i
Hence,
n
i
n
i
n
i
n
i EcfEcf11
),(),( . Taking the same steps, fornDEc ),( , such that
E =E, )1,,1(,0 nici L and Nn cc , we can show
n
i
n
i
n
i
n
i EcfEcf
11
),(),( .
7/27/2019 dp06-04
19/28
19
Therefore,nf must satisfy CON.
Hence, in the case of (1),
n
f satisfies CON and DEC-k. Therefore,n
f must be ageneralized proportional rule by theorem4.2.
Step 2. Let n5 and k=n-2. I fnf satisfies SDCE-k, then nf satisfies SDEC-2.
Proof of Step 2. FornDEc ),( , consider a problem nDEc ),( such that E =E,
01 c , 212 ccc , 1,,3,0 nici L and )( 21 cccc Nn . Sincenf
satisfies SDEC-k and CON by step 1,
).,(),(),(),(),(),(),(),( 213131
21 EcfEcfEcfEcfEcfEcfEcfEcfnn
n
i
n
i
n
i
n
i
n
i
n
i
n
i
n
i
nn
Hence, )()( 21 nn ff is dependent only on Nc , E and )( 21 cc . For any Nji , ,
ji , we can similarly show )()( njn
i ff is dependent only on Nc , E and )( ji cc .
By step 1 and step 2, nf satisfies CON and DEC-2 in the case of (2). Therefore, nf
must be a generalized proportional rule by theorem4.2.
By, theorem 4.5, we have proved that generalized proportional rules can be
characterized based on SDEC-k if kn-2 with an exception of 4-person problems. In the
last of this section, we show an example of a rule of 4-person claims problems that
satisfies SDEC-2 and OB but is not a generalized proportional rule.
Example 4.6. Let k=1 and n=4. Let ipN
i
cc for each iN. Consider a rule nf such
that,
Epppppp
Ecfn
2
1)()()(),(
2
41
2
31
2
211
,
Epppppp
Ecfn
2
1)()()(),(
2
41
2
31
2
212
,
Epppppp
Ecfn
2
1)()()(),(
2
41
2
31
2
213
, and
7/27/2019 dp06-04
20/28
20
Epppppp
Ecfn
2
1)()()(),(
2
41
2
31
2
214
.
This rule
n
f satisfies SDEC-2 and OB, but is not a generalized proportional rule.
5. Game Theoretic AnalysisIn this section, we introduce a new coalitional form game which describes a
bankruptcy problem after a division is established as a law. Before defining it, we
review the previous coalitional game theoretic analyses of bankruptcy problems. Given
an n-person bankruptcy problem (c,E), ONeill [16] defined a coalitional form game
),( ),( EcvN as a bankruptcy game corresponding to a problem(c,E)6 such that,
0,max)( /),( SNEc cESv , for each SNS , .
Aumann=Maschler [3] explain this characteristic function )(),( Sv Ec indicates what
S can get without going to court. In the nature state where there is no rule to solve
claims problems, it is natural for the members of S to evaluate the worth of S by what S
can get without going to court. So, this game ),( ),( EcvN might be called a bankruptcy
game under the nature state.
However, once a rule is established as an act, the members of S will evaluate the worth
of S by what S can get by going to court, as is mentioned by Atsumi [2]. So, we should
redefine the worth of S after a rulenf is established as a law. In modern laws, claims
are in principle transferable among agents. Then, the members of each coalition may
manipulate the sum of their awards by reallocating their claims. Since we should take
such manipulations into consideration, we call a coalitional form game, which describes
a bankruptcy problem under an established rule, a claims reallocation game.
Applying Aumann=Peleg [3]s idea of coalitional form representation of strategic
games, we evaluate the worth of S by the following two ways.
6 ONeill [16] considered only bankruptcy problems. A bankruptcy problem is defined by
the pair RREcn),( such that EcN and 0Nc . Let
n denote the set of
n-person bankruptcy problems, thenn 0|),( NNn candEcRREc ,
which has a rich domain.
7/27/2019 dp06-04
21/28
21
One way is to evaluate the worth of S by the assurable level. Given a rulenf and a
problem (c,E), an aggregated award RxS is assurable for S , if
Si SNjjSiiniSSNSNSNjjSSSii Eccfxcccccc ),,(,,)(,,)( //// .
Clearlywhen
Si
SNjjSii
n
icccccc
S EccfxSNSNSNjjSSSii
,,minmax)/(|| ///
,
Sx is assurable for S. Then, we define an assurable(or)representation of a claims
reallocation game as follows.
Definiton 5.1.n
f EcvN ),(, is an assurable (or) representation of a claims reallocation
game corresponding to a problems (c,E) under a rulenf , if
Si
SNjjSii
n
icccccc
f
Ec EccfSvSNSNSNjjSSSii
n
,,minmax)()/(||
),(///
, for each SNS , .
Another way is to evaluate the worth of S by the unpreventable level. Given a rule
nf and a problem (c,E), an aggregated award RxS is unpreventable for S , if
SiSNjjSii
n
iSSSSiiSNSNSNjjEccfxcccccc ),,(,,,,)(
////.
Clearlywhen
Si
SNjjSii
n
icccccc
S EccfxSSSiiSNSNSNjj
,,maxmin)/(|| ///
,
Sx is unpreventable for S. Then, we define an unpreventable(or)representation of
a claims reallocation game as follows.
Definiton 5.2. nf EcvN ),(, is an unpreventable (or ) representation of a claimsreallocation game corresponding to a problems (c,E) under a rule
nf , if
Si
SNjjSii
n
icccccc
f
Ec EccfSvSSSiiSNSNSNjj
n
,,maxmin)()/(||
),(///
, for each SNS , .
It is clear that )()( ),(),( SvSvnn f
Ec
f
Ec . This fact indicates that claims reallocation
games are based on the more pessimistic evaluation of the worth of each coalition than
7/27/2019 dp06-04
22/28
22
claims reallocation games.
We should note that claims reallocation games are always superadditive but that
additional assumptions are required for claims reallocation games to satisfysuperadditivity.
Claim 5,3.For any nDEc ),( and nf , nf EcvN ),(, is superadditive.
Proof of Claim 5.3. For each ,,, TSNTS take Siic )( and Tiic
)( such that
Si
SNijSii
n
icccccc
Sii EccfArgcSNSNSNjjSSSii
,,minmax)()/(|| ///
Ti
SNijTii
n
icccccc
Tii EccfArgcTNTNTNjjTTTii
,,minmax)()/(|| ///
.
Then, for each TSNTSNTSNjj ccc //\ , ,
SiTSNjjTiiSii
n
i
SiSNjjSii
n
iccc
EcccfEccfSNSNSNjj
,,,,,min/)/(| ///
TiTSNjjTiiSii
n
i
TiTNijTii
n
iccc
EcccfEccfSNSNSNjj
,,,,,min/)/(| ///
.
Therefore,
)(,,min)()( ),(/|)(),(),( /// TSvEccfTvSv
n
TSNTSNTSNjj
nn f
Ec
TSiTSNjjTSii
n
iccc
f
Ec
f
Ec
,
by definition ofnf
Ecv ),( .
Claim 5.4. nf EcvN ),(, is superadditive if
(1) for any iN,n
if is concave on c.
(2) for any SN,Si
n
if is quasi-concave on Sc and for any iNn
if is continuous on c.
Proof of Claim 5.4. For each ,,, TSNTS take TSNjjc
/)( such that
TSi
TSNijTSii
n
icccccc
TSNjj EccfArgcTSTSTSiiTSNTSNTSNjj
,,maxmin)()/(||
////
And define correspondences SSSiiTTTiiS cccccc ||: and
7/27/2019 dp06-04
23/28
23
TTTiiSSSiiT cccccc ||: such that,
Si
TSNjjTiiSii
n
iccc
TiiSEcccfArgc
SSSii
),,(max)(/|
and
Ti
TSNjjTiiSii
n
iccc
SiiTEcccfArgc
TTTii
),,(max)(/|
.
In both case (1) and case (2), )(S and )(T are non-empty, convex, and upper
hemi-continuous. SSSii ccc | and TTTii ccc | are compact. Then, by
Kakutanis fixed point theorem (Kakutani [11]), there must exist Siic )( and Tiic
)(
such that )()(TiiSSii
cc
and )()(
SiiTTiicc
. Therefore,
Ti
TNjjTii
n
iccc
SiSNjjSii
n
iccc
f
Ec
f
Ec EccfEccfTvSvTTTiiSSSii
nn
,,max,,max)()(/|)(/|)(
),(),(
)(),( ),( TSvEcfnf
Ec
TSiNii
n
i
.
Next, we analyze claims reallocation games by considering their cores. Given a
coalitional form game (N,v), an award vector nNii
Rx is a feasible allocation if
)(NvxN . A feasible allocation Niix is dominated by S if there exists a feasible
allocation Nii
y such that Siyx ii , and )(),( Svynf
EcS . Since )(),( Svynf
EcS ,
the members of S can get at least Sii
y by themselves under the assurable scenario.
So, if a division plan Nii
x , which is dominated by S, is proposed, the members of S
must have an objection to this proposal.
The core of a claims reallocation game is the set of feasible allocations which is not
dominated by any SN. If an allocation which belongs in core is proposed, then
no coalition can improve their awards by themselves in the assurable level. Therefore,
under the assurable scenario, no coalition has an incentive to reject this allocation. In
this sense, an allocation which belongs in core is a stable consequences of the
bargaining among N. Since an claims reallocation games under a rulenf is super
additive, we can define the core ofnf
EcvN ),(, , which denotesnf
EcvC ),( , as
SNSNvxandNvxRxvC nnn f EcSf EcNnf Ec .),()(| ),(),(),( .
We can similarly define the core of a claims reallocation game.
7/27/2019 dp06-04
24/28
24
If there exists a problem (c,E) such that the core ofnf
EcvN ),(, is empty, then for each
feasible allocation, there must exist SN which can improve their awards byreallocating their claims. Then, it becomes difficult to make an agreement on how to
divide a resource under an established rule nf . Therefore, we require the following
axiom to division rules in order for claimants to bargain smoothly how to divide E.
Core NonEmptiness(-CNE). For any nDEc ),( , nf EcvN ),(, has the nonemptycore.
In the context of implementation theory, we can interpret core nonemptiness in
another way. Let us regardnf as a social choice function social planners want to
implement. I f there exists a problem (c,E) such that Nin
i Ecf ),(nf
EcvC ),( , then
there is no guarantee forNi
n
i Ecf ),( to be actually implemented even if this social
choice functionnf is established as a law, since there exists SN the members of
which have a objection to the allocation Nin
i Ecf ),( under a rule
n
f . Then,
Ni
n
i Ecf ),( should be in the core ofnf
EcvN ),(, in order to guarantee that
Ni
n
i Ecf ),( is actually implemented. Core nonemptiness is of course a necessary
condition fornf to be implemented in the core allocation.
Core Implementabil ity (-CI).For any nDEc ),( , Ni
n
i Ecf ),(
nf
EcvC ),( .
I f for each SN the members of S evaluate the worth of S by unpreventable level,
core nonemptiness and implementability can be rewritten as follows.
Core NonEmptiness(-CNE). For any nDEc ),( , nf EcvN ),(, has the nonemptycore.
7/27/2019 dp06-04
25/28
25
Core Implementabil ity (-CI).For any nDEc ),( , Nini Ecf ),(nf
EcvC ),( .
Since core core, we can easily check that -CNE implies -CNE and that
-CI implies -CI.
Next, we characterize generalized proportional rules based on the axioms related to
the core of claims reallocation games. Since -CNE is the weakest among these axioms,
we first study the axiomatization of the proportional rules based on -CNE.
Theorem 5.5. Let n3. A rule satisfies-CNE, CON and OB if and only if it is ageneralized proportional rule.
Proof. To proof this theorem, we make use of theorem 4.1. By theorem 4.1, it is
sufficient to prove that -CNE is equivalent to DEC.
Claim 5.6. Assume CON, then-CNE is equivalent to DEC.Proof. For any SN and (c,E), (c,E )
nD , if NN cc , E= E and SS cc , consider
(c,E
)
n
D such that ii cc
,iS and jj cc
,jN/S. Assuming CON,
n
f
must satisfy NAR. Then,
SNi
n
i
Ni
n
i
Si
n
i
Si
n
i EcfEcfEcfEcf/
),(),(),(),(
Si
n
i
SNi
n
i
Ni
n
i EcfEcfEcf ),()','()','(/
by CON and NAR. Therefore,
Si
n
i
SiSNjjSii
n
i
cccccc
f
EcEcfEccfSv
SNSNSNjjSSSii
n
),(,,minmax)()/(
||
),(///
,
for each SNS , .
So, an allocation Nii
Ecf ),( must be in the core of )(),( Svnf
Ec which implies the core
ofnf
EcvN ),(, is non-empty.
Next we prove the converse. Assumenf does not satisfy DEC. Then there must exist
Ni and nDEcEc ),(),,( satisfying ii cc , NN cc , and EE , such that
7/27/2019 dp06-04
26/28
26
),(),( EcfEcf ii . For each Nj , by CON,
EccfEcfjNv jNkkjnj
ccc
Ni
n
i
f
EcjNkNjNkk
n
,)(,min),())/(( )/(|)(
),()/()/()/(
)/( ),(,),( jNkn
k
n
jNk
n
k EcfEcfEcf .
And for Ni ,
EccfEcfiNv iNkkini
cccNi
n
i
f
EciNiNiNkk
n
,)(,min),())/(( )/(|)(
),()/()/()/(
)/( ),(,),(,),( iNkn
i
n
iNk
n
k
n
iNk
n
k EcfEcfEcfEcfEcf .
Therefore,
)(),(),(1
1))/((
1
1),()/(),(
NvEcfEcfn
kNvn
nn f
EcNj
n
jNk kNj
n
jNk
f
Ec
.
This inequity implies thatnf
EcvN ),(, is not balanced. So the core ofnf
EcvN ),(, must be
empty. (Bondereva [6] or Shapley [17].) Thus, -CNE must imply DEC.
As corollaries of theorem, we can conduct easily show the following statements.
Corollary 5.7. Let n3. A rule satisfies-CI, CON and OB if and only if it is ageneralized proportional rule.Corollary 5.8. Assume n3. A rule satisfies-CNE , CON and OB if and only if it is ageneralized proportional rule. -CNE can be replaced by-CI.
In the last of this chapter, we show two examples which show that -CNE is not
necessarily equivalent to DEC.
Example 5.9. Consider a rule nf such that Ec
cEcf
N
i
Ni
n
),( , for each iN. This
rule does not satisfy DEC. But since En
SSv
n
f
Ec
n
1)(),( for each SN and
En
NNv
n
f
Ec
n
1)(),( ,
E
nE
n
nn1
,,1
L must be in the core of nf EcvN ),(, .
7/27/2019 dp06-04
27/28
27
Example 5.10. Assume n=3. Consider a problem
E
EEE,
3
.
3
.
3
and a rule3f such
that
otherwise
ccifE
fN
ii
02
3 , for each iN.
Of course, this rule3f satisfies DEC. But, the core of
nf
EcvN ),(, must be empty since
ESvnf
Ec )(),( , for any SN,| S|2.
6. Concluding Remarks
In this paper, we focus on four axioms which a division rule should be satisfied: no
advantageous reallocation, decentralizability, core nonemptiness and core
implementability. Though these four axioms are derived from different requirements,
they are mathematically equivalent to one another if a division rule satisfies
constantness. This fact makes it possible to characterize generalized proportional rules
in different ways.
References
[1] J . Aczel, J . Dhombres, Functional Equations in Several Variables: With Application
to Mathematics, Information Theory and he Natural and Social Sciences, Cambridge
University Press, Cambridge, 1989.
[2] H. Atsumi, Interpretation of Talmud Bankruptcy Problem, (in J apanese) Sougou
Seisaku K enkyu1 (1996) 151-159.
[3] R. Aumann, M. Maschler, Game Theoretic Analysis of a Bankruptcy Problem from
the Talmud,J ournal of Economic Theory36 (1985) 195-213.
[4] R. Aumann, B. Peleg, Von Noumann-Morgenstern Solution to Cooperative Games
without Sidepayments, Bulletin of the American Mathematical Society 66 (1960)
173-179.
[5] R. D. Banker, Equity Consideration in Traditional Full Cost Allocation Practices: An
7/27/2019 dp06-04
28/28
Axiomatic Perspective, in: S. Moriarity (Ed.), J oint Cost Allocations, Center for
Economic and Management Reserch, Norman, 1981.
[6] O. N. Bondereva, Some Applications of L inear Programming Methods to the Theoryof Cooperative Games, (in Russian) Problemy K ibernetiki10 (1963) 119-139.
[7] Y. Chun, The Proportional Solution for Rights Problems, Mathematical Social
Sciences15 (1988) 231-246.
[8] M. de Frutos, Coalitional Manipulation in a Bankruptcy Problem, Review of
Economic Design4 (1999) 255-272.
[9] B. G. J u, Manipulation via Merging and Splitting in Claims Problems, Review of
Economic Design8 (2003) 205-215.
[10] B. G. J u, E. Miyagawa, T. Sakai Non-Manipulable Division Rule in Claims
Problems and Generalizations, forthcoming inJ ournal of Economic Theory, 2005.
[11] S. Kakutani, A Generalization of Browers Fixed Point Theorem, Duke
Mathematical J ournal8 (1941) 457-458.
[12] H. Moulin, Egalitarianism and Utilitarianism in Quasi-Linear Bargaining,
Econometrica53 (1985) 49-67.
[13] H. Moulin, Binary Choices with Compensations, mimeo., University of California
San Diego, 1985.
[14] H. Moulin, Equal or Proportional Division of a Surplus, and Other Methods,
International J ournal of Game Theory16 (1987) 161-186.
[15] H. Moulin, Axioms of Cooperative Decision Making, Cambridge University Press,
Cambridge, 1988.
[16] B. ONeill, A Problem of Rights Arbitration from the Talmud, Mathematical Social
Sciences2 (1982) 345-371.
[17] L. S. Shapley, On Balanced Sets and Cores,Naval Research Logistics Quarterly14
(1967) 453-460.
[18] W. Thomson, Axiomatic and Strategic Analysis of Bankruptcy and Taxation
Problems: A Survey, Mathematical Social Sciences45 (2003) 249-297.