www.elsevier.com / locate / econbase
Single-peaked preferences, endowments and
population-monotonicity
*
Bernardo Moreno
´ ´ ´ ´
Departamento de Teorıa e Historia Economica, Universidad de Malaga, Campus El Ejido s /n, 29071 Malaga, Spain
Received 5 May 1999; received in revised form 4 September 2001; accepted 24 September 2001
Abstract
We consider economies with a finite set of agents with single-peaked preferences and with private endowments of an infinitely divisible commodity. We look for solutions that distribute fairly across agents the gains made possible by redistributions of the commodity under the consideration of changes in the number of agents. We show that the uniform rule is the only selection from the weak no-envy and Pareto efficient solution satisfying one-sided population-monotonicity. 2002 Elsevier Science B.V. All rights reserved.
Keywords: Endowments; Population-monotonicity; Uniform rule
JEL classification: D6
1. Introduction
We consider economies with a finite set of agents with single-peaked preferences and with private endowments of an infinitely divisible commodity. We look for solutions that distribute fairly across agents the gains made possible by redistributions of the commodity. The model in which there are no private endowments but there is a social endowment to be allocated has been analyzed, among others,
`
by Sprumont (1991) and Thomson (1994a,b, 1995a). Barbera et al. (1997) characterize the class of rules that are strategy-proof and Pareto efficient and satisfy an individual monotonicity property. Thomson (1995b) presents economies in which there are private and social endowments in a unified framework of open economies. The model in which there are private endowments is also studied in Klaus et al. (1997, 1998). Among other desirable properties, they focus on envy-freeness and
population-monotonicity. Klaus et al. (1997) use an envy-freeness condition in which an agent
compares either the net allotment change of another agent or the feasible part of the agent’s allotment
*Tel.: 134-95-213-1247; fax:134-95-213-1299. E-mail address: [email protected] (B. Moreno).
0165-1765 / 02 / $ – see front matter 2002 Elsevier Science B.V. All rights reserved. P I I : S 0 1 6 5 - 1 7 6 5 ( 0 1 ) 0 0 5 7 6 - 6
change. We use a weaker envy-freeness condition in which an agent only compares redistributions of other agents which are feasible for him (Klaus et al. (1997) refer to this condition as weak no-envy).
1
We show that the uniform rule is the only selection from the weak no-envy and Pareto efficient solution satisfying one-sided population-monotonicity (Theorem 1 below). Klaus et al. (1997) characterize the uniform rule by means of Pareto efficiency, envy-freeness, and one-sided
population-monotonicity using the fact that the uniform rule is the only selection from the envy-free and Pareto efficient solution which satisfies endowment-peak-only. However, it can be shown by means of an
example that it is not possible to use weak envy-freeness in the characterization of the uniform rule as
2
the only weak no-envy and peaks-only rule .
This paper proceeds as follows. In Section 2, we introduce the model and in Section 3, we present the results.
2. The model
We follow Thomson (1995a) and Moreno (1996). There is an infinite population of ‘potential
˜
agents’, indexed by the positive integers, :. Let N denote the class of non-empty finite subsets of :. Each agent i[: is equipped with a continuous preference relation R defined over R and an
i 1
endowment v [R . Let P denote the strict preference relation associated with R , and I the
i 1 i i i
indifference relation. These preference relations are single-peaked: for each R , there is a number,i
9
9
9
9
denoted by p(R )[R , such that for all z ,z [R if z ,z <p(R ), or p(R )<z ,z , then z P z .
i 1 i i 1 i i i i i i i i i
The preference relation R can be described in terms of the function r :R <h`j→R <h`jdefined
i i 1 1
as follows: given zi<p(R ), r (z )i i i >p(R ) and z I r (z ) if such a number exists, and r (z )i i i i i i i 5 ` otherwise; given zi>p(R ), r (z )i i i <p(R ) and z I r (z ) if such a number exists, and r (z )i i i i i i i 50 otherwise. Let r (`)5lim r (z ). LetR be the class of single-peaked preference relations onR . For
i z→` i i 1
i
N
˜
all N[N, let R denote the cartesian product of N copies ofu u R, indexed by the members of N.
N
Similarly, let R denote the cartesian product of N copies ofu u R . We write R 5(R ) ,
1 1 N i i[N
N N ˜
p(R )5( p(R )) , and v 5(v) . An economy is a pair e5(R ,v )[R 3R . For all N[N,
N i i[N N i i[N N N 1
N
let j denote the set of admissible economies.
N N
˜
For all N[N and all e5(R ,v )[j , let Z(e)5h(z ) [R :o z 5o vj denote the set of
N N i i[N 1 N i N i
˜
feasible allocations of e. A solution is a mapping w which associates with every N[N and
N
ˆ
e5(R ,NvN)[j , a non-empty subset of Z(e). The intersection of twow solutions and w is denoted
ˆ ww.
N
˜
For all N[N and all e5(R ,NvN)[j , the allocation z[Z(e) is Pareto efficient for e if there is no
ˆ ˆ ˆ
z[Z(e) with z R z for all ii i i [N, and z P z for some ii i i [N. Let P(e) be the set of these allocations.
N
˜
For all N[N and all e5(R ,NvN)[j , the allocation z[Z(e) is individually rational for e if z Ri ivi
for all i[N. Let I(e) be the set of these allocations. The allocation z[Z(e) is envy-free in final consumptions for e if for all i, j[N, z R z (Foley, 1967).i i j
1
Moreno (1996) shows that there is no population-monotonic selection from the weak no-envy and Pareto efficient solution. However, he provides a solution, the proportional-sacrifice solution, that it is a selection from the individually rational and Pareto efficient solution satisfying population-monotonicity under a minor additional restriction on preferences.
2
Note that the intersection of the sets of envy-free and individually rational allocations may be empty. However, we can ask how to distribute fairly across agents the gains made possible by redistributions. We will therefore be looking for notions of equitable redistributions.
N
˜
Definition 1. For all N[N and e5(R ,NvN)[j , the allocation z[Z(e) is obtained through an
3
envy-free redistribution if for no i, j[N, such that v 1(v 2z )[R , we have v 1(v 2z )P z .
i j j 1 i j j i i
Let F(e) be the set of allocations obtained through envy-free redistributions.
N
˜
For all N[N and e5(R ,NvN)[j , the set of Pareto efficient allocations obtained through envy-free
4
redistributions is non-empty. Indeed, the uniform allocation, introduced next, always exists, and
satisfies weak no-envy and Pareto efficiency.
N
˜
Definition 2. For all N[N and e5(R ,NvN)[j , the allocation z[Z(e) is the uniform allocation of e if there is l[R such that (i) when o p(R )>o v, then for all i[N, z 5minhl1v, p(R )j,
1 N i N i i i i
and (ii) when oNp(R )i <oNvi then for all i[N, zi5maxhvi2l, p(R )i j.
Let U(e) denote the uniform allocation of e.
3. One-sided population-monotonicity
We now consider changes in the number of agents. We require the arrival of new agents to affect all agents present before the change in the same direction. A general version of this requirement was proposed and studied by Thomson (1983a,b). Chun (1986) considered a quasi-linear model of cost allocation and proposed the condition that all agents be affected in the same direction by the arrival of additional agents.
However, population-monotonicity is a strong requirement when combined with weak no-envy and
Pareto efficiency. In fact, there is no selection from the weak no-envy and Pareto efficient solution
satisfying population-monotonicity. The next step is to weaken the population-monotonicity con-dition. The following requirement is an adaptation of Thomson (1995a) to our context.
Definition 3. A solution is one-sided population-monotonic if for all N,N9[: with N9,N, and all
N N9
e5(R ,NvN)[j and e9 5(R ,N9vN9)[j if either (i)oNp(R )i <oNvi andoN9p(R )i <oN9vi or (ii)
oNp(R )i >oNvi and oN9p(R )i >oN9vi, then for all i[N9, wi(e)Riwi(e9), or for all i[N9,
wi(e9)Riwi(e).
When we impose one-sided population-monotonicity on selections from the no-envy and Pareto
efficient solution, the unique solution that remains acceptable is the uniform rule, as we show in the
5
next theorem .
3
The concept of envy-freeness in terms of allotment changes, called fair net trade, was formulated by Kolm (1972) and Schmeidler and Vind (1972) in the more general context of exchange economies.
4
In the sequel, when we say weak no-envy allocations, we refer to allocations obtained through envy-free redistributions.
5
Theorem 1. The uniform rule is the only selection from the weak no-envy and Pareto efficient
solution satisfying one-sided population-monotonicity.
Proof. It is clear that the uniform rule is weak envy-free and Pareto efficient and satisfies one-sided
population-monotonicity. Now we show that the uniform rule is the only one satisfying the above
properties. We will make the proof in two steps.
N
˜
Step 1. Let N[N and e5(R ,v )[j be such that o p(R )>o v. Let w7FP be a one-sided
N N N i N i
population-monotonic solution. We prove thatw(e)5U(e). Let z5w(e). Since w7P this implies that for all i[N, zi<p(R ). The proof follows from the four claims below:i
Claim 1.1. (Fig. 1) For all j[N such that vj>p(R ), zj j5p(R ). Suppose not. Then, for some jj [N
such that vj>p(R ), zj j,p(R ). By feasibility and efficiency, for some kj [N, vk,zk<p(R ). Sincek ˜
w#F, z Rj jvj1(zk2vk). Let N9[N be a set of agents such that uN9u5u uN and N>N9 5 [. Let
b:N→N9 be a bijection. Let Rb(i )5R for all ii [N\hjj, let Rb( j )[R be such that p(Rb( j ))5p(R ),j
and vb( j )1(zi2vi)Pb( j ) jz , and let vb(i )5vi, for all i[N. Let N0 5N<N9. Let e0 5(RN99,vN0)[
N0
j . Let z0 5w(e0). Note that oNp(R )i 1oN9p(Rb(i ))>oNvi1oNvb(i ), and sincew#P, for all i[N
99
99
zi <p(R ), and zi b(i )<p(Rb(i )). Since w#F, vb(i )5vi for all i[N\hjj, and p(Rb(i ))5p(R ) for alli
99
99
i[N\hjj, we have zb(i )5zi for all i[N\hjj. Sincew#FP,vb( j )5vj, and p(Rb( j ))5p(R ), we havej
99
99
99
99
99
z 5z . Since w#F, z 5z , and v 1(z 2v)P z , we have z ±z . Thus, in the change
b( j ) j b( j ) j b( j ) i i b( j ) j b( j ) j
99
99
from e to e0, if zb( j ).z (zj b( j ),z ) agent j gains (loses) and at least one ij [N\hjj loses (gains) in violation of one-sided population-monotonicity.
Claim 1.2. (Fig. 1) For all k[N such that vk<p(R ), zk k>vk. Suppose not. Then, for some k[N
such thatvk<p(R ), zk k,vk. By feasibility and efficiency, for some i[N\hkj, vi,zi<p(R ). Sincei
w#F, z Rk kvk1(zi2vi). Note that this is possible only if vk1(zi2vi)>r (z ). Let us nowk k
introduce an additional agent, agent n11. Let vk5vn11 and let Rn11[R be such that p(R )k 5
N<hn11j p(Rn11) and rn11(vn11).vn111(zi2vi)1(vk2z ). Let ek 9 5(R ,RN n11,vN,vn11)[j be the
9
9
resulting economy and z9 5w(e9). Sincew#FP,vk5vn11 and p(R )k 5p(Rn11), we have zk5zn11.
9
9
9
9
Note that zj 5z for all jj [N\hij, zn115z and zk i 5zi1(vn112zn11) is a feasible allocation. Since
w#F and rn11(vn11).vn111(zi2vi)1(vk2z ), the above allocation is not obtained through ank
9
envy-free redistribution. Indeed, if vn11.zn11.z , agent k gains and by one-sided population-k
9
9
monotonicity, all agents gain. Therefore, zi >z . By feasibility, zi i ,zi1(vn112z ) and sincek
rn11(vn11).vn111(zi2vi)1(vk2z ), agent nk 11 envies the redistribution of agent i, in violation
9
9
ofw#F. Thus, in the change from e to e9, if zn11>vn11 (zn11,z ) agent k gains (loses) and at leastk
one i[N\hkj loses (gains), in violation of one-sided population-monotonicity.
Claim 1.3. For all i, j[N such that vi<p(R ),i vj<p(R ) and p(R )j i 2vi>p(R )j 2vj, we have
zi2vi>zj2vj. Suppose not. Then, for some i, j[N such that vi<p(R ),i vj<p(R ), and p(R )j i 2
vi>p(R )j 2vj, we have zi2vi,zj2vj. Sincew#P, it is clear that vi1(zj2vj)P z , in violationi i
of w#F.
Claim 1.4. (Fig. 2) For all i, j[N such that vi<p(R ),i vj<p(R ) and p(R )j i 2vi>p(R )j 2vj, we have zi2vi.zj2vj if and only if zj5p(R ). Suppose not. Then, for some i, jj [N such that vi<p(R ),i vj<p(R ), and p(R )j i 2vi>p(R )j 2vj, we have zi2vi.zj2vj and zj,p(R ). Sincej
w#F, z Rj jvj1(zi2vi). Note that this is possible only ifvj1(zi2vi)>r (z ). Let us now introducej j
two additional agents, agents n11 and n12. Let vj5vn11 and let Rn11[R be such that
p(R )j 5p(Rn11), vn111(zi2vi)Pn11 jz . Let Rn12[R and vn12 be such that vn122p(Rn12)5zj2
N<hn11j<hn12j
vj. Let e9 5(R ,RN n11,Rn12,vN,vn11,vn12)[j be the resulting economy. Note that
9
oNp(R )i 1p(Rn11)1p(Rn12)>oNvi1vn111vn12 and let z9 5w(e9). Since w#P, we have zi <9
9
p(R ) for all ii [N, zn11<p(Rn11), and zn12<p(Rn12). Sincew#F, vj5vn11 and p(R )j 5p(Rn11),
9
9
9
9
9
we have zj 5zn11. By Claim 1.1, we have zn125p(Rn12). Note that zi5z for all ii [N, zn115zj
9
9
9
and zn125p(Rn12) is a feasible allocation. Since w#F, zn115z , andj vn111(zi2vi)Pn11 jz , we
9
9
9
have z ±z . Thus, in the change from e to e9, if z .z (z ,z ) agent j gains (loses) and at
n11 j n11 j n11 j
least one i[N\hjj loses (gains), in violation of one-sided population-monotonicity.
N
˜
Step 2. Let N[N and e5(R ,NvN)[j be such that oNp(R )i <oNvi. Let w#FP be a one-sided population-monotonic solution. We prove thatw(e)5U(e). Let z5w(e). Since w#P this implies that
for all i[N, zi>p(R ). The proof follows from the four claims below:i
Fig. 3. Characterization of the uniform rule on the basis of one-sided population-monotonicity (Theorem 1, Claim 2.1).
Claim 2.1. (Fig. 3) For all j[N such that vj<p(R ), zj j5p(R ). Suppose not. Then, for some jj [N
such that vj<p(R ), zj j.p(R ). By feasibility and efficiency, for some kj [N, vk.zk>p(R ). Sincek ˜
w#F, for all v 1(z 2v )[R , z Rv 1(z 2v ). Let N9[N be a set of agents such that
j k k 1 j j j k k
uN9u5u uN and N>N9 5 [. Let b:N→N9 be a bijection. Let Rb(i )5R andi vb(i )5vi for all i[N\hjj, and let Rb( j )[R andvb( j ) be such that p(Rb( j ))2vb( j )5p(R )j 2vj, andvb( j )1(zi2vi)Pb( j )vb( j )1
N0
(zj2vj). Let N0 5N<N9. Let e0 5(R ,N0vN0)[j . Let z0 5w(e0). Note that oNp(R )i 1
99
99
oN9p(Rb(i ))<oNvi1oNvb(i ), and since w#P, for all i[N zi >p(R ), and zi b(i )>p(Rb(i )). Since
99
99
w#F, vb(i )5vi for all i[N\hjj, and p(Rb(i ))5p(R ) for all ii [N\hjj, we have zb(i )5zi for all
99
99
i[N\hjj. Since w#FP, and p(Rb( j ))2vb( j )5p(R )j 2vj, we have zb( j )2vb( j )5zj 2vj. Since
99
99
99
w#F, z 2v 5z 2v, and v 1(z 2v)P v 1(z 2v), we have z 2v ±z 2
b( j ) b( j ) j j b( j ) i i b( j ) b( j ) j j b( j ) b( j ) j
99
99
vj. Thus, in the change from e to e0, if zb( j )2vb( j ),zj2vj (zb( j )2vb( j ).zj2vj) agent j gains (loses) and at least one i[N\hjj loses (gains) in violation of one-sided population-monotonicity.
Claim 2.2. (Fig. 4) For all k[N such that vk>p(R ), zk k<vk. Suppose not. Then, for some k[N
such thatvk>p(R ), zk k.vk. By feasibility and efficiency, for some i[N\hkj, vi.zi>p(R ). Sincei w#F, for allv 1(z 2v)[R , z R v 1(z 2v). Let us now introduce an additional agent, agent
j i i 1 k k k i i
n11. Let Rn11[R and vn11 be such that vn112p(Rn11)5vk2p(R ) and rk n11(vn11),vn111
N<hn11j
(zi2vi)1(vk2z ). Let ek 9 5(R ,RN n11,vN,vn11)[j be the resulting economy and z9 5w(e9).
9
9
Since w#FP, and vn112p(Rn11)5vk2p(R ), we have zk k2p(R )k 5zn112p(Rn11). Note that
9
9
9
9
zj 5z for all jj [N\hij, zn112vn115zk2vk and zi 5zi1(vn112zn11) is a feasible allocation. Since w#F and rn11(vn11),vn111(zi2vi)1(vk2z ), the above allocation is not obtainedk
9
through an envy-free redistribution. Indeed, if vn11,zn11,vn111(zk2vk), agent k gains and by
9
9
one-sided population-monotonicity, all agents gain. Therefore, zi <z . By feasibility, zi i .zi1
9
(vn112zn11) and since rn11(vn11).vn111(zi2vi)1(vk2z ), agent nk 11 envies the
redistribu-9
tion of agent i, in violation ofw#F. Thus, in the change from e to e9, if zn11<vn11 (zn11.vn111 (zk2vk)) agent k gains (loses) and at least one i[N\hkj loses (gains), in violation of one-sided
population-monotonicity.
Claim 2.3. For all i, j[N such that vi>p(R ),i vj>p(R ) andj vi2p(R )i >vj2p(R ), we havej vi2zi>vj2z . Suppose not. Then, for some i, jj [N such that vi>p(R ),i vj>p(R ), andj vi2
p(R )i >vj2p(R ), we havej vi2zi,vj2z . Sincej w#P, it is clear that vi1(zj2vj)P z , ini i violation of w#F.
Claim 2.4. (Fig. 5) For all i, j[N such that vi>p(R ),i vj>p(R ) andj vi2p(R )i >vj2p(R ), wej
have vi2zi.vj2z if and only if zj j5p(R ). Suppose not. Then, for some i, jj [N such that vi>p(R ),i vj>p(R ), andj vi2p(R )i >vj2p(R ), we havej vi2zi.vj2z and zj j.p(R ). Sincej w#F, for all v 1(z 2v)[R , z Rv 1(z 2v). Let us now introduce two additional agents,
j i i 1 j j j i i
agents n11 and n12. Let Rn11[R and vn11 be such that vn112p(Rn11)5vj2p(R ), andj
vn111(zi2vi)Pn11vn111(zj2vj). Let Rn12[R andvn12 be such that p(Rn12)2vn125vj2z .j
N<hn11j<hn12j
Let e9 5(R ,RN n11,Rn12,vN,vn11,vn12)[j be the resulting economy. Note that
9
oNp(R )i 1p(Rn11)1p(Rn12)>oNvi1vn111vn12 and let z9 5w(e9). Since w#P, we have zi >9
9
p(R ) for all ii [N, zn11>p(Rn11), and zn12>p(Rn12). Since w#F, and vn112p(Rn11)5vj2
9
9
9
9
p(R ), we havej vj2zj 5vn112zn11. By Claim 2.1, we have zn125p(Rn12). Note that zi 5z for alli
9
9
9
i[N, vn112zn115vj2z and zj n125p(Rn12) is a feasible allocation. Since w#F, vj2zj 5
9
9
vn112zn11, and vn111(zi2vi)Pn11vn111(zj2vj), we have vn112zn115vj2z . Thus, in thej
9
9
change from e to e9, if vn112zn11.vj2z (vj n112zn11,vj2z ) agent j gains (loses) and at leastj
one i[N\hjj loses (gains), in violation of one-sided population-monotonicity. h
Note that here, in contrast to the problem of fair allocation (Thomson, 1995a), we have not used
replication-invariance to characterize the uniform rule as the only one-sided population-monotonic
selection from the weak no-envy and Pareto efficient solution. By Theorem 1, since the uniform rule is replication-invariant, it is clear that if a subsolution of the weak no-envy and Pareto efficient solution satisfies one-sided population-monotonicity, then it is replication-invariant. In the problem of fair division (Thomson, 1995a), the counterpart of the above statement is not true.
Acknowledgements
´ ´ ´
I would like to thank Pablo Amoros, Salvador Barbera, Carmen Bevia, Luis Corchon, Carmen ´ ¨
Herrero, Cristina Mazon, Jorg Naeve and William Thomson for their very helpful comments. I thank the referee and the associate editor, whose comments led to an improvement of the paper. Financial
´
support from the Instituto Valenciano de Investigaciones Economicas is gratefully acknowledged. All remaining errors are mine.
References
`
Barbera, S., Jackson, M., Neme, A., 1997. Strategy-proof allotment rules. Games and Economic Behavior 18, 1–21. Chun, Y., 1986. The solidarity axiom for quasi-linear social choice problems. Social Choice and Welfare 3, 297–320. Foley, D.K., 1967. Resource allocation and the public sector. Yale Economic Essays 7, 45–98.
Klaus, B., 1998. Fair allocation and reallocation: an axiomatic study. Ph.D. dissertation University of Maastricht, The Netherlands.
Klaus, B., Peters, H., Storcken, T., 1997. Reallocation of an infinitely divisible good. Economic Theory 10, 305–333. Klaus, B., Peters, H., Storcken, T., 1998. Strategy-proof division with single-peaked preferences and initial endowments.
Social Choice and Welfare 15, 297–311. ´
Kolm, S.C., 1972. Justice et Equite. CNRS, Paris.
Moreno, B., 1996. The uniform rule in economies with single-peaked preferences, endowments and population monotonicity. WP-AD-96-10. Working paper, IVIE, Valencia.
Schmeidler, D., Vind, K., 1972. Fair net trades. Econometrica 40 (4), 637–642.
Sprumont, Y., 1991. The division problem with single-peaked preferences: a characterization of the uniform allocation rule. Econometrica 59, 509–519.
Thomson, W., 1983a. The fair division of a fixed supply among a growing population. Mathematics of Operations Research 8, 319–326.
Thomson, W., 1983b. Problems of fair division and the egalitarian solution. Journal of Economic Theory 31, 211–227. Thomson, W., 1994a. Resource-monotonic solutions to the problem of fair division when preferences are single-peaked.
Social Choice and Welfare 11, 205–223.
Thomson, W., 1994b. Consistent solutions to the problem of fair division when preferences are single-peaked. Journal of Economic Theory 3, 219–245.
Thomson, W., 1995a. Population-monotonic solutions to the problem of fair division when preferences are single-peaked. Economic Theory 5, 229–246.
Thomson, W., 1995b. Endowment-monotonicity in economies with single-peaked preferences. Working paper, University of Rochester, revised May 1996.