You are on page 1of 15

For the Abandonment of Symmetry in Game Theory

T. C. Schelling

The Review of Economics and Statistics, Vol. 41, No. 3. (Aug., 1959), pp. 213-224.

Stable URL:

The Review of Economics and Statistics is currently published by The MIT Press.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact
Fri Apr 4 06:16:57 2008
T h e Review of Economics and Statistics



T. C. Schelling

T H E first part of this paper argues that the

pure "moveless" bargaining game analyzed
by Nash, Harsanyi, Luce and Raiff a, and others
monly these operations are sketched in by ref-
erence to the notion of "binding agreements"
and the notion of free communication in the
may not exist or, if it does, is of a different char-
process of reaching agreement. Thus, to say that
acter from what has been generally supposed; two players may divide $100as soon as they can
the point of departure for this argument is the agree on how to divide it, and that they may
operational meaning of "agreement," a concept discuss the matter fully with each other, is
that is almost invariably left undefined. The generally considered sufficient to define a game.2
second part of the paper argues that symmetry A game of this sort is symmetrical in its move
in the solution of bargaining games cannot be structure, even though it may be asymmetrical
supported on the notion of "rational expecta- in the configuration of payoffs. The two players
tions"; the point of departure for this argument have identical privileges of communication, of
is the operational identification of irrational refusing offers, and of reaching agreement. If
expectations. instead of dividing $100the players are to agree
on values X and Y contained within a boundary,
the payoff function may not be symmetrical but
the move structure is. Harsanyi, to emphasize
The logical structure of a zero-sum game is
this, has even added explicitly the postulate of
completely defined by the game's payoff matrix
symmetrical moves : "The bargaining parties
or function; and, if one is attracted to the mini-
follow identical (symmetric) rules of behavior
max solution, the payoff matrix contains every-
(whether because they follow the same prin-
thing significant in the game. I n the case of the
ciples of rational behaviour or because they are
tacit (non-cooperative) non-zero-sum game,
though not all of the essentials of the game are subject to the same psychological laws) ." (Op.
cit., 149.)
necessarily contained in the payoff matrix, the
What I want to do is to look a t this notion of
entire move structure is. There are no moves
"agreement" on the asumption of perfect sym-
in this game except the unilateral selection of
metry in the move structure of the game, paying
strategies, and the timing of such selection is
close attention to the "legal details" of the bar-
immaterial since there is no outcome until both
gaining process. We m u s ~ a l s olook at the mean-
players have selected. But the non-tacit (co-
ing of "nonagreement." Since any well-defined
operative) non-zero-sum game is not defined by
game must have some rule for its own termina-
its payoff matrix; the operations by which
choices are made must still be specified. Com- Luce and Raiffa, in effect, define cooperative two-person
games by reference to a payoff matrix and the following
John F. Nash, "The Bargaining Problem," Econornetrica, three stipulations:
xvnr (April 1g50), 155-62, and "Two-Person Cooperative (i) All preplay messages formulated by one player are
Games," Econornetrica, XXI (January 1953)~128-40; John transmitted without distortion to the other player.
Harsanyi, "Approaches to the Bargaining Problem Before (ii) All agreements are binding, and they are enforceable
and After the Theory of Games: a Critical Discussion of by the rules of the game.
Zeuthen's, Hicks', and Nash's Theories," Econornetrica, (iii) A player's evaluations of the outcomes of the game
x x ~ v(April 1956), 144-57; R. Duncan Luce and Howard are not disturbed by these preplay negotiations.
Raiffa, Games and Decisions (New York, 1957)~114ff. Games and Decisions, I 14.

tion, let us look a t the rules for termination corded in some manner that will be visible to the
first.3 referee when the bell rings. Perhaps he keeps
If we are to avoid adding a whole new dimen- it written on a blackboard that the other player
sion to our payoff matrix, in the form of dis- can see; perhaps he keeps it in a sealed enve-
count rates, we must suppose that the game is lope that is surrendered to the referee when the
terminated soon enough so that nothing like the bell rings; perhaps he keeps it punched into a
interest rate enters the picture. We do not want private keyboard that records his current offer
to have to consider the time a t which agreement in the referee's room. When the bell rings the
is reached, in addition to the agreement itself. blackboard is photographed, or the envelope
This is not only a matter of convenience; the surrendered, or the keyboard locked, so that the
game ceases to be moveless except in very spe- referee need only inspect the two "current" of-
cial cases, unless we make this stipulation. For fers as they exist a t midnight to see whether
if the players' time preferences take any shape they are compatible. If they are compatible,
except that of a continuously uniform discount the gains are divided in accordance with the
rate, the game itself changes with the passage "agreement"; if the two players have jointly
of time, and a player can, in effect, change the claimed more than is available, "disagreement"
game itself by failing to reach agreement. The exists and the players get nothing. (Defer, for
notion of a continuously uniform discount rate a moment, ruling on what happens if the two
is probably far too special to treat as a neces- players together have claimed less than the total
sary condition, and anyway has not been made available -whether they get as much as they
an explicit postulate in the models under exami- have claimed or get nothing for lack of proper
nation; so we must assume that the game is agreement. And, in what follows, it will not
somehow gotten over with. matter whether an exhaustive agreement reached
Perhaps the simplest way to terminate the before midnight -i.e., compatibility of the cur-
game is to have a bell ring at a time specified in rent offers occurring before midnight -termi-
advance. There are other ways, such as having nates the game.)
the referee roll dice every few minutes, calling There are other ways of defining "agreement"
off the game whenever he rolls boxcars. (We in terms of the operations by which it is reached
could have the game terminate after a specified or recorded; but if we adhere to the notion of a
number of offers have been refused, but this perfectly symmetrical move structure they will
would change the character of the game by generally, I think, have the property that I am
making certain kinds of communication "real trying to single out for attention. That property
moves" that leave the game different from what is this. There must be some minimum length of
it was before, and perforce lead us into such time that it takes a player to make, or to change,
tactics as the exhaustion of offers.) his current offer. (For simplicity again, let us
For simplicity, suppose that the game will be suppose that the same operation either makes
terminated a t a specified time, known in ad- an offer or changes it, so that we may always
vance to the players, and for convenience let us assume that a '(current offer" exists.) There
call that final moment "midnight." If agree- must then be some critical moment in time, a
ment exists when the midnight bell rings, the finite period before the midnight bell rings, that
players divide the gains in the way they have is the last moment a t which a player can begin
agreed; if no agreement exists, the players re- the operations that record his final offer. That
ceive nothing. is, there is some last moment before the bell
Next, what do we mean by "agreement"? rings, beyond which it is too late to change one's
For simplicity, suppose that each player keeps existing offer. Under the rules of the game and
(or may keep) his current "official" offer re- the rationality postulate both players know this.
And by the rule of symmetry this moment must
8The model discussed here is quite abstract, artificial, and
unrealistic; but it does have the advantage of helping to test be the same for both players.
whether even in an artificially abstract model it is fruitful to From this follows the significant feature. The
postulate perfect symmetry in the move structure and to last offer that it is mechanically and legally
treat asymmetry as a special case, symmetry as the more
general case. possible for a player to make is one that he

necessarily makes without knowing what the From this it follows that the solution of the
other player's final offer is going to be; and the cooperative game must be identical with that of
last offer that a player can make is one that the the corresponding tacit game (if the latter has
other player cannot possibly respond to in the a predictable and efficient solution). I t must,
course of the game. Prior to that penultimate because the tacit game comes as an inevitable,
moment no offer has any finality; and a t that mechanical sequelTo the cooperative game.
last moment players either change or do not At this point it looks as though the coopera-
change their current offers, and whatever they tive feature of the game is irrelevant; the play-
do is done in complete ignorance of what the ers really need not show up until I I :59 ; in fact
other is doing, and is final.4 they do not need to show up at all. The preplay
This must be true. If either could get a communication and ability to reach binding
glimpse of the other's final offer in time to do agreements, which were intended to character-
anything about it, or if either could give the ize the game, prove to be irrelevant; the co-
other a glimpse of his own final offer in time operative game as a distinct game from the tacit
for the other to respond, it is not -and is game does not exist.6
known to be not-a final offer.5 But this conclusion is unwarranted. First, the
But now we have reached an important con- corresponding tacit game may not have a con-
clusion about the perfectly move-symmetrical fidently predicted efficient s ~ l u t i o n .More
~ than
bargaining game. I t is that it necessarily gives that, certain details of the cooperative game that
way, a t some definite penultimate moment, to a might have seemed to be innocuous from the
tacit (non-cooperative) bargaining game. And point of view of explicit negotiation may affect
each player knows this. ' I n his 1953 article, "Two-person Cooperative Games,"
The most informative way to characterize the Nash presents a model that is explicitly tacit in its final
stage. The model's relation to the cooperative game was
game, then, is not that the players must reach heuristic: it was to help to discover what might constitute
overt agreement by the time the final bell rings "rational expectations" (and hence the indicated rational
or forego the rewards altogether. I t is that they outcome) in the corresponding cooperative game. The ar-
gument of the present paper is that the relation is likely t o
must reach overt agreement by a particular be mechanical rather than intellectual if a symmetrical move
(and well identified) penultimate moment - structure is strictly adhered to, and that with strict sym-
metry it is difficult, perhaps impossible, to define the corre-
when the "warning bell" rings - or else play sponding non-tacit game that was the ultimate subject of
the tacit variant of the same game. study.
Each player must be assumed to know this I t should be emphasized that bargaining-game solutions
that (like the Nash and Harsanyi solutions) depend on a
and may, if he wishes, by simply avoiding overt clearly recognized zero point - i.e. on an unambiguous out-
agreement, elect to play the tacit game instead. come that reigns in the absence of overt agreement -cannot
And if we assume (for the moment) that the necessarily be applied to a cooperative game that is based
on a matrix of choices. A matrix (unless perhaps all payoffs
tacit game has a clearly recognized solution, and are zero except in the diagonal) does not have a zero point
that the solution is efficient, each player has a defined by the rules. There is consequently no "normal
pure minimax behavior strategy during the ear- form" consisting of a convex region and associated zero point
unless there is available a fully adequate theory that "solves"
lier stage. Either player can enforce this tacit the tacit game (and does so in a manner that the players can
solution by abstaining from agreement until the take for granted). One may, following Luce and Raiffa
warning-bell rings; neither can achieve any- (e.g., page 137) take the players' "security levels" (maximin
values) as the zero point; but this is either arbitrary or based
thing better from a rational opponent by verbal on the hypothesis that, left to themselves, the players could
bargaining. succeed in doing no better than this in the tacit game. The
latter hypothesis, especially where there are pure-strategy
'Incidentally, the argument is unaffected by supposing efficient points (as in Braithwaite's game, and as in the
that a player can change his offer "instantaneously" as long Luce-Raiffa matrix discussed in note zo below), is a weak
as we keep the symmetrical rule that both can do it "equally hypothesis that can be empirically refuted; it assumes that
instantaneously" as the final bell rings. rational players are incapable of correlating strategies with-
There is a mechanical assumption here that in the proc- out communicating, while in fact this is something they often
ess of making a new offer one can stop and start over. The can do even in the face of conflicting preferences. (This point
case is slightly more complicated if an offer started one and is taken u p again in note 20.) The potential ambiguity of
one-half minutes before midnight is necessarily the last offer the zero point is the issue between Harvey Wagner and John
because the process cannot be started again until a minute Harsanyi in the former's, "Rejoinder on the Bargaining
has passed and by then the critical point has been passed. Problem," Southern Economic Journal, xxw (April 1958),
This case will be looked at again below. 48-82.
the character of the tacit game; similarly, pre- point yields zero for both players. Thus the
play communication that has no binding effect variant game, which seemed to differ only in-
on the players themselves may also affect the consequentially, is in fact drastically different
character of the tacit game. For example, just from the original game; but it does not appear
consider the following variant of the cooperative so until we have identified the terminal tacit
game. game as a dominating influence.'
Instead of saying that the players may divide T o take another example, suppose there are
a set of rewards if they can reach agreement on IOO individual objects to be divided and that,
an exhaustive division, let us say that the play- although they are fungible as far as value is
ers may divide a set of rewards to the extent concerned, the agreement must specify precisely
that they have reached agreement on a division; which individual items go to which individual
they may divide such portion of the available players. If the rules require that full and ex-
rewards as they have already agreed on by the haustive agreement be reached, then in the tacit
time the bell rings. If, for example, there are game the players are dependent on their ability
one hundred indivisible objects and they have not only to divide the total value of the objects
reached agreement on how to divide eighty of in coordinated fashion but to sort out the IOO
them when the bell rings, the twenty items in individual objects into two piles in identical
dispute revert to the house, while the eighty on fashion. If, then, one of the players has de-
which agreement was reached will be divided in manded specific items worth 8 0 per cent of the
accordance with the agreement.s total and the other player has refused, the for-
Now, in the explicit-bargaining (cooperative) mer has an advantage in the tacit game. The
case, if we had already concluded that there was only extant proposal for dividing the IOO ob-
an efficient solution to this game - i.e., that the jects is the one player's specification of 80 that
players would in fact reach an exhaustive agree- would satisfy him; the chances of their concert-
ment -we should probably have considered ing identically on any other division of the IOO
this reformulation of the problem inconsequen- objects, equal or unequal between them, may be
tial. I t only says, in effect, that bargaining so small that they are forced for the sake of
should take the form of each player's writing agreement into accepting the only extant pro-
down the totality of his claim and that conces- posal in spite of its bias. Thus preplay com-
sions shall take the form of each player's delet- munication has tactical significance in that it
ing items from his list of claims, with full agree- can affect the means of coordination once the
ment being reached when no more items are in tacit stage of the game has been reached.
conflict on the lists of claims. But when we If now, in considering the tactical implica-
look at the tacit case, the game is drastically tions of this last point, we insist on a rule of sym-
altered by this reformulation. The tacit game metrical behavior, we must conclude that if
now has a perverse incentive structure. There either player opened his mouth to drown out
is no rational reason for either player to demand what the other was about to say, he would al-
less than the whole of the available reward; wavs find the other player
- - also with his mouth
each knows this and knows that the other knows
might seem that we can draw a by-product from the
it' There is incentive reduce analysis here, namely, the observation that in order to set
because any residual dispute costs the player no up a "truly" cooperative (non-tacit) game, the legal defini-
more than he would losiif he claim tion of agreement must be such as to make the ultimate tacit
game perverse, so that the players must reach binding agree-
to the The sing1e ment before the warning bell or suffer complete loss. But
there is still a problem. Assuming that the players themselves
In the case of a single divisible object like money, the can define "agreement" for purposes of agreement prior to
corresponding rule might be that they divide the money in the final bell, and that the perverse rule only governs the
accordance with their offers after the house has removed the definition of "agreement" at termination if no prior agree-
"overlap." Each player obtains as much as the other im- ment exists, we must now provide (or assume the players to
plicitly accords him; if one is demanding 65 per cent of the provide) an operational definition of agreement. If it is like
money at the end of the game, and the other 5 5 per cent, the our earlier definition, all they accomplish is to make the
second has been accorded 35 per cent and the first 45 per perverse cooperative game into a benign one, one minute
cent; these amounts are outside the range of dispute and shorter, which is equivalent to a tacit game two minutes
constitute the LLagreement." shorter than the original.

open, both knowing that if either spoke the Is the nondiscriminatory, move-symmetrical

other would be found to be speaking, neither game a ('general" game, one that gets away from

able to hear the other, and so on. I n other words, "special cases"? Or is it a special, limiting case

the assumption of complete symmetry of be- in which the most interesting aspects of the co-

havior as a recognized foregone conclusion seems operative game have vanished?

to preclude the very kind of action that might I t should be emphasized that the fruitful al-

have seemed to enrich the game at the stage of ternative to symmetry is not necessarily the

preplay communication. assumption of asymmetry, but just nonsymme-

But by now we have certainly pressed the try, admitting both symmetry and asymmetry as
perfect move-symmetrical game as far as is possibilities without being committed to either
worthwhile.1 We could go on to analyze this as a foregone conclusion.
game in more detail, considering such things as An illustration may help. Suppose we were to
alternative ways of terminating the game or of analyze the game in which there is $100at the
defining "agreement," etc.; it seems more worth- end of the road for the player who can get there
while, however, to raise at this point the question first. This game of skill is not hard to analyze:
of whether the perfectly "moveless" or "move- the money goes to the fastest, barring accidents
symmetrical" game is a profitable one to study. and random elements. We can predict rational
l o o n e detail may be worth pursuing, in line with an
behavior (running) and the outcome (money
earlier footnote. Suppose that it takes one minute to make to the fastest). Ties will occasionally occur;

or change an offer and (in contrast to the earlier version) but they will occur a t the end of a race, not be
that the process of recording a new offer, once started, can-
not be stopped before it is completed. Under this procedure, taken for granted a t the outset. We need an
any offer initiated during the next to last minute of the game auxiliary rule to Cover ties, but it would not
is one's final offer. If this final offer cannot be communi- dominate the game or the analysis.
cated to the other player before the expiration of the minute,
the game is essentially the same as before; "simultaneous" the same game played in a
now means within a minute of each other for practical pur- tion in which everybody can run exactly as fast
poses, and again neither can see the other's final offer as he anybody else, and everybody knows it. N~~
initiates his own, no matter what time during the final min-
ute the offers are initiated. But suppose one punches his what race ends in a tie, the
offer into a visible board which remains locked for one auxiliary rules are all that matter. But since a
minute while the offer is recorded, so that the other player tie is foregone conclusion, why would they
can see one's offer in a few seconds although one cannot
initiate a change until the minute's delay is up. (And sup- run?
pose that neither can make himself visibly incapable of seeing The perfectly move-symmetrical cooperative
the other's offer once it is so recorded.) In this case, if the game little like that foot race. B ~ ~ -
two offers during that final minute are not simultaneous, the
player who moves second makes his final offer in full knowl- gaining in the one case is as unavailing as 1%-
edge of the other's; and since his only chance of winning work in the other; every player knows in ad-
anything is to accept it, he must accept whatever the other
has offered. Thus "second move" loses if the first mover
vance that all moves and tactics are foredoomed
knows that the other is waiting. We now have a game that to neutralization by the symmetrical ~otentiali-
can be characterized as follows: the players dally around ties available to his opponent. The interesting
for 23 hours 58 minutes and then play a game lasting one
minute, this game allowing each player one and only one elements that we might inject in the bargaining
offer which he can make a t any time during the minute. This game are meaningless if perfect symmetry, and
game offers, in effect, three strategies to a player, namely, its acceptance as inevitable by both players, are
( I ) assume the other will wait, and demand 99 per cent;
( 2 ) assume both will make simultaneous offers, and demand
imposed on the game by its definition.
whatever is indicated by the tacit game; (3) wait. ~f both What should we add to the game to enrich it
wait, the game is still to be played. If there is a finite num- if the assumption of symmetry is dropped?
ber of potential waits, we have strategies of wait-once-then-
demand-99-per-cent, wait-once-then-demand-tacit-solution; There are many moves that are often available,
wait-twice-demand-99-per-cent, wait-twice-demand-tacit-SO- but not necessarily identically available to both
lution; etc. This game (the "tacit supergame" consisting of
all strategies for playing the one-minute game) is then the
players, in actual situations. Moves would
game; and it has, if we wish to accept it, its own "solution include commitments, threats, promises; tam-
in the strict sense" which consists of all strategies (all lengths pering with the communication system; invoca-
of waits) that end in demands that correspond to the s o h - tion of penalties on promises, commitments, and
tion of the tacit game. (For the definition of a solution in
the strict sense in a tacit two-person game, see Lute and threats; conveyance true
Raiffa, 106ff.) identification; and the injection of contextual
detail that may constrain expectations, particu- symmetry as an interesting possibility, but not
larly when communication is incomplete.ll as a foregone conclusion; stalemate and the
To illustrate, suppose in the earlier coopera- anticipation of it become interesting possibilities
tive game there is a turnstile that permits a if the actions and information structure are in
player to leave but not to return; his current fact conducive to ties. But with non-symmetry
offer as he goes through the turnstile remains as our philosophy, we need not be obsessed with
on the books until the bell rings. Now we have the possibility of ties.
a means by which a player can make a "final" Again, if one player can make an offer and
offer, a "commitment"; whoever can record an destroy communication, he may thereby win
offer favorable to himself and known to the the ensuing tacit game by having provided the
other, and leave the room, has the winning only extant offer that both players can converge
tactic. Of course it may win for either of them; on when they badly need to concert their choices
but this may mean that we end up with something later during the final tacit stage. T o be sure, we
like a foot race, and the one closest to the turn- can consider what happens when identical ca-
stile wins. By analyzing the tactic, and its in- pacities for destruction of communication are
stitutional or physical arrangements, we may present, and both players must recognize that
determine who can make first use of it. they may simultaneously destroy communica-
We have not, it should be noted, converted tion without getting messages across; but this
the game of strategy into a game of skill by interesting symmetrical case seems to be better
letting them race for the turnstile. I t remains considered a special one, not the general case.
true that one wins when he gets to the turnstile I n summary, the perfectly move-symmetrical
first only through the other's cooperation, only bargaining game is not a fruitful general case,
by constraining the other player's choice of but a limiting case that may degenerate into an
strategy. He does not win legally or physically ordinary tacit game. The cooperative game is
by going through the turnstile; he wins strate- rich and meaningful when "moves" are admit-
gically. He makes the other player choose in his ted; and much of the significance of the moves
favor. I t is a tactic in a game of strategy, even will vanish if complete symmetry in their avail-
though the use of it may depend on skill or ability to the players is stamped into the defini-
locational advantage. tion of the game. It is the moves that are in-
We can even put a certain kind of symmetry teresting, not the game without moves; and it is
into the game now, without destroying it; we the potential asymmetry of the moves that makes
can flip a coin to see who is nearest the turnstile them most interesting.
when the game begins, or let the players be
similarly located and of similar speed but with
random elements determining who gets to the
turnstile first. Though the game is now non- Symmetry is not only commonly imposed on
discriminatory, the outcome would still be asym- the move structure of games but adduced as a
metrical, because each player has an incentive plausible characteristic of the solution of the
to run to the turnstile, leaving behind a standing game or of the rational behavior with which the
offer in his own favor .I2 solution must be consistent. Nash's theory of
We can include some risk of a tie, especially the two-person cooperative game explicitly pos-
if there are two turnstiles and the players might tulates symmetry, as does Harsanyi's.13 The
go to them simultaneously. This constitutes symmetry postulate is certainly expedient; it
often permits one to find a solution to a game
"For an extensive analysis of several such moves, see the and to stay - if he wishes -within the realm
author's "Strategy of Conflict: Prospectus for a Reorienta-
tion of Game Theory," Journal of Conflict Resolution, 11 of mathematics. There are few similarly potent
(September 1958).
I2It could be argued a t this point that the expected value " I t is not always clear whether symmetrical behavior, as
of the game is still symmetrically divided between the play- prescribed for example in the earlier quotation from Har-
ers, and that the analyst may consequently still view the sanyi, is to be considered a rule of the game under analysis
game as symmetrical in terms of average outcomes. But if or a behavior postulate separate from the definition of the
he does he commits himself to a minimum of insight into the game; but for the purpose of the present argument it does
game and the way the game will be played. not matter.

concepts that compete with it as bases for solv- Now this intuitive formulation involves two
ing a game. But the justification for the sym- postulates. First, that one bargainer will not
metry postulate has not been just that it leads concede more than he would expect to get if he
to nice results; it has been justified on grounds himself were in the other position. second, that
that the contradiction of symmetry would tend the only basis for his expectation of what he
to contradict the rationdity of the two players. would concede if he were in the other position
This is the underpinning that I want to attack. is his perception of symmetry.
What I am going to argue is that though sym- The intuitive formulation, or even a careful
metry is consistent with the rationality of the formulation in psychological terms, of what it is
players, it can not be demonstrated that asym- that a rational player expects in relation to an-
metry is inconsistent with their rationality, other rational player, poses a problem in sheer
while the inclusion of symmetry in the definition scientific description. Both players, being ra-
of rationality begs the question. I then want to tional, must recognize that the only kind of
offer what I think is an argument in favor of "rational" expectation they can have is a fully
symmetrical solutions, an argument that tends shared expectation of an outcome. I t is prob-
to make symmetry but one of many potential ably not quite accurate -as a description of
influences on the "rational outcome" with no the psychological phenomenon - to say that
prima facie claim to pre-eminence. one expects the second to concede something or
Explicit statements of the relation between to accept something; the second's readiness to
symmetry and rationality have been given by concede or to accept is only an expression of
Harsanyi. He says, "The bargaining problem what he expects the first to accept or to concede,
has an obvious determinate solution in a t least which in turn is what he expects the first to
one special case: viz., in situations that are expect the second to expect the first to expect,
completely symmetric with respect to the two and so on. T o avoid an "ad infinitum" in the
bargaining parties. I n this case it is natural to descriptive process, we have to say that both
assume that the two parties will tend to share sense a shared expectation of an outcome; one's
the net gain equally since neither would be pre- expectation is a-belief that both identify the
pared to grant the other better terms than the same outcome as being indicated by the situa-
latter would grant him."14 I n a later paper he tion, hence as virtually inevitable. Both play-
refers to the symmetry axiom as the "funda- ers, in effect, accept a common authority - the
mental postulate" and says, "Intuitively the as- power of the game to dictate its own solution
sumption underlying this axiom is that ; rational through their intellectual capacity to perceive
bargainer will not expect a rational opponent to it -and what they expect is that they both
grant him larger concessions than he would perceive the same solution.16
make himself under similar conditions." l5
Foundation Discussion Paper No. 46, December 11, 1957,
"Harsanyi, op. cit., 147. He goes on to say, "For in- quoted by permission of the author.)
stance, everybody will expect that two duopolists with the lU Viewed in this way, the intellectual process of arriving
same cost functions, size, market conditions, capital resources, at "rational expectations" in the full-communication bar-
personalities, etc., will reach an agreement giving equal prof- gaining game is virtually identical with the intellectual proc-
its to each of them." ess of arriving at a coordinated choice in the tacit game.
"The full quotation deserves to be given: "What the The actual solutions might be different because the game con-
Zeuthen-Nash theory of bargaining essentially proposes to texts might be different, with different suggestive details;
do is to specify what are the expectations that two rational but the nature of the two solutions seems virtually identical
bargainers can consistently entertain as to each other's bar- since both depend on an agreement that is reached by tacit
gaining strategies if they know each other's utility functions. consent. This is true because the explicit agreement that is
The fundamental postulate of the theory is a symmetry reached in the full-communication game corresponds to a
axiom, which states that the 'functions defining the two par- priori expectations that were reached (or in theory could have
ties' optimal strategies in terms of the data (or, equivalently, been reached) jointly but independently by the two players
the functions defining the two parties' final payoffs) have before the bargaining started. And it is a tacit agreement in
the same mathematical form, except that, of course, the the sense that both can hold confident rational expectations
variables associated with the two parties have to be inter- only if both are aware that both accept the indicated solu-
changed. Intuitively the assumption underlying this axiom tion in advance as the outcome that they both know they
is that a rational bargainer will not expect a rational op- both expect.
ponent to grant him larger concessions than he would make There is a qualification to this point. With full informa-
himself under similar conditions." (Harsanyi, 'IBargaining tion about each other's value systems and a homogeneous
in Ignorance of the Opponent's Utility Function," Cowles set of gains to be divided, there may be an infinity of equiv-
I n these terms the first (explicit) part of the part of the definition of rationality; to do so
Harsanyi hypothesis might be rephrased: that would destroy the empirical relevance of the
there is, in any bargaining-game situation (with theory and simply make symmetry an independ-
perfect information about utilities), a particular ent axiom. We must have a plausible definition
outcome such that a rational player on either of rationality that does not mention symmetry,
side can recognize that any rational player on and show that asymmetry in the bargaining
either side would recognize it as the indicated expectations would be inconsistent with that
solution. The second (implicit) part of the definition. For our present purpose we must
hypothesis is that the particular outcome so suppose that two players have picked $80 and
recognized is determined by mathematical sym- $20 by agreement, and see whether we can iden-
metry. The first we might call the "rational- tify any kind of intellectual error, misguided
solution postulate"; it is the second that con- expectations, or disorderly self-interest, on the
stitutes the "symmetry postulate." part of one or both of them, in their failure to
The question now is whether the symmetry pick a symmetrical point.
postulate is derived from the players' rationality Specifically, where is the "error" in B's con-
- the rationality of their expectations -or cession of $80 to A? He expected -he may
must rest on other grounds. Additionally, if it tell us, and suppose that we have means to check
rests on other grounds, what are they and how his veracity ( a modest supposition if full in-
firm is the support? formation of utilities is already assumed!) -
T o pursue the first question, whether sym- that A would "demand" $80; he expected A to
metry can be deduced from the rationality of expect to get $80; he knew that A knew that he,
the players' expectations, we can consider the B, expected to yield $80 and be content with
rationality of the two players jointly, and in- $20; he knew that A knew that he knew this;
quire whether a jointly expected non-symmetri- and so on. A expected to get $80, knew that B
cal outcome contradicts the rationality postu- was psychologically ready because he, B , knew
late. If two players confidently believe they that A confidently expected B to be ready, and
share, and do share, the expectation of a par- so on. That is, they both knew-they tell us-
ticular outcome, and that outcome is not sym- and both knew that both knew, that the outcome
metrical in a mathematical sense, can we dem- would ineluctably be $80 for A and $ 2 0 for B.
onstrate that their expectations are irrational, Both were correct in every expectation; the
hence that the rationality postulate is contra- expectations of each were internally consistent
dicted? Specifically, suppose that two players and consistent with the other's. We may be
may have $roo to divide as soon as they agree mystified about how they reached such expecta-
explicitly on how to divide it; and they quite tions; but the feat claims admiration as much
readily agree that A shall have $80 and B shall as contempt. The rational-solution postulate is
have $20; and we know that dollar amounts in beautifully borne out; the game seems to have
this particular case are proportionate to utilities, dictated a particular outcome that both players
and the players do too: can we demonstrate confidently perceived. If, a t this point, we feel
that the players have been irrational? we ourselves wouldn't have perceived the same
We must be careful not to make symmetry outcome, we can conclude that one of four hy-
alent solutions, all yielding the same values to the two players,
potheses is false: ( I ) the rational-solution pos-
but no difficulty in agreeing on an arbitrary choice among tulate, (2) the rationality of A and B, ( 3 ) our
this indifferent set. But tacit bargaining often requires a own rationality, (4) the identity (in all essential
further degree of coordination, namely, a coordinated choice respects) of the game that we introspectively
even among equivalent divisions of the gains. Negotiation
over a boundary line in homogeneous territory is thus dif- play with the game that A and B have just
ferent from the simultaneous dispatch of troops to take up played. But we cannot, on the evidence, declare
positions representing claims; such claims may overlap and
cause trouble even though the terrain values claimed are
the second to be the false one - the rationality
consistent. Thus the coordination ~ r o b l e mis different: and of A and B.
there is no a priori assurance that'the solution to the'tacit Note that if B had insisted on $50, or if A
game (or to games with somewhat incomplete communica-
tion, information, etc.) would be in the set of equivalent
had been content to demand $50, claiming to be
solutions to the fully explicit game. rational and arguing in terms of confidence in

a shared expectation of that outcome, both coordinating device that presents itself to them.
players would have been in "error"; and we They must, consciously or unconsciously, use a
cannot tell, on the evidence, which one is irra- selection procedure that leads to unique results.
tional or whether they both are. Unless we There must be something about the point they
make symmetry the definition of rationality we pick that distinguishes it - if not in their con-
can only conclude that a t least one of the play- scious reasoning, at least in our conscious analy-
ers is irrational or that the rational-solution sis - from the continuum of all possible alter-
postulate does not hold. What we have is at best natives.
a single necessary condition for the rationality Now, is it possible for two rational players,
of both players jointly; we have no sufficient through anything other than sheer coincidence
condition, and no necessary condition that can or magic, to focus their attention on the same
be applied to a single player. particular outcome and each rationally be con-
Nor can we catch them up if we ask them fident that the other is focussed on the same
how they arrived a t their expectations. Any outcome with the same appreciation that it is
grounds that are consistent would do, since any mutually expected? And, if so, how can they?
grounds that each expects the other confidently The answer is that they can; this was dem-
to adopt are grounds that he cannot rationally onstrated by the fact that in a sample of over
eschew. Consistent stories are all they need; 40 people, instructed to concert tacitly on pick-
and if they say that a sign on the blackboard ing the same positive number from among the
said A-$80, B-$20, or that they saw in a bulle- infinity of positive numbers, 40 per cent picked
tin that two other players, named A' and B', the same number; and that, instructed to con-
split $80-$20, and that they confidently per- cert tacitly on picking a single cell from a 3 X 3
ceived that this was clear indication to both of matrix or a 4 X 4 matrix, a substantial majority
them of what to expect-that this was the only managed to do so in spite of sizable odds against
"expectable" outcome -we cannot catch them them in a random sense. How can they? By
in error and prove them irrational. They may using any means that is available: any clue,
be irrational but the evidence will not show it. any suggestion, any rule of elimination, that
There is, however, a basis for denying my leads to an unambiguous choice or a high prob-
present argument. Since I have not actually ability of concerted choice. And one of these
applied an independent test of rationality to rules, or clues, or suggestions, is mathematical
two players, given them the game to play, and symmetry.17
observed the 80-20 split, but have only posed I n a game that has absolutely no details but
it as a possibility to see whether it would imply its mathematical structure, in which no inad-
irrationality if it occurred, one could raise the vertent contextual matter can make itself ap-
objection that it could not occur. The argument preciated by a player as something that the
would rest on the problem of coordination; it other can appreciate too, there may be nothing
would run as follows. to work on but a continuum of numbers. And
If two players jointly expect a priori the same all the numbers can be sorted according to
outcome, and confidently recognize it as their 17The basic intellectual premise, or working hypothesis,
common expectation, they must have the intel- for rational players in this game seems to be the premise that
lectual capacity to pick a particular point in some rule must be used if success is to exceed coincidence,
and that the best rule to be found, whatever its rationaliza-
common. If the whole $100 can be divided to tion, is consequently a rational rule. This premise would
the nearest penny, there are 9,999 relevant divi- support, for example, Nash's model that views an "un-
smoothed" tacit game as the limit of a "smoothed" game as
sions to consider, one of which would have to be the smoothing approaches zero. While this view of the un-
picked simultaneously but separately by both smoothed game is in no sense logically necessary it is a
players as their expectation of the outcome. But powerfully suggestive one that can, in the absence of any
better rationale for converging on a single point, command
how can two people concert their selections of the attention of players in need of a common choice. The
one item out of 9,999, in the sense that their limiting process provides a clue for picking one of the in-
expectations focus or converge on it, except with finitely many equilibrium points that actually exist in the
unsmoothed game. Of course, the premise equally supports
odds of 9,999 to I against them? The answer any other procedure that produces a candidate for election
must be that they utilize some trick, or clue, or among the infinitely many potential choices.
whether they correspond to symmetrical or expectations. (These other aspects are com-
asymmetrical divisions. If all numbers but one monly not contained in the mathematical struc-
represent an asymmetrical split, then sheer ture of the game but are part of the topical
mathematical symmetry is a sufficient rule and content; i.e., they usually depend on the "label-
a supremely helpful one in concerting on a com- ling" of players and strategies, to use the term
mon choice. And it may be possible to set up a of Luce and Raiffa.)
game in such sanitary fashion, with suppression I have no basis for arguing with what force,
of identity of players and of all contextual de- or in what percentage of interesting games,
tail, that there is literally no other visible basis mathematical symmetry does dominate "ration-
for concerting unless impurities creep in.'' al expectations." But I think that the status of
In other words, mathematical symmetry may the symmetry postulate is qualitatively changed
focus the expectations of two rational players by the admission that symmetry has competitors
because it does - granted the other assumed in the role of focussing expectations. For if it
features of the game, like full information on were believed that rational players' expectations
each other's utility systems-provide one means could be brought into consistency only by some
of concerting expectations. Whether it is a po- mathematical property of the payoff function,
tent means may depend on what alternatives are then symmetry might seem to have undisputed
available. claim, particularly if it is possible to find a
That there are other means of concerting, in- unique definition of symmetry that meets cer-
cluding some that may substantially outweigh tain attractive axioms. But if one has to admit
the notion of symmetry, seems amply demon- that other things - things not necessarily part
strated by some of the experiments already of the mathematical structure of the payoff func-
referred to.19 So it is demonstrably possible to tion - can do what symmetry does, then there
set up games in which mathematical symmetry is no a priori reason to suppose that what sym-
does provide the focus for coordinated expecta- metry does is 99 per cent or I per cent of the
tion, and demonstrably possible to set up games job. The appeal of symmetry is no longer mathe-
in which some other aspect of the game focusses matical, it is introspective; and further argu-
''In this view, the theory of Nash (leading to the maxi- ment is limited to the personal appeal of par-
mum-utility-product solution) is a response to the fact that ticular focussing devices to the game theorist
even in the realm of mathematics there are offhand too many as game player, or else to empirical observation.
types of uniqueness or symmetry to provide an unambiguous
rule for selection, hence a need to adduce plausible criteria Thus a normative theory of games, a theory
(axioms) sufficient to yield an unambiguous selection. of strategy, depending on intellectual coordina-
Braithwaite's theory can be characterized the same way. The tion, has a component that is inherently empiri-
fact that the two solutions conflict implies that mathemati-
cians may not have a sufficiently common mathematical cal; it depends on how people can coordinate
aesthetic to satisfy the first part of the Harsanyi postulate, their expectations. I t depends therefore on skill
i.e., to coordinate their expectations on the same outcome.
(R. B. Braithwaite, Theory of Games as a Tool for the Moral and on context. The rational player must ad-
Philosopher, Cambridge, England, 1955; Braithwaite's solu- dress himself to the empirical question of how,
tion is described in Luce and Raiffa, Games and Decisions, in the particular context of his own game, two
145ff.) Braithwaite's construction of the problem as a one-
person arbitration problem, and Luce and Raiffa's reformu- rational players might achieve tacit coordina-
lation of Nash's theory in terms of arbitration rather than tion of choices, if he is to find in the game a
strategy (pages 121-154), seem to emphasize that intellectual basis for sharing an a priori expectation of the
coordination is at the heart of the theory. A legalistic solu-
tion requires some rationalization of a unique outcome; pure outcome with his partner. The identification
casuistry is helpful if the alternative is vacuum. of symmetry with rationality rests on the as-
In For a report and analysis of some of the experiments,
see the author's "Bargaining, Communication, and Limited
sumption that there are certain intellectual
War," Jozcrnal of Conflict Resolution, I (March 1957), 19- processes that rational players are incapable of,
36. Regarding the formal extension of game theory to cover namely, concerting choices on the basis of any-
this type of problem, see the author's "The Strategy of
Conflict: Prospectus for a Reorientation of Game Theory," thing other than mathematical symmetry, and
op. cit., esp. 204-19. Some further elaboration is in the that rational players should know this. I t is a n
author's RAND Corporation paper, P-1385, "Re-interpreta- empirical question whether rational players can
tion of the Solution Concept for the 'Non-Cooperative'
Game" (Santa Monica, 1958). actually do what such a theory denies they can

do and should consequently ignore the strategic explicit bargaining game. And imagine a game's
principles produced by such a theory.20 potential payoffs as consisting of all the points
on or within some boundary in the upper-right
I11 quadrant. We now consider some variants of
this game.
An game>which be sub- First, we are to play the same game in tacit
mitted to experiment, may illustrate the point.
form; each of us picks a point along his own
Let us -whether or not we are strongly at-
axis, and if the resulting point is on or within
tracted to the 'ymmetry postulate) and whether
the boundary, we get the amounts (utilities)
or not we are especially attracted to the particu-
denoted by the coordinates we pick. I conjec-
lar symmetry the Nash -put Our- ture that, in the frame of mind I have asked for,
selves in a frame of mind congenial to accepting
we should probably pick the Nash point. With-
the "Nash point" as the rational outcome of an
out asking precisely why, let us go on to the next
m is interesting that in demanding a symmetrical solu- variant. This game is tacit too, but it differs in
tion to an ostensibly symmetrical tacit (non-cooperative) that We get nothing unless the point whose co-
game, Luce and Raiffa dismiss the two most promising can- ordinates we pick is exactly on the boundary.
didates. They consider (Games and Decisions, 90-94) a
matrix, We get nothing unless we exhaust the available
gains. Caution gets us nowhere; each must
choose exactly as the other expects him to.
Again I propose that, in our present frame of
mind, we ought to take the Nash point.
Finally consider another variant: we are
shown the diagram of the game that has just
been played and told that we are now to be
perfect partners, winning and losing together.
and note that it has pure-strategy equilibrium points in the COnSCiouSof the fact that our present game is
upper-left and lower-right corners. These are ruled out on
grounds that LLwhateverrationalization I give for either a1 On a bargaining game we are to pick)
or an there is, by the symmetry of the situation, a similar without communicating, coordinates of a point
rationalization for player 2 , and so it seems inevitable that that lies exactly on the boundary; if we do, we
we both lose." They then look at a pair of maximin strate-
gies, which are unsatisfactory because they do not produce both win prizes - the same prizes no matter
an equilibrium point, and a minimax strategy which they what point we succeed in picking together -
find even inferior. But the important question is whether and if we fail to pick a point on the boundary
players who are both rational and imaginative are quite as
impotent as Luce and Raiffa insist. Can players correlate We get In this pure game,
strategies without communicating? This is an empirical 1 conjecture again that we should (would) in
question. (The writer's experiments, mentioned above, dem- present frame of mind pick the ~~~h point.
onstrate that, at least sometimes, they can.) Offhand it may
seem hard for them to concert on a non-symmetrical pair of because we need ration-
strategies. But much the hardest part is just recognizing alization that leads to a unique point; and in
that they have to ; the question of how to do it then becomes the ,,,text, the bargaining analogy provides it.
a practical matter. They must jointly and tacitly find a clue
to the concerting of their choice. Of course, a non-symmetri- Unless there is a sharp corner (which is then
cal solution in the above matrix is a discriminatory one; it likely to be the Nash point anyway) ; or a sim-
quite arbitrarily condemns one of the players to a smaller ple mid-point as when the boundary is a straight
gain that the other for reasons that may seem purely acci-
dental or incidental. But we have to suppose that a rational line or circular arc (which again coincides with
player can discipline himself to accept the lesser share if the the Nash point) ; or some especially suggestive
clue points that way. Only a discriminatory clue can point form that seems to point towards a particular
to a concerted choice; to deny the discrimination is to deny
the premise that a clue can be .jointly. found and jointly point; or unless there is an impurity (such as
acted on in the interest of an outcome that is jointly fir a dot on the boundary, from a printer's error,
superior to any symmetrical outcome. Luce and Raiffa con-
clude their discussion of this particular game with the remark
or a single point whose coordinates are whole
that "although this seemingly innocuous game possesses some numbers, etc.), we may be led to search for a
symmetries it is difficult to see how to exploit them." But "uniqueJ' definition of symmetry to fall back on,
the real key to this seemingly innocuous game is that it may,
particularly when presented in a context, possess some asym- Nash-type symmetry being as plausible as any
metries, and the object is to exploit them. I can think of, not as simple as some (like the
intersection of a 45-degree line with the origin demanding tacit bounded-area variant of the
and others of that ilk) but less ambiguous on game; and that in turn takes the heart out of
its own level of sophistication. any player in the explicit bargaining game who
And if the Nash point appeals to us power- might hope that expectations could focus any-
fully in the bargaining game, it must do so be- where else.
cause we are confident that it appeals equally I n other words, by postulating the need for
to our partner who in turn we believe to be coordination of expectations we seem to have
aware that our views coincide; it must therefore an empirical reference for something like the
appeal to us in the pure-coordination game as a Nash axioms. What a theory like Nash's needs
unique point that the partner will consider to be is the premise that a solution exists; it is the
obviously obvious. observable phenomenon of tacit coordination
What does this prove or suggest? I am not that provides empirical evidence that (some-
arguing for the Nash point. I am arguing rather times) rational expectations can be tacitly fo-
that the appeal of the Nash point to a game cussed on a unique (and perhaps efficient) out-
theorist (as introspective game player) may be come, and that leads one to suppose that the
the reverse of the sequence I have just run same may be possible in a game that provides
through. I t may be the focal quality of the nothing but mathematical properties to work on.
Nash-point in the pure coordination game - The Nash theory is vindication of this supposi-
the unequivocal usefulness of a uniquely defined tion -complete vindication if it dominates all
symmetry concept, when no non-mat.hematica1 competing mathematical solutions in terms of
impurities are available to help - that makes mathematical aesthetics. The resulting focal
it a controlling influence in the tacit and terribly point is limited to the universe of mathematics,
cooperative boundary-line variant of the game; however, which should not be equated with the
that in turn makes it a reliable guide in the less universe of game theory.

- Page 1 of 2 -

You have printed the following article:

For the Abandonment of Symmetry in Game Theory
T. C. Schelling
The Review of Economics and Statistics, Vol. 41, No. 3. (Aug., 1959), pp. 213-224.
Stable URL:

This article references the following linked citations. If you are trying to access articles from an
off-campus location, you may be required to first logon via your library web site to access JSTOR. Please
visit your library's website or contact a librarian to learn about options for remote access to JSTOR.


The Bargaining Problem
John F. Nash, Jr.
Econometrica, Vol. 18, No. 2. (Apr., 1950), pp. 155-162.
Stable URL:

Two-Person Cooperative Games
John Nash
Econometrica, Vol. 21, No. 1. (Jan., 1953), pp. 128-140.
Stable URL:

Approaches to the Bargaining Problem Before and After the Theory of Games: A Critical
Discussion of Zeuthen's, Hicks', and Nash's Theories
John C. Harsanyi
Econometrica, Vol. 24, No. 2. (Apr., 1956), pp. 144-157.
Stable URL:

Rejoinder on the Bargaining Problem
Harvey M. Wagner
Southern Economic Journal, Vol. 24, No. 4. (Apr., 1958), pp. 476-482.
Stable URL:
NOTE: The reference numbering from the original has been maintained in this citation list.

- Page 2 of 2 -

Approaches to the Bargaining Problem Before and After the Theory of Games: A Critical
Discussion of Zeuthen's, Hicks', and Nash's Theories
John C. Harsanyi
Econometrica, Vol. 24, No. 2. (Apr., 1956), pp. 144-157.
Stable URL:

NOTE: The reference numbering from the original has been maintained in this citation list.