Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Bargaining with ordinal and cardinal players¤
Emilio Calvoy Hans Petersz
June 2004 (This version)
Corresponding author: Hans Peters, Department of Quantitative Eco-
nomics, University of Maastricht, 6200 MD Maastricht, Netherlands. E-
mail address: [email protected]. Telephone: +31-43-3883288. Fax:
+31-43-3884874.
¤We thank the associate editor and two referees for their elaborate comments, which
led to a substantial revision of this paper.yDepartamento de Analisis Economico, Campus dels Tarongers, Avinguda dels
Tarongers s/n, Edi¯cio Departamental Oriental, 46011 Valencia, Spain. E-mail address:
[email protected]. Grants from the University of the Basque Country (UPV/EHU
00031.321-HA7924/2000) and the Spanish Ministry of Science and Technology (BEC2000-
0875 and (BEC2000-1429) are gratefully acknowledged.zDepartment of Quantitative Economics, University of Maastricht, 6200 MD Maas-
tricht, Netherlands. E-mail address: [email protected].
1
Abstract
We consider bargaining problems with at least one cardinal player and with
ordinal players, and provide a complete description of utility invariant solu-
tions of such problems for two players. For the n-person case we provide a
procedure that: (i) returns a given cardinal solution if there are only cardinal
players; (ii) is based on the ordinal solution for gradual bargaining problems,
introduced by O'Neill et al. (2001), for the ordinal players. Finally, we in-
troduce the so-called cardinal concession solution as another example of a
utility invariant solution.
Keywords: Bargaining, cardinal invariance, ordinal invariance, gradual bar-
gaining
2
1 Introduction
We consider n-person pure bargaining problems, that is, situations where the
only possible outcomes involve either complete cooperation of all players or
complete breakdown of cooperation.
Most of the game theoretic literature on bargaining assumes that play-
ers are cardinal. Speci¯cally, it is assumed that they have von Neumann-
Morgenstern preferences over outcomes: their preferences are represented by
expected utility functions, which are unique up to positive a±ne transfor-
mations. Therefore, it is natural to impose the requirement that bargaining
solutions be invariant under such transformations. The most prominent so-
lutions in the literature (Nash, 1950; Kalai and Smorodinsky, 1975; Perles
and Maschler, 1981) satisfy this property of cardinal invariance.
In contrast, traditional economic theory considers bargaining problems
in which the players are ordinal, for instance the exchange of commodities
between consumers. In that case, a bargaining solution should be invari-
ant under all monotonically increasing transformations of the players' utility
functions. This property is called ordinal invariance.
A remarkable consequence of the fact that bargaining theory abstracts
from the physical environment and considers only the image in utility space,
is the impossibility, observed by Shapley (1969), of ordinally invariant two-
person bargaining solutions that are nondictatorial and do not assign the
disagreement point. This impossibility result, however, is no longer true if
there are more than two bargainers. Shubik (1982, pp. 90{98) describes a
three-person ordinally invariant solution, attributed to Shapley. Sprumont
(2000) uses the associated construction to characterize ordinally equivalent
bargaining problems by deriving a complete set of representative problems.
3
Basically, this characterization solves the question of describing all ordinally
invariant bargaining solutions. Kibris (2001) gives an axiomatic characteriza-
tion of the solution in Shubik (1982), based on the construction in Sprumont's
paper. See also Safra and Samet (2001, 2004) for extensions.
The assumption that bargaining theory should only use the information
available in the feasible set of utility tuples is sometimes called welfarism.
Dropping this assumption makes it much easier to ¯nd natural ordinally
invariant solutions, as recently demonstrated in Nicolµo and Perea (2000), but
at the same time makes the theory less general. Also in the present paper we
relax the welfarist assumption: we assume that at least one player is cardinal
and the other players are ordinal. (Obviously, if the names of the ordinal and
cardinal players are ¯xed and known, we are back in the welfarist framework.)
We show by an example of a bargaining problem between a manager who is
paid in risky stocks and a worker who is paid in riskless money, that such a
situation can make a lot of sense.
A bargaining solution will be called utility invariant if it is ordinally
invariant for the ordinal players and cardinally invariant for the cardinal
players. We give a complete description of utility invariant solutions for
bargaining problems with one cardinal and one ordinal player, also involving
additional conditions like individual rationality or Pareto optimality. This
is easy since all such problems are equivalent under cardinal-ordinal utility
transformations.
There are many ways to extend these two-person solutions to n-person
problems with at least one cardinal player. Here, we describe one such pro-
cedure. This procedure returns a given cardinal solution (e.g., Nash, Kalai-
Smorodinsky,...) if there are only cardinal players. If there are also ordinal
4
players, we proceed, roughly, as follows. The cardinal players receive a ¯xed
percentage of the gains associated with the cardinal solution payo®s they
would receive if the ordinal players would receive their disagreement payo®s.
The line segment connecting the disagreement payo®s for the cardinal players
and their cardinal solution payo®s determines a gradual bargaining problem
for the ordinal players in the sense of O'Neill et al. (2002). Then, we apply
the ordinal solution proposed by these authors to determine the payo®s for
the ordinal players.
Finally, we present an example of a solution, the so-called cardinal con-
cession solution, that arises in a somewhat di®erent way. It is de¯ned by
considering integral curves of a speci¯c vector ¯eld, coincides with the Perles-
Maschler solution (for two players) and an extension thereof (for n players)
if there are only cardinal players, and with the solution produced by the
above procedure if there is exactly one cardinal player. In the other mixed
cases, however, it treats cardinal players in a more sophisticated way. We
include this example to illustrate that there are many ways, apart from the
procedure above, to obtain reasonable utility invariant solutions.
The organization of the paper is as follows. In Section 2 we present a
two-person manager-worker problem as an example of a situation with a
cardinal and an ordinal player. After the preliminaries in Section 3, we treat
the two-person case in Section 4 and the extension to the n-person case in
Section 5. Section 6 is concerned with the cardinal concession solution, and
Section 7 concludes.
5
2 A two-person example
Suppose that the owner of a ¯rm wants to distribute a reward between a
worker (player 1) and a manager (player 2). The total amount to distribute
is one monetary unit. The payo®s for the worker are in money (m units),
and for the manager in stocks of the ¯rm (s units). Denoting the stock price
by p, we thus have m+ ps · 1.We assume that the worker's preferences are ordinal, represented by a
continuous and strictly increasing utility function u1 for money. We also
assume that the manager must wait until the next period to sell the stocks,
and that there is uncertainty about the future price: it will be either p1
or p2, each with probability one half, and the manager has expected utility
of the form Eu2(s) =12u2(p1s) +
12u2(p2s), where u2 is again a continuous
strictly increasing function. The disagreement outcome corresponds to the
worker and the manager both receiving zero, and the disagreement vector is
d = (u1(0); u2(0)). The bargaining problem is summarized by the pair (S; d),
where S is the feasible utility set
S = fx 2 IR2 : x1 = u1(m); x2 = Eu2(s); m+ ps · 1g:
A bargaining solution is a function à that assigns to a bargaining problem
(S; d) a feasible point Ã(S; d) 2 S: In our example a reasonable solution
should satisfy Ã(S; d) ¸ d (individual rationality) and Ã(S; d) 2 @S (Paretooptimality), where @S denotes the boundary of S. Let I(S; d) be the `ideal
point', as indicated in Figure 1. (All concepts will be introduced more for-
mally in the next section.) For every real number r 2 [0; 1], de¯ne the solution'r such that 'r(S; d) 2 @S and 'r2(S; d) = rI2(S; d) + (1 ¡ r)d2 (see Figure1). It is straightforward to check that this solution is utility invariant: more
6
I(S; d)
d
u'r(S; d)
'r1(S; d)
'r2(S; d) = rI2(S; d) + (1¡ r)d2
Card
Ord
Figure 1: A two-person example
precisely, if u1 is transformed by a strictly increasing continuous function
(player 1 is ordinal) and u2 by a positive a±ne transformation (player 2 is
cardinal) then to obtain the solution of the transformed bargaining problem
we can just apply the two transformations to the point 'r(S; d). In other
words, the underlying agreement in terms of money for the worker and stocks
for the manager has not changed.
In Section 4 we show that any individually rational, Pareto optimal, and
utility invariant solution must be equal to 'r for some r 2 [0; 1]. Hence,where there is an abundance of solutions if the two players are cardinal,
and only two dictatorial solutions (namely, '0 and '1) if the two players
are ordinal (cf. Shapley, 1969), there is basically a one-dimensional family of
solutions in the mixed case.
7
3 Preliminaries
Let N = f1; : : : ; ng (with n ¸ 2) be the set of players, and let ; 6= C µ N .Members of C are called cardinal and members of NnC are called ordinal.
A bargaining problem (for N) is a pair (S; d) where:
(i) the feasible set S µ IRN is closed and comprehensive, i.e., if x 2 S andy 2 IRN with x ¸ y then y 2 S;1
(ii) S is unbounded in every coordinate;
(iii) the set fx 2 S : x ¸ yg is bounded for every y 2 S;
(iv) the boundary @S of S is non-level , that is, if x 2 @S and y 2 IRN withy ¸ x and y 6= x, then y =2 S;
(v) the disagreement point d is an element of int(S) (the interior of S).
The interpretation is that the players either agree on some point x in S,
yielding utility xi to player i, or disagree, in which case each player i ends
up with utility di.
Observe that there is no convexity condition on the feasible set: convexity
does not make sense if one or more of the players are ordinal, since it is not
preserved under arbitrary monotonic transformations.2
We denote by BN the family of bargaining problems (S; d) for N .For (S; d) 2 BN and every i 2 N we de¯ne
Ii(S; d) := maxfyi : y 2 S and yj = dj for all j 2 Nnfigg:
The point I(S; d) := (Ii(S; d))i2N is called the ideal point of (S; d).
8
A bargaining solution à is a map à : BN ! IRN such that for every
(S; d) 2 BN , Ã(S; d) 2 S.An order-preserving transformation is a strictly increasing bijection on
IR. Such a function is necessarily continuous, and its inverse is again an
order-preserving transformation. Let f = ffi : i 2 Ng be a set of order-preserving transformations. For x 2 IRN and (S; d) 2 BN denote f(x) =
(f1(x1); : : : ; fn(xn)), and f(S) = ff(x) : x 2 Sg. Observe that (f(S); f(d)) 2BN .A subclass of order-preserving transformations is the class of the positive
a±ne transformations of the form ¿ 7! b+ a¿ for all ¿ 2 IR, where a; b 2 IR,a > 0.
We say that a bargaining solution à is utility invariant if for every set
f of order-preserving transformations such that fi is positive a±ne for all
i 2 C, we have Ã(f(S); f(d)) = f(Ã(S; d)) for all (S; d) 2 B.A bargaining solution à is Pareto optimal if Ã(S; d) 2 @S for every
(S; d) 2 BN , and individually rational if Ã(S; d) ¸ d for every (S; d) 2 BN .
4 One cardinal and one ordinal player
Let N = f1; 2g and C = f2g, so player 1 is ordinal and player 2 is cardinal.We start with the following observation.
Lemma 4.1 Let (S; d); (T; e) 2 BN . Then there is a pair of order-preservingtransformations f = (f1; f2), with f2 positive a±ne, such that (T; e) =
(f(S); f(d)).
Proof. Let g2 be a positive a±ne transformation such that e2 = g2(d2) and
I2(T; e) = g2(I2(S; d)). Next, for every ¿ 2 IR let h(¿) 2 IR be de¯ned by
9
(¿; h(¿)) 2 @S. Then h : IR ! IR is a strictly decreasing function. Also, for
every ¿ 2 IR de¯ne j(¿) by (j(¿ ); ¿ ) 2 @T . Again, j : IR ! IR is a strictly
decreasing function. Now the desired transformations are f1 := j ± g2 ±h andf2 := g2. 2
Fix an arbitrary bargaining problem ( ¹S; ¹d) 2 BN , and let x 2 ¹S be an
arbitrary point. By Lemma 4.1, there is a unique utility invariant bargaining
solution that assigns x to ( ¹S; ¹d). Denote this solution by Ãx, then we have
just derived:
Corollary 4.2 A bargaining solution à : BN ! IRN is utility invariant if
and only if there is an x 2 ¹S with à = Ãx.
Obviously, the family of all utility invariant Pareto optimal [individually
rational] bargaining solutions is obtained by restricting x to @ ¹S [to fx 2¹S : x ¸ dg]. A nicer way to describe the intersection of these families is asfollows. For every r 2 [0; 1] let 'r be the bargaining solution with for every(S; d) 2 BN : 'r(S; d) 2 @S and 'r2(S; d) = rI2(S; d) + (1¡ r)d2.
Corollary 4.3 A solution à on BN is utility invariant, Pareto optimal, andindividually rational if, and only if, à = 'r for some r 2 [0; 1].
Proof. Obviously, every 'r satis¯es the three conditions in the corollary.
For a solution à satisfying these conditions, by Pareto optimality and in-
dividual rationality there is an r 2 [0; 1] such that Ã( ¹S; ¹d) = 'r( ¹S; ¹d). By
utility invariance and Lemma 4.1, Ã = 'r. 2
Corollary 4.3 provides a simple but complete description of all utility in-
variant, Pareto optimal, and individually rational bargaining solutions. If
10
N = C, then of course any of the well-known two-person bargaining solutions
(Nash, Kalai-Smorodinsky, Perles-Maschler,...) satis¯es these conditions. In
the next section we provide an extension to more than two players.
5 More than two players
In this section we extend the family of bargaining solutions described in
Corollary 4.3 to more than two players. There are many ways to do this but
here we concentrate on a solution (or rather, class of solutions) that satis-
¯es the following requirements: (i) if there are only cardinal players, then
the solution should coincide with a given `cardinal' solution; (ii) the cardinal
players receive a ¯xed fraction of the gain associated with the cardinal solu-
tion outcome that would result if the ordinal players would be kept down to
their disagreement payo®s; (iii) subject to the constraint imposed by (ii), the
ordinal players receive payo®s according to the ordinal solution of O'Neill et
al. (2002). Implicit in this procedure is that the obtained solution should be
utility invariant, individually rational, and Pareto optimal.
Concerning the ¯rst requirement, let C be any set of cardinal players, let
DC µ BC , and let ' : DC ! IRC be a utility invariant, individually ratio-
nal, Pareto optimal, and symmetric solution. (Symmetry means that the
players should receive equal payo®s in a symmetric bargaining problem, i.e.,
a bargaining problem that is invariant under any permutation of the coor-
dinates.) For brevity, we call such a solution classical. For example, DCcould be the set of all bargaining problems with (cardinal) player set C and
with convex feasible sets, and ' could be the jCj-person3 Nash bargainingsolution, assigning to each (S; d) 2 DC the unique maximizer of the product
11
Qi2C(xi ¡ di) over the set fx 2 S : x ¸ dg.4
In order to proceed with requirements (ii) and (iii), we need to be explicit
about the ordinal solution of O'Neill et al. (2002). Let O be a (¯nite) set
of (ordinal) players (possibly, jOj = 1). An agenda is a function ® : IRO !IR that is continuously di®erentiable, with (strictly) positive gradient that
satis¯es a local Lipschitz continuity condition. (See O'Neill et al. (2002) for
a discussion of these and other assumptions, and for more details about the
ordinal solution.) If ® is an agenda and d 2 IRO, then the pair (®; d) is calleda gradual bargaining problem. Note that in a gradual bargaining problem
there is a sequence of increasing feasible sets (fx 2 IRO : ®(x) · tg)t2IR. Theordinal solution associates with each gradual bargaining problem (®; d) a
(continuously di®erentiable) function (`path') ! : IR ! IRO satisfying the
system of di®erential equations
!0i(t) = [jOj®0i(!(t))]¡1 ; i 2 O; (1)
(where ®0i denotes a partial derivative) with initial condition
!(®(d)) = d:
For every gradual bargaining problem this path is unique. Denote this path
by ! = (®; d). Then, ®(!(t)) = t for every t 2 IR (Pareto optimality),
!(t) ¸ d for every t 2 IR with t ¸ ®(d) (individual rationality), and
is ordinally invariant: if f = (fi)i2O is a collection of order-preserving
transformations5, then for every gradual bargaining problem (®; d), we have
(®; d) = f((¯; e)), where d = f(e) and ¯ = ® ± f . In spite of its technicalde¯nition the ordinal solution has a simple interpretation: the ratio of player
i's and j's marginal increments of utility at `time' t, !0i(t)=!0j(t), is equal to
12
the marginal rate of substitution of i's and j's utilities along the boundary
fx 2 IRO : ®(x) = tg of the feasible set at time t.
Before we continue we need some notation. For a vector y 2 IRN and a
proper subset M µ N , let yM denote the restriction of y to the coordinates
in M . For S µ IRN and ¹x 2 IRNnM , let
SM;¹x = fy 2 IRM : there is an x 2 S with y = xM and xNnM = ¹xg;
the `slice' of S for coalition M where the complement NnM receives ¹x.
Let now N be a player set, with ; 6= C µ N the subset of cardinal players,
and O = NnC the subset of ordinal players. Let ' : DC ! IRC be a classical
solution for some DC µ BC . We will de¯ne a utility invariant, individuallyrational, and Pareto optimal solution Ã'; on a subset B0N of BN . We de¯nethis subset of bargaining problems along the way. If N = C, then Ã'; := ';
hence, for this case we de¯ne B0N = DN . Otherwise, let (S; d) be a bargainingproblem for N and assume that (SC;dO ; dC) 2 DC (hence, this is a restrictionon BN), so that '(SC;dO ; dC) is well-de¯ned. Let zC := '(SC;dO ; dC). Hence,zC is the outcome for the cardinal players if the ordinal players would receive
their disagreement payo®s. For every t 2 IR, de¯ne the point xC(t) 2 IRC byxC(t) := (1¡ t)zC + tdC . Under appropriate conditions on BN (to guaranteesmoothness and local Lipschitz continuity|so this is another restriction on
BN ) the points xC(t) generate a gradual bargaining problem with feasible
sets SO;xC(t) (t 2 IR) and disagreement point dO. Let the ordinal solution forthis gradual bargaining problem be given by the path !. Let t¤ := jOj=jN jand de¯ne
Ã';(S; d) := (xC(t¤); !(t¤)):
13
Hence, the cardinal players receive the fraction jCj=jN j of the gain zC ¡dC , which they would obtain if the ordinal players were kept down to their
disagreement utilities dO. The number t¤ plays a similar role as the number r
in Corollary 4.3. The choice for t¤ proposed here guarantees that the solution
is symmetric at least on the bargaining problem (fx 2 IRN : Pi2N xi · 1g; 0).Given that the cardinal players receive xC(t
¤), the ordinal players play the
induced gradual bargaining problem and receive utilities according to the
ordinal solution .
The solution Ã'; proposed here is utility invariant (obvious for the cardinal
players, and for the ordinal players this follows from ordinal invariance of
the ordinal solution6), Pareto optimal, and individually rational. It can be
varied by choosing di®erent classical solutions ', or other solutions instead
of for the gradual bargaining problem.
If there is exactly one cardinal player, say player n, then according to
Ã'; this player receives (1=n)(In(S; d) ¡ dn) (this is independendent of theclassical solution ' since for only one player all these solutions coincide). The
gradual bargaining problem for O = Nnfng then consists of all slices of Swith ¯xed utility for player n. (See Figure 2 in Section 6 for an illustration.)
Thus, in this gradual bargaining problem the feasible sets are endogenously
determined by the utility levels of the cardinal player. In this sense, adding
one cardinal player results in an `endogenization' of the feasible sets in a
gradual bargaining problem.
14
6 The cardinal concession solution
The cardinal concession (cc) solution is constructed in a di®erent way than
the solution(s) in the previous section. If there are only cardinal players, then
the solution coincides with the Perles-Maschler solution (for two players) or
an extension thereof (for more than two players). If there is exactly one
cardinal player, then the solution coincides again with Ã'; (for any classical
solution '). In the remaining mixed cases, however, the cardinal players are
treated in a more sophisticated way. Thus, this shows that the procedure for
¯nding utility invariant solutions, discussed in the preceding section, certainly
does not exhaust the possibilities.
The cc-solution will be de¯ned on the class BNs of bargaining problems, char-acterized as follows. A bargaining problem (S; d) 2 BN is in BNs if and only ifthere is a function g : IRN ! IR with continuously di®erentiable and positive
partial derivatives, such that @S = fx 2 IRN : g(x) = 0g. The i-th partialderivative of g is denoted by g0i.
Let (S; d) 2 BNs . For a point x 2 S and j 2 N , let xj := (xNnfjg; Ij(S; x)).(In this notation we suppress dependence of the point xj on the set S.) Hence,
xj is the best point for j if the other players i are kept to xi. We de¯ne the
vector ¯eld v on the interior of S, int(S), as follows:
vS(x)i :=
264Yj2C
24Yk2C
g0k(xj)
g0i(xj)
35 1jCj375
1jCj
; for all i 2 N , x 2 int(S). (2)
The vector ¯eld v can be interpreted as assigning to an interior point x a
kind of geometric mean of the exchange rates of utilities at associated points
xj on the boundary of S.
15
We proceed by showing that through each point of the interior of S there
is a unique integral curve of v. First, we construct a new vector ¯eld ¹v on
all of IRN , as follows. Extend the de¯nition of the `ideal point' I(S; x) to all
of IRN (we can take exactly the same de¯nition as in Section 3, noting that
the feasible set S is assumed unbounded in every coordinate), and de¯ne the
distance function DS(x) :=Pi2N [Ii(S; x) ¡ xi] for all x 2 IRN . Obviously,
DS(x) > 0 if x 2 int(S), DS(x) = 0 if x 2 @S, and DS(x) < 0 if x =2 S.Then let
¹vS(x) :=
8>>>><>>>>:DS(x)vS(x) if x 2 int(S),0 if x 2 @S,DS(x)vS(I(S; x)) if x =2 S.
Lemma 6.1 Let (S; d) 2 BNs , let x0 2 IRN , and let T be a real interval with0 2 int(T ). Then there is a unique di®erentiable function »(¢; x0) : T ! IRN
with »(0; x0) = x0 and with (d=dt)»(t; x0) = ¹vS(»(t; x0)) for all t 2 T .
Proof. This lemma follows from standard results in the theory of ordinary
di®erential equations. See, for example, Chapter 8 of Hirsch and Smale(1974).
It is su±cient that the map x 7! ¹vS(x) is locally Lipschitz continuous on IRN .
If g is the function associated with S, then by assumption, the partial deriva-
tives g0i(xj) never vanish and have continuous partial derivatives. Therefore,
¹vS(x) has continuous ¯rst-order partial derivatives, and this implies Lipschitz
continuity. 2
It is now straightforward to de¯ne the cardinal concession solution · : BNs !IRN . Let (S; d) 2 BNs and let »(¢; d) be the unique integral curve of the vector¯eld ¹vS passing through d. Note that (the graph of) this curve is strictly
increasing, since the partial derivatives of the function g describing @S are
16
positive. This curve must cross the boundary @S at some point »(t¤; d): if
not, then it would converge to a point z 2 int(S), but then it can be uniquelyextended by the integral curve of ¹v passing through z, a contradiction. Note
that »(¢; d) is also an (hence, the) integral curve of the vector ¯eld vS onthe interior of S, since on int(S) these vector ¯elds do not di®er in direction.
De¯ne ·(S; d) := »(t¤; d). The solution · de¯ned this way is obviously Pareto
optimal and individually rational. Further, ·(S; e) = ·(S; d) whenever e is
a point on the integral curve »(¢; d). Another direct consequence of Lemma6.1 is that di®erent curves never intersect and, in particular, never cross @S
at the same point.
We show now that · is utility invariant with respect to those order-
preserving transformations that map bargaining problems in BNs to problemsin the same set. Therefore, we restrict attention to transformations f =
ffi : i 2 Ng such that each fi is twice continuously di®erentiable, withpositive ¯rst derivative everywhere. Of course, positive a±ne transformations
have these properties. In the remainder of this section utility invariance is
understood to be invariance with respect to transformations of this kind.
The following lemma gives the simple relation between the partial deriv-
atives associated with a bargaining problem and the transformed problem.
Lemma 6.2 Let (S; d) 2 BNs and let f be a collection of order-preserving
transformations. Let g and h describe @S and @f(S), respectively. Then, for
all i 2 N and x 2 @S,h0i(f(x)) =
g0i(x)f 0i(xi)
:
Proof. Note that we can write h = g ± f¡1 (where f¡1 = ff¡1i : i 2 Ng),hence for i 2 N and x 2 @S, we have h0i(f(x)) = (g ± f¡1)0i(f(x)) = g0i(x) ¢(f¡1i )
0(fi(x)) = g0i(x)=f0i(xi). 2
17
Theorem 6.3 · is utility invariant.
Proof. In order to show that · is utility invariant, we have to show that the
integral curves of the vector ¯eld vS are utility invariant. Let (S; d) 2 BNs ,let f = ffi : i 2 Ng be a set of order preserving transformations, and let»(¢; d) and »f (¢; f(d)) be the integral curves of vS passing through d and ofvf(S) passing through f(d), respectively. Then it is su±cient to show that
d
dtfi(»i(t; d)) = ¹(f(d))
d
dt»fi (t; f(d)) for all i 2 N; (3)
where ¹ : int(f(S)) ! IR is some positive and continuously di®erentiable
function (this function may change the speed through the integral curve, but
does not change its direction). Note that for every i 2 N ,d
dtfi(»i(t; d)) = f
0i(»i(t; d)) ¢
d
dt»i(t; d) = f
0i(»i(t; d)) ¢ vS(»(t; d))i: (4)
Denoting for any two vectors x; y 2 IRN , x ¤ y = (x1y1; : : : ; xnyn), and
f 0(x) = (f 01(x1); : : : ; f0n(xn)), (3) can be rewritten as
f 0(d) ¤ vS(d) =
Ãd
dtfi(»i(t; d))
!i2N
= ¹(f(d))
Ãd
dt»fi (t; f(d))
!i2N
= ¹(f(d)) ¢ vf(S)(f(d)); (5)
where the ¯rst equality follows from (4). Hence, we are left to show that (5)
holds.
First note that the fi are positive a±ne for all i 2 C, say fi(xi) = bi+aixi.Recall that dji = di if i 6= j, and dii = Ii(S; d), for all i. Hence, f 0i(dji ) = f 0i(di)for all i 6= j. Moreover, if i 2 C, f 0i(Ii(S; d)) = ai = f 0i(di). By this and
Lemma 6.2 we have h0i(f(dj)) = g0i(d
j)=f 0i(di) (where, as in Lemma 6.2, g
18
describes @S and h describes @f(S)) if i 6= j or if i = j and i 2 C. Hence,for all i 2 N ,
vf(S)[f(d)]i =
264Yj2C
24Yk2C
h0k(f(dj))
h0i(f(dj))
35 1jCj375
1jCj
=
264Yj2C
24Yk2C
g0k(dj)=ak
g0i(dj)=f 0i(di)
35 1jCj375
1jCj
=f 0i(di)
[Qk2C ak]
1jCj¢ vS(d)i:
So [Qk2C ak]
1jCj ¢ vf(S)[f(d)] = f 0(d) ¤ vS(d), and (5) holds. 2
For the case of exactly one cardinal player the solution · coincides with the
solution Ã'; for any classical solution '. See Figure 2.
Corollary 6.4 Let jCj = 1 and let ' be a classical solution. Then ·(S; d) =Ã';(S; d) for every (S; d) 2 BNs .Proof. For (S; d) 2 BNs , let »(¢; d) be the integral curve of vS throughd. Then ·(S; d) is the point of intersection of the curve with @S. We may
assume d = »(0; d) and ·(S; d) = »(1; d). Let 1 be the cardinal player and
let O = Nnf1g. For 0 · t · 1, the slices SO;»1(t;d) form (part of) a gradual
bargaining problem, and projected on these sets the curve f»(t; d) : 0 · t ·1g determines a curve f»¡1(t; d) : (»2(t; d); : : : ; »n(t; d)) : 0 · t · 1g, wherefor each t 2 [0; 1], the point »¡1(t; d) is on the boundary of the feasible set attime t. Let g be the function describing @S, then by de¯nition of the vector
¯eld v in (2) it follows that
d
dt»i(t; d) ¢ g0i(»(t; d) =
d
dt»j(t; d)) ¢ g0j(»(t; d)) for all i; j 2 N:
But by (1), this implies that the curve f»¡1(t; d) : 0 · t · 1g is exactly thecurve that determines the ordinal solution the gradual bargaining problem
for O.
19
¡¡¡¡¡¡¡¡¡¡¡¡
¡¡
¡¡
¡¡
¡¡
¡¡¡
¡¡
¡¡
¡¡
¡¡
¡¡ u uuu
P
P
d = »(0; d)
-»¡1(t; d)
-»(t; d)
¡¡ª
·(S; d) = »(1; d)
¾ @SO;»1(1;d)¾ @SO;»1(t;d)
Ord
Ord
Card (player 1)
*µ
¸
u·(S; d)
d
S
S
»
Ord
Card
Figure 2: The upper diagram illustrates the proof of Corollary 6.4. The lower
diagram illustrates the case n = 2 with one cardinal player: the graph of » is
obtained by rotating the part of @S above ·(S; d), so ·(S; d) = 'r(S; d) with
r = 1=2.
20
We are left to show that ·1(S; d) = d1 +1n[I1(S; d) ¡ d1]. Consider the
paths » and P as in Figure 2. We ¯rst construct a set of order preserving
transformations f such that the transformed path »f is a straight line.
Let f1 be a positive a±ne transformation such that f1(d1) = 0 and
f1 (I1(S; d)) = 1. For all i 6= 1, de¯ne fi such that fi (»i(t; d)) = f1 (»1(t; d)).Because » is a di®erentiable function, the corresponding fi are twice contin-
uously di®erentiable, with positive ¯rst derivatives everywhere. De¯ned in
this way, and because · is utility invariant, the integral curve of vf(S) through
0 is a straight line passing through f1 (·1(S; d)) ¢ (1; :::; 1) . Hence
vf(S)³»f (t; 0)
´i=d
dt»fi (t; 0) =
d
dt»fj (t; 0) = vf(S)
³»f(t; 0)
´j; (6)
for all i; j 2 N .Now we show that the corresponding P f is also a straight line. Denote by
h the function that describes @f(S). To simplify notation, denote by pf (t)
the corresponding point in P f such that pfi (t) = »fi (t; 0), for each t 2 [0; 1]
and i 6= 1. Recall that, by de¯nition of ·, we have
vf(S)³»fi (t; 0)
´i=h01 (p(t))h0i (p(t))
; (7)
for all i 2 N . Then (6) and (7) imply h0i (p(t)) = h0j (p(t)) for all i; j 2 N . Thefact that the normal vector to @f(S) in every pf(t) has all its components
equal and positive implies that P f is a straight line through (1; 0; :::; 0) and
f1 (·1(S; d)) ¢ (1; :::; 1). Moreover, P f is included in the hyperplane H = fx 2IRN :
Pi2N xi = 1g, which ¯nally implies that f1 (·1(S; d)) = 1
n, and thus
·1(S; d) = d1 +1n[I1(S; d)¡ d1]. 2
The lower diagram in Figure 2 illustrates the two-player case with one cardi-
nal player. In this case, the graph of the curve » is obtained by rotating the
21
upper part of @S along the horizontal line through '1=2(S; d) (cf. Section 4),
as can be inferred directly from (2).
We conclude this section with three further remarks on the cc-solution.
Remark 6.5 (Two players, cardinal) If there are only two players, both
cardinal, then the cc-solution · coincides with the Perles-Maschler (1981)
solution. This can be seen as follows. Denoting by ³ the Perles-Maschler
solution, the di®erential equation for ³ using our notation is given by
³S(x)1 :=
"g02(x
2)
g01(x2)
#12
; ³S(x)2 :=
"g01(x
1)
g02(x1)
#12
; for all x 2 int(S); (8)
where, as before, g is the function describing @S. (See the de¯nition of
the solution by Procedure 2 in Perles and Maschler (1981), or Section 2 in
Calvo and Guti¶errez (1994).) Applying de¯nition (2) to the case in which
C = N = f1; 2g, vS reduces to
vS(x)1 =
24"g02(x1)g01(x1)
# 12
¢"g02(x
2)
g01(x2)
# 12
3512
; vS(x)2 =
24"g01(x1)g02(x1)
# 12
¢"g01(x
2)
g02(x2)
# 12
3512
;
for all x 2 int(S). Therefore, it is straightforward to check that
vS(x)2vS(x)1
=³S(x)2³S(x)1
=
"g01(x
1)
g02(x1)¢ g
01(x
2)
g02(x2)
#12
:
Hence, both vector ¯elds, vS and ³S, are collineal and therefore produce the
same integral curves, so · and ³ coincide.
Remark 6.6 (More than two players, cardinal) In Perles and Maschler
(1981) the super-additive solution was supported by a set of axioms: ef-
¯ciency, symmetry, continuity, cardinal invariance and superadditivity. In
22
Perles (1982) it is proved that for three-person problems a solution that sat-
is¯es e±ciency, cardinal invariance, symmetry and superadditivity does not
exist. In Calvo and Guti¶errez (1994) it was shown that (8) implies a prop-
erty that can be used to give a suitable extension of this solution to n-person
problems. It can be shown that this extension is a solution di®erent from ·.
Thus, · restricted to cardinal problems can be considered as an alternative
extension of the super-additive solution.
Remark 6.7 (General case) In general, according to the solution ·, car-
dinal players no longer receive a ¯xed proportion of the gains they would
obtain if the ordinal players were kept to their disagreement payo®s, as was
the case for the solutions studied in the preceding sections. This can be seen
by considering (2) for a bargaining problem with two cardinal players and
one ordinal player. The details are left to the reader.
7 Conclusion
The purpose of this paper was twofold: to extend the bargaining problem
by allowing both cardinal and ordinal players, and two ¯nd utility invariant
solutions for it. The procedure presented in Section 5 can be `axiomatized'
on a `meta-level' by the following choices: choose a classical solution for the
cardinal players; choose the percentage of the gains they would receive if
the ordinal players were held down to their disagreement payo®s; and choose
a solution for the gradual bargaining problem. The ordinal solution to the
gradual bargaining problem is characterized by O'Neill et al. (2001). The
de¯nition of the cc-solution was inspired by the de¯nition of a (purely) ordinal
solution by Shapley (1995).
23
References
Calvo, E., Guti¶errez, E., 1994. Extension of the Perles-Maschler solution
to N-person bargaining games. International Journal of Game Theory
23, 325{346.
Hirsch, M.W., Smale, S., 1974. Di®erential equations, dynamical systems,
and linear algebra. Academic Press, San Diego.
Kalai, E., Smorodinsky, M., 1975. Other solutions to Nash's bargaining
problem. Econometrica 43, 513{518.
Kibris, ÄO, 2001. Preference-based egalitarianism in bargaining. Mimeo,
CORE, University Louvain-la-Neuve, Belgium.
Nash, J.F., 1950. The bargaining problem. Econometrica 18, 155{162.
Nicolµo, A., Perea, A., 2000. A non-welfarist solution for two-person bar-
gaining situations. Mimeo, Universidad Carlos III de Madrid.
O'Neill, B., Samet, D., Wiener, Z., Winter, E., 2001. Bargaining with an
agenda. Forthcoming in Games and Economic Behavior.
Perles, M.A., 1982. Non-existence of super-additivity solution for the Nash
bargaining game. International Journal of Game Theory 10, 151{161.
Perles, M.A., Maschler, M., 1981. The super-additive solution for the Nash
bargaining game. International Journal of Game Theory 10, 163{193.
Safra, Z., Samet, D., 2001. Ordinal solutions to bargaining problems.
Mimeo.
24
Safra, Z., Samet, D., 2004. An ordinal solution to bargaining problems with
many players. Games and Economic Behavior 46, 129{142.
Shapley, L., 1969. Utility comparison and the theory of games, in: G.Th.
Guilbaud (ed.), La D¶ecision. Editions du CNRS, Paris.
Shapley, L., 1995. Oral communication at Center for Rationality and Inter-
active Decision Theory, The Hebrew University, Jerusalem.
Shubik, M., 1982. Game Theory in the Social Sciences: Concepts and
Solutions. MIT University Press, Cambridge Mass.
Sprumont, Y., 2000. A note on ordinally equivalent Pareto Surfaces. Jour-
nal of Mathematical Economics 34, 27{38.
25
Footnotes
1 x · y means xi · yi and x < y means xi < yi for every i. Analogously for¸ and >.
2 In Sections 5 and 6 we impose a few additional conditions on bargaining
problems.
3 For a ¯nite set X, jXj denotes its cardinality.
4 If jCj = 1, then a bargaining problem is de¯ned as a pair ((¡1; a]; d),where a; d 2 IR with a > d.
5 Here, only those oder-preserving transformations are allowed that map
gradual bargaining problems into gradual bargaining problems.
6 It is assumed that only those order-preserving transformations are con-
sidered such that the resulting bargaining problems are again in the class
B0N .
26