Tnv Co~sv Lvc:uvv Svvivs

The Coase Lecture series, established in honor of Ronald H.

Coase, Clifton R. Musser Professor Emeritus of Economics at
the University of Chicago Law School, is intended to provide
law students and others with an introduction to important
techniques and results in law and economics. The lectures pre-
suppose no background in the subject.
An Introduction to Game Theory and the Law
Randal C. Picker *

I am pleased to have the opportunity to give the third of the

three lectures in the Law Schools inaugural Coase Lecture Series. I
have to confess I am still at a stage in life when I think about how
things will look on my resume, and to put down the Coase Lecture,
I suspect, adds real luster to it. Nonetheless, we might want to call
these lectures something else. My suggestion is The Bar Stool
Lecture Series. That wouldnt sound as distinguished and hence
wouldnt do much for my resume, but it more accurately captures
what the mission of the talk is. Here is the test for this talk: Given
two bar stools and a stack of cocktail napkins, could the ideas in this
talk be explained to an intelligent person in a crowded bar with a
bank of TVs showing the Bulls and the Blackhawks? I f this talk
succeeds at that level, I will have accomplished my mission; if it does
not, then I will have to consult with the Dean to get a larger budget
for eld research for my next big talk.
The bar stool test is a test of simplicity, of making an idea acces-
sible to someone who is not a specialist in an area. I t is a test that all
of Ronald Coases work that I know passes quite easily. I t is the re-
markable combination of simplicity and depth, which I guess travel
together if you are smart enough, that defines Coases work. The
material that I will discuss today is, I think, fairly simple, though
some of it is relatively new. And to give credit where credit is due,
some of the work I will describe today is part of a joint effort wi th
Doug Baird and Rob Gertner.
This will be an eight-cocktail-napkin talk: I want to talk about
two basic forms for games, the normal form and the extensive form;
four ways of predicting the outcomes of games, through dominance
arguments, Nash arguments, backwards induction, and forward i n-
duction; and two interesting ideas about game theory and the law.

*Randal C. Picker is Professor of Law, University of Chicago Law

School. I have benefited from extensive discussions with Douglas G. Baird
and Robert H. Gertner. I thank the Sarah Scaife Foundation and the Lynde
and Harry Bradley Foundation for their generous research support.
a Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
. GameTheory and StrategicBehavior
As a discipline, law and economics advanced on the strong back
of classical microeconomics. I ndividual decisionmakers maximized
utility or profits subject to constraints. These individuals were treated
either as pricetakers in competitive settings or pricesetters in mo-
nopolies. They were also perfectly informed. A sizable and largely
successful academic legal literature grew out of taking first derivatives
and ruthlessly applying the discipline of the microeconomists
marginal analysis to a vast array of legal problems.
.
The last twenty years have seen a major shift in the fundamental
methodological tools used by microeconomic theorists. Game theory
has emerged to augment the standard, polar approaches of pure
competition and monopoly. I n a competitive setting, individuals or
firms are seen as having no real decisions to make. Prices are given,
and individuals and firms are pricetakers. The other production
paradigm, monopoly, treats the monopolist as a pricesetter for a
given demand curve. I n a game-theoretic setting, rational actors
need worry about the actions of othersthis is the fundamental
strategic interdependence that game theory addresses. Other settings
lack the back-and-forth quality that characterizes strategic settings.
Game theory sounds like funvisions of the gamut from
Candyland to Monopoly spring to mind. A definition might be
useful; as a rough cut, try: gametheory is a set of tools and a language
for describing and predicting strategicbehavior. I will discuss in a bit
what these tools are and how to apply them, but I want to focus
first on the core concept in the definition, strategic behavior.
Strategic settings are situations in which one person would like to
take into account how a second person will behave in making a de-
cision, and the second person would like to do likewise. Strategic
settings typically involve two or more decisionmakers, and the pos-
sibility of linking one decision to a second decision, and vice versa.

.
A sample of well-known textbooks and research monographs makes the
point. Look at Mitch Polinskys Introduction to Law and Economics, which i s
now in its second edition; Cooter and Ulens Law and Economics textbook,
which came out in .; and two torts monographs, Landes and Posners The
EconomicStructureof Tort Law and Steven Shavells EconomicAnalysisof Accident
Law, both of which were published in .,.
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w
Consider the airlines industry. Whether Northwest will cut fares
may depend on how American and United will respond, and the
same, of course, is true for them. I ndeed, Northwest recently filed
suit against American, claiming that Americans introduction of a
new pricing schedule was part of a scheme of predatory pricing de-
signed to put Northwest out of business.
a
Oligopolistic
industriesairlines, computer microprocessors or operating systems,
for exampleare natural settings for strategic interactions.
But so is a country road. I have risen for an early-morning walk.
I would like to enjoy the view, take in the scenery, and generally ig-
nore the cars going by me. You unfortunately are driving your new
Mazda Miata. You want to see how the car handles, to test how it
drives through turns and its acceleration. I f I knew that you were
driving like a maniac, I would want to take that into account in de-
ciding whether to pay much attention to the road. I f you knew that
I was soaking in the countryside and ignoring the road, you would
want to take that into account as well. Our behavioral decisions are
intertwined, and we need to take that fact into account when we
seek to predict likely outcomes. The legal system should take this
into account as well when it establishes antitrust laws for oligopolis-
tic industries or a torts scheme for ordinary accidents.
. Normal FormGames, Dominant Strategies, and theHidden Roleof
Law
~. ThePrisoners Dilemma
The best known bit of game theory is the Prisoners Dilemma. I
will go through the analysis to make clear how much game theory
has already crossed over and to establish some terminology, and will
then move on to more natural settings. So consider the following
game:

a
See Bridget OBrian, Predatory Pricing I ssue is Due to be Taken Up i n
American Airs Trial, Wall Street Journal, July .a, ., ~..
Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
Prisoner 1
Silent
Confess
Silent Confess
-2, -2
0, -10
-10, 0
-6, -6
Prisoner 2
Payoffs: (Prisoner r , Prisoner a )
Figure r : Prisoners Dilemma
Here is the story that this game is trying to capture. We have two
prisoners, or, more generally, two players. They both have commit-
ted a serious crime, but the district attorney cannot convict either
one of them of this crime without extracting at least one confession.
The district attorney can, however, convict them both on a lesser
offense without the cooperation of either. The district attorney tells
each prisoner that if neither confesses, they will both be convicted
for the lesser offense. Each will go to prison for two years. This out-
come is represented in the upper left cell.
I f, however, one of the prisoners confesses and the other does
not, the prisoner who confesses will go free and the other will be
tried for the serious crime and given the maximum penalty of ten
years in prison. This applies to both prisoners and is represented i n
the of-diagonal cells. Finally, if both confess, the district attorney
will prosecute both for the serious crime, but not ask for the maxi-
mum penalty. They will both go to prison for six years. This is the
final cell, the lower right cell.
This is a normal formgame. We have identified the players, our
two prisoners; the choices, or strategies, available to them (here, to
be silent or confess); and the outcomes associated with the four
different strategy pairs. The layout here in the bimatrix is the stan-
dard way of representing this normal form game.
Now the solution of the game. Each prisoner wants to mini-
mize time spent behind bars and has no other goal. Moreover, each
is indifferent to how much time the other spends in prison. I ignore
the possibility of altruism or spite. I also ignore the reputational is-
sues that might arise from being known as a snitch or fear of reprisal
for confessing. Finally, the two prisoners have no way of communi-
cating with each other. Each must decide without knowing what
the other will do.
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w
This is a game in which each prisoner has a strictly dominant
strategy. Each is better off confessing regardless of what the other
does. One can solve the game by recognizing that each prisoner is
likely to reason in the following way: I f the other prisoner has de-
cided to keep silent, I am better off confessing. That way I spend no
time behind bars at all, rather than two years. What about the other
possibility? I f the other prisoner confesses, I am also better off con-
fessing. As bad as serving a six-year sentence might be, serving a ten-
year sentence is worse. No matter what the other person does, I am
better off confessing. No prison is better than two years and six years
is better than ten years. Because both prisoners will likely engage i n
this reasoning, both are likely to confess.
The outcomeboth prisoners confessseems counterintuitive
at first because the prisoners would have been better off if both had
remained silent. But this result follows once we assume that we have
structured the payoffs correctly. Even if each prisoner erroneously
believed that the other was altruistic and would confess, we would
still have the same outcome, given our assumption that theprisoners
careonly for themselves. I f a prisoner believes (for whatever reason)
the other will remain silent, confessing is a way of avoiding prison
altogether, the best outcome of all. (Again, if the prisoners care
about something in addition to the length of time spent in jail, we
have specified the payoffs incorrectly. The premise of the game is
that the players are both selfish.) The result is not at all odd once
one recognizes that the prisoners lack a means of committing
themselves to remaining silent. As long as the two prisoners cannot
reach any agreement with each other and as long as their only con-
cern is time spent in prison (and not, let us say, their reputations as
finks), their individual interest will lead them to confess, even
though they are jointly better off remaining silent.
The power of the Prisoners Dilemma comes from the incon-
gruence between private benefit and the collective good. I ndividually
rational decisionmaking leads to collective disaster. The Prisoners
Dilemma is thus often seen as one of the main theoretical
justifications for government intrusion into private decisionmaking.

See, e.g., Cass R. Sunstei n, After theRights Revolution: Reconceiving the

Regulatory State-. (Harvard Univ. Press, .c).
6 Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
Legislation almost appears attractive given the collective disaster that
results from individual decisionmaking in the dilemma.
I say almost for two reasons. First, the existence of private fail-
ure tells us nothing about whether government decisionmaking en-
joys a comparative advantage over private decisionmaking. The
Churchill line about democracydemocracy is the worst form of
Government except all those other forms that have been tried from
time to timemay apply here as well. We need to know much
about the quality of government decisionmaking before we can
summarily abandon private decisionmaking. The second reason for
being cautious about relying on a simple game-theoretic model such
as the Prisoners Dilemma to justify legal intervention will require
more hardware, so I will return to it at the end of this talk.
n. An ExamplefromtheLaw of Torts
Many legal settings can be represented as normal form games
and solved by identifying dominant strategies. Consider an accident
on a country road involving a motorist and a pedestrian. The likeli-
hood of an accident turns both on how much care the motorist uses
in driving and how much care the pedestrian uses in crossing the
street. We do not expect the motorist to drive so slowly that there is
never any possibility of hitting a pedestrian. Nor do we insist that
the pedestrian cross only when there is no car in sight. We want
them both to take sensible precautions. I f both act reasonably, the
chances of an accident as well as the inconvenience to both parties
are minimized. I f they could bargain with each other, we would ex-
pect that each would agree to act in this way. The problem arises, of
course, because the two are strangers and they cannot communicate
with each other. The motorist and the pedestrian both recognize
that the actions of the other influence what will happen, and that
basic fact must be recognized if we are to have a sensible analysis of
the situation. Game theory is the right tool for this problem.
To jump right in, consider the following game:
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w ,
Pedestrian
No Care
Due Care
No Care Due Care
-100, 0
-110, 0
-100, -10
-20, -10
Motorist
Payoffs: (Pedestri an, Motori st)
Figure a : No Reallocation Law
Here are the stylized facts that this game is seeking to represent. I f
an accident takes place between the motorist and the pedestrian, the
motorist and her car will not be hurt, but the pedestrian will of
course suffer harm. Assume that we can represent the harm to the
pedestrian as a dollar amount and set that amount at $.cc. Both the
motorist and the pedestrian decide on how much care to take.
Assume that they each choose between taking no care and due
care. Representing the decision of how much care to take as a bi-
nary choice oversimplifies greatly, but it is the natural place to start.
Assume that it costs nothing to exercise no care but costs $.c to
exercise due care. Due care is really a legal term for a physical level
of care. Consistent with the convention, due care is the level of
care that minimizes the total expected costs of the accident. We also
need to know how the care choices relate to the probability of an
accident occurring. Assume that the accident is certain to happen
unless both the motorist and the pedestrian exercise due care, but
that there is still a one in ten chance of an accident occurring even if
both exercise due care.
So far, we have set out the brute facts of nature: the choices
available to our players (the motorist and the pedestrian), or what a
game theorist would call the strategies of the players, and the physi-
cal consequences associated with those strategies (whether an acci-
dent takes place and the resulting harm). To fully specify this game,
we need one more item, and it is this item that determines the pre-
cise structure of the game set forth above. We need to know the le-
gal rule for allocating the harms of an accident. The problem of
strategic behavior that the legal analyst faces is a simple problem of
simultaneous decisionmaking. The amount of care that the motorist
and pedestrian each take would turn on the amount of care each ex-
pects the other to take. The amount of care that each takes will turn
in some measure on the legal rule that is in placewhen and to
Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
what extent the motorist will have to pay damages to the pedestrian
in the event of an accident. The first question for the legal analyst
concerns the effect of changes in the legal rule on the behavior of
the parties. Start with a rule of no liability, or of letting the parties
bear their own losses. I n this case, if an accident occurs, the motorist
is not harmed and the pedestrian is harmed, and the legal rule of no
liability does not reallocate any of the harm by having the motorist
pay damages.
We can now explain the game in figure a and determine how to
solve it. I n a legal regime of no liability, a regime in which the mo-
torist was never liable for the accident, the motorist would enjoy a
payo of $c and the pedestrian a payoff of $.cc if neither exercised
care. The cost of no care is zero, an accident is certain to happen,
and the accident harms the pedestrian to the tune of $.cc. I f both
exercised care, the motorist would receive a payoff of $.c and the
pedestrian a payoff of $ac. (The pedestrian invests $.c in care and,
assuming the pedestrian is risk neutral, still faces $.c in expected ac-
cident costs, a one in ten chance of a $.cc accident.) I f the motorist
exercises care and the pedestrian does not, the motorist receives a
payoff of $.c (the cost of taking care) and the pedestrian a payoff
of $.cc (the cost of the accident, which by assumption is certain to
arise unless both take care). Finally, if the motorist does not take care
and the pedestrian does, the motorist has a payoff of $c and the
pedestrian a payoff of $..c (the pedestrian invests $.c in taking care
and still suffers a $.cc injury).
What is the likely outcome of this game? I n this model, taking
care costs the motorist $.c and provides no benefit to the motorist
in return. The motorist always does better by not taking care than
by taking care. We can predict the motorists likely choice of strategy
because there is a single strategy (no care) that, in the context of
this model, is better for the motorist no matter what choice the
pedestrian makes. I n the language of game theory, this is a dominant
strategy (really a strictly dominant strategy). I n corresponding fash-
ion, a strategy which is always worse than another strategy, again
regardless of what the other player does, is a dominated strategy. I n
figure a, due care is a dominated strategy for the motorist. We
should predictas we did in analyzing the Prisoners
Dilemmathat a player will embrace a dominant strategy wherever
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w
possible and will not embrace any strategy that is dominated by
another.
This idea by itself, however, tells us only what the motorist is
likely to do in this model. We cannot use this concept to predict the
pedestrians behavior. Neither of the strategies available to the
pedestrian is dominated by the other. I t makes sense for the pedes-
trian not to take care when the motorist does not, but to take care
when the motorist does. The pedestrian lacks a dominant strategy
because either course of action could be better or worse than the
other depending upon what the motorist does. Note that this game
differs from the Prisoners Dilemma in this regard, as in that game,
both players had a dominant strategy. To predict the pedestrians
behavior, we need to take the idea that players play dominant
strategies one step further. Not only will a player likely adopt a
strictly dominant strategy, but a player will predict that the other
player is likely to adopt such a strategy and will act accordingly. We
can predict, in other words, that the pedestrian will choose a strategy
based on the idea that the motorist will not choose a strategy that is
strictly dominated by another. This idea travels under the name of
iterated dominanceand allows us to solve this game. The pedestrian
should understand that the motorist has a dominant strategyplay
no careand therefore the pedestrian should play no care as well.
Given that the motorist plays no care, the payoff to the pedestrian
from playing due care is $..c and that from playing no care is
$.cc. (Recall that the accident is certain to happen unless both
players play due care; once the motorist will not, the pedestrian is
better off by not wasting any money on care.) The pedestrian should
play no care as well. Neither player exercises care. Note that to
reach this solution, we proceeded iteratively: we first identified the
strategy that the motorist would play using dominance argu-
mentsthis is the first iterationand we next identified the
pedestrians strategy given the motorists strategy as determined i n
the first stage of the argumentthis is the second iteration. This is
the logic of iterated dominance.
This extension of the idea that dominated strategies are not
played requires us to make a further assumption about the rationality
of the players. Players not only act rationally and do the best they
can given their preferences, but they also believe that others act ra-
.c Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
tionally and do the best they can given their preferences. This solu-
tion concept seems plausible if the number of iterations is small.
After all, most people act rationally most of the time and we can
choose our own actions in anticipation that they will act this way. I f
we accept this solution concept, we can solve the game in figure a.
The pedestrian will not exercise care because the pedestrian will be-
lieve that the motorist will not exercise care and, in that event, the
pedestrian, under our assumptions, is better off not exercising care
either. We cannot, however, make this prediction as condently as
we can predict the motorists behavior. The solution to the game
turns not only on the motorist acting in a way that advances her
self-interest, but also on the pedestrian anticipating that the mo-
torist will in fact act in this way.
You might think that these results are specific to the particular
numbers set forth in figure a. The specific result is, though the result
that matters is not. I n the example in figure a, the pedestrian
chooses to exercise no care when the motorist exercises no care.
That outcome is tied directly to the particular probability function
for accidents, which makes it worthless for one player to exercise any
care if the other player is exercising no care. I n generalmeaning
for different probability functions for accidentsthe pedestrian
might choose more or less than due care. The general result is the
result that matters: under a rule of no reallocation of losses and
where any harm from the accident will be borne by the pedestrian,
the motorist lacks an appropriate incentive to take care. I ndeed, as
shown aboveand this is a general resultexercising no care is a
dominant strategy.
Thus, play under a rule of no liability puts us far from the social
goal of having both players exercise due care. This result in itself is
hardly startling. To say that the strategy of taking due care is domi-
nated by another strategy of taking less than due care restates in the
language of game theory a familiar insight from law and economics,
the insight that in a world without tort law, parties tend to take less
than due care because they do not fully internalize the costs of their
actions.

The motorist enjoys all the benefits of driving fast, but

See, e.g., W. Landes and R. Posner, TheEconomicStructureof Tort Law

6a (Cambridge, Harvard Univ. Press, .,).
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w ..
does not bear all the costs (the danger of injuring the pedestrian). By
capturing the problem of the pedestrian and the motorist in the
form of a two-by-two game, however, not only are the incentives of
the motorist made manifest, but we can readily understand how a
change in the legal rules alters the incentives of the motorist and the
pedestrian at the same time.
To see this, consider the legal regime of negligence coupled wi th
contributory negligence. This is the regime that Anglo-American
law has embraced for a long time. Under this regime, the pedestrian
can recover only if the motorist is negligent and if the pedestrian is
not. This rule of law leads to the normal form game set out in figure
:
Pedestrian
No Care
Due Care
No Care Due Care
-100, 0
-10, -100
-100, -10
-20, -10
Motorist
Payoffs: (Pedestri an, Motori st)
Figure : Negl i gence wi th contri butory negl i gence
Compare figure with figure a. The two figures are identical except
in the box in which the pedestrian exercises due care and the mo-
torist fails to do so. I n this event, the motorist rather than the
pedestrian bears the cost of the accident. The pedestrian bears the
cost of the accident whenever the pedestrian fails to exercise care and
in the case in which both players exercise care. The legal rule does
not change the strategies available to the players or the sum of the
payoffs in each box. All that changes is the allocation of the cost of
the accident between the parties.
I n this game, unlike the game in figure a, the pedestrian has a
dominant strategy. The pedestrian is always better of taking care.
The motorist no longer has a dominant strategy. Whether the mo-
torist is better off taking care turns on whether the pedestrian also
takes care. I f we accept the idea of iterated dominance, however, we
can predict the strategy that the motorist will choose. The motorist
will recognize that the pedestrian will play due care and then de-
cide to play due care. Hence, under this legal regime, both pedes-
trian and motorist will take due care.
.a Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
A comparison between the two models focuses our attention on
the way in which this legal rule works and reveals a counterintuitive
insight about the role of law. The only difference between figure a
and figure is in the box representing the strategy combination i n
which the pedestrian exercises due care and the motorist does not.
I n figure a, the payoffs were $..c and $c for the pedestrian and the
motorist respectively. I n figure , they are $.c and $.cc. This
strategy combination is not the solution to either game: in figure a,
neither player exercises care, while in figure both players exercise
care. Yet it is how the negligence/contributory negligence regime
reallocates the harm when the pedestrian takes care and the motorist
does notan outcome that is not reached in either gamethat
completely alters the expected play of the game. Under either liability
rule, we would never expect to observe the pedestrian exercising due
care and the motorist exercising no care, but it is precisely how the
law treats the outcome that will not happen that determines
whether the efficient due caredue care outcome occurs. A legal rule
brings about changes through theconsequences it attaches to behavior
that never happens either when thelegal ruleis in placeor when it i s
not.
This model also focuses on a central assumption underlying the
Anglo-American rule. To believe that this rule works, we must be-
lieve both that the motorist acts rationally and that the motorist be-
lieves that the pedestrian acts rationally as well. The motorist will
take care in order to avoid liability only if the motorist believes that
the pedestrian is similarly motivated to act in a way that tries to avoid
bearing the cost of accidents and will take care as well. I f the mo-
torist believed that the pedestrian would not take care, the motorist
would not take care either. This liability rule turns crucially on the
assumption that the motorist believes that the pedestrian will exer-
cise due care.
This explicit game-theoretic approach isolates two features of
the law in a useful way. First, it makes clear the rationality assump-
tions required. We must assume not only that individuals behave ra-
tionally, but that individuals expect others to behave rationally as
well. Second, this way of looking at the problem reveals one of the
important but subtle ways in which a legal rule works. A change in a
legal rule can alter the behavior of both parties even by changi ng
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w .
outcomes that are never seen under either the new or the old
regime.

. ExtensiveFormGames and Backwards Induction

Not all games or legal situations can be resolved using domi-
nance arguments. For example, consider the problem of choosing on
which side of the road to drive. I n this country, we drive on the
right hand side, in England, on the left. Think of two players faced
with that choice in the absence of a governmental setting:
Player 1
Left
Right
Left Right
4, 3
0, 0
0, 0
3, 4
Player 2
Payoffs: (Player r , Player a )
Figure | : Driving Coordination Game (Normal Form)
Player . has a slight preference for driving on the left, player a for
the right, but both care most about making the same decision. (For
that reason, this game is often labeled a coordination game.) Neither
player has a dominant strategy nor is any strategy a dominated strat-
egy. What then is the likely outcome? There is another important
approach to solving games, though it will be of only some help here.
Consider the following idea: I f player . knew that player a were to
play left, player . would play left also, and the flipside of that is
true as well. The same is true of the combination (right, right):
player . would play right in response to player as right and player
a would play right in response to player .s right. (Left, right) and
(right, left) lack this quality: if player . chose to play left but before
committing learned that player a was going to play right, player .
would abandon left and instead play right. (Left, left) and (right,
right) have a stability that the other two outcomes lack. The game

For additional analysis of torts issues from the perspective of dominant

and dominated strategies, see Daniel Orr, The Superiority of Comparative
Negligence: Another Vote, ac J. Legal Stud. .. (..); Tai-Yeong Chung,
Eciency of Comparative Negligence: A Game Theoretic Analysis, Mimeo,
Department of Economics, Social Science Center, Univ. of Western Ontari o,
London, Ontario (.a).
. Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
theory lingo for this is that both (left, left) and (right, right) are
Nash equilibria, Nash coming from the great game theorist John
Nash. This game has two pure strategy Nash equilibria. (Pure strat-
egy is more lingo for saying that neither player is playing in a proba-
bilistic fashion.)
I n some settings, a game will have a unique Nash equilibrium
and it is perhaps understandable that such an equilibrium is consid-
ered the most natural outcome to the game. Unfortunately, as i n
figure , many games have multiple Nash equilibria and the games
themselves offer no good means for the players to coordinate on
those equilibria. As a consequence, if the game in figure were
played in an experimental setting, I would expect to see a sizable
number of non-Nash (left, right) and (right, left) outcomes. The
players would not be happy about this, as this is the worst outcome
for them, but the problem with the game is that the players lack any
good means for coordinating their choices. Sometimes player .
would hope that the (left, left) Nash outcome was going to be
played while player a would be hoping for the (right, right) Nash
outcome and that puts the players squarely on (left, right).
Subject to the Churchill caveat, legal intervention might again
be appropriate. To get at this and to introduce another form for
representing games, suppose, for example, the government gave the
first person the right to set the rules of the road. This game could be
represented in the following way:
Left
Player 2
Right
(0, 0) (4, 3) (3, 4)
Right
(0, 0)
Player 1
movesright
Player 1
movesleft
Left
Payoffs: (Player r , Player a )
Figure : Driving Sequential Game (Extensive Form)
This game represents the players choices through something akin to
a decision tree. This representation is known as the extensiveformof
a game. Figure differs from a decision tree in that it represents de-
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w .
cisions by two players, but the basic idea is the same. Pursuant to
governmental edict, player . chooses first, player a second, and each
still chooses between left and right. I n this game, player a ob-
serves player .s choice, which is the essential difference between this
game and our prior game in figure .
This game can be solved using another solution technique, back-
wards induction. I f player . moves left, player a will choose be-
tween left, with a payoff of , and right with a payoff of c.
Player a would clearly play left. I f player . moves right, player a
will choose between left, with a payoff of c, and right with a
payoff of , and hence will choose right. Player . thus faces mov-
ing left, and receiving and moving right and receiving , and
hence would move left. Legislation changing the sequence of
moves turns a simultaneous decisionmaking game into a sequential
game and establishes a clear outcome. The indeterminacy of the si-
multaneous game is eliminated. Note that the government alloca-
tion of the right to move first has distributional consequences. I n
this game, player . receives and player a gets . I f the right to move
rst were allocated to player a, player a would get and player .
would receive .
Standard setting, such as establishing the rules of the road, is a
conventional use of governmental power. The games in gures
and should make clear the possible benets associated with these
activities.
. Embedded Games: Caveat Legislator
I started the analysis with the Prisoners Dilemma, as it is easily
the best-known game and is most often invoked in defense of legal
intervention. Such an analysis often does little more than to suggest
that a particular situation has the form of the dilemma and then to
claim that intervention would be appropriate. This may be a serious
mistake. Whether a Prisoners Dilemma creates problems depends
on the larger structure in which the game exists. Put dierently, a
small game, such as the Prisoners Dilemma, may arise in a much
larger game. The very existence of the Prisoners Dilemma in the
large game may have benecial, rather than negative, consequences.
A simple example should make this clear. Consider the games set
forth in gure 6:
.6 Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
Player 1
Up
Down
Left Right
2, 2
1.5, 3
3, 1.5
2.5, 2.5
Player 2
Player 1
Up
Down
Left Right
6, 1.5
0, 0
0, 0
1, 3
Player 2
Prisoners Dilemma
Coordination Game
Payo s: (Player r , Player a )
Figure 6
Figure 6 illustrates a Prisoners Dilemma and a coordination game.
(I have changed the payos from the prior versions of these games,
but that is irrelevant here.) I n the rst game in gure 6, player . will
play up, as that is his dominant strategy. (I f player a were to play
left, player . gets a payo of a from up and a payo of .. from
down; if player a were to play right, player . would get a payo of
from up and of a. from down; up is therefore a dominant
strategy.) Players . and a are in symmetric positions in the rst game,
so player a has a dominant strategy of left. Both players have
dominant strategies, resulting in the payo of (a, a), which is worse
than (a., a.) from (right, right).
Game a in gure 6 is a coordination game, meaning here, as
before, that the game has two pure strategy Nash equilibria. The
strategy combination (up, left) is one equilibrium: if player . were to
play up, player a would want to play left, as that results in a payo
of .. rather than the payo of c obtained by playing right. And if
player a were to play left, player . would prefer up and 6 to
down and c. Thus, (up, left) forms a Nash equilibrium. A similar
analysis holds for (down, right). As before in gure , game theory
oers us little basis for choosing between these two equilibria.
Thats where the Prisoners Dilemma comes in; it will take us
two steps to get there. Start with the game set forth in gure ,:
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w .,
Player 1 movesleft
Player 1 movesright
(2, 2)
No. 1
Up
Down
Left Right
6, 1.5
0, 0
0, 0
1, 3
No. 2
Mixed Form
Payo s: (Player r , Player a )
Figure y: Embedded Coordination Game
I have embedded the coordination game from gure 6 into a larger
game. I n this game, player . makes an initial move in which player .
has a chance to decide between taking a certain payo of a or play-
ing a coordination game. I f the coordination game is played, player a
knows that player . has elected to forego the certain payo of a and
has instead chosen to play the coordination game with player a. This
coordination game is identical to that in gure 6. I n that game,
players . and a move simultaneously, and, most importantly, neither
can observe the choice of the other.
Now consider how players . and a should reason. Player a decides
whether to play left or right only after observing that player . has
moved right. Player a does not know whether player . moved up
or down, but player a should not expect player . ever to move
down after having moved right. Moving down is dominated by
any strategy in which player . moves left. Player .s maximum
payo of . in the game that follows after playing right followed by
down is dominated by the payo from playing left. Hence, if
player . moves right, player . should follow that move by moving
up. Were player . to do otherwise, player . would have adopted a
dominated strategy. Believing that others would not play dominated
strategies, player a will play left in response to player .s initial move
of right. Because player a believes player . will move up after
moving right, player a ensures a payo of .. rather than c by
moving up. Player ., recognizing that player a will move left, will
play the strategy of moving right and up and enjoy a payo of 6,
rather than one in which player . moves left and enjoys a payo of
only a. Even though this coordination game standing alone does
. Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
not have a unique solution, it does have one when it is part of a
larger game.
6
Now take the next step. Replace the solitary payo of (a, a) wi th
our Prisoners Dilemma game from gure 6:
Player 1 moves left
Player 1 moves right
No. 1
Up
Down
Left Right
6, 1.5
0, 0
0, 0
1, 3
No. 2
No. 1
Up
Down
Left Right
2, 2
1.5, 3
3, 1.5
2.5, 2.5
No. 2
Payo s: (Player r , Player a )
Figure 8 : Embedded Prisoners Dilemma and Coordination
Games
I n this game, player . moves left or right rst, and this move is
observed by player a. I f player . moves left, the Prisoners Dilemma
game is played. I f player a moves right, the coordination game is
played.
How should this game be solved? I n the same way we solved the
game in gure ,. I n the Prisoners Dilemma, each player has a
dominant strategy and a payo of (a,a) should result. I f player . were
to play left, he would obtain a. That payo is better than any
payo that can result by playing right followed by down. Hence,
player . would follow right only with up. Player a should under-
stand this and play left following player .s initial right. This
would result in a payo of 6 to player .. Player . should therefore play
right followed by up and player a should play left. This results
in payos of 6 and .., for a total of ,., the maximum available on
these particular (and cooked) numbers.
Step back and note what has happened. We started with two
games in gure 6, the Prisoners Dilemma and a coordination game.
Taking either of these as freestanding games would suggest that le-
gal intervention might be appropriate. The Prisoners Dilemma

6
This is an example of forward induction. For an introduction, see Drew
Fudenberg and Jean Tirole, GameTheory, ... (MI T Press, ..).
Ax I x:vobuc:iox :o G~:v Tnvovv ~xb :nv L~w .
plays out ineciently, and the existence of multiple equilibria in the
coordination game means we can have little condence of an e
cient outcome there. Yet bring these two games together in a single
larger game, and private decisionmaking leads to an ecient out-
come. The very existence of the Prisoners Dilemma makes it possi-
ble to coordinate on a particular Nash equilibrium in the coordina-
tion game.
The punch line here is that game structure matters, and often
matters a lot. I dentication of the game itself is of great importance.
Misidentication usually occurs when the small, freestanding game
is viewed as thegame. A modeler who focused on the interaction
captured in the Prisoners Dilemma in gure rather than the entire
game would be misled. I t is a mistake to suggest that a Prisoners
Dilemma may arise in a particular context and to use that to justify
legal intervention. The larger game structure must be understood, as
these rather stylized games suggest. The counterintuitive (at least to
me) suggestion of gure is that the existence of a scenario i n
which a Prisoners Dilemma game might arise actually helps the
players to achieve the best outcome.
All of this should introduce a level of caution into willy-nilly i n-
vocations of the Prisoners Dilemma as a basis for legislation. More
generally, it is critical to understand the context in which a particular
game occurs and the extent to which it is embedded in a larger
game.
,
Understanding that may make it clear that the very form of
the game is up for grabs. For example, the dominant theoretical
justication for bankruptcy is that creditors of the failing rm face a
collective action problem akin to that in the Prisoners Dilemma.

(This is often called the common pool problem.) One solution is a

government-created collective procedure, the modern bankruptcy
proceeding. Nonetheless, to accept that the creditors of the rm
must play the nancial equivalent of the Prisoners Dilemma is a
mistake. Together with the debtor, the creditors have an interest i n

,
For a similar point in a political science context, see George Tsebelis,
Nested Games: Rational Choice in Comparative Politics , (Uni v. of Cali forni a
Press, .c).

See Thomas H. Jackson, The Logic and Limits of Bankruptcy Law .c

(Harvard Univ. Press, .6).
ac Cnic~co Wovxixc P~vvv ix L~w a Ecoxo:ics
taking actions ahead of time to mitigate the possible harms of the
dilemma. Security interests can be understood as one important way
of completely avoiding the dilemma.

Again, the point here is that

we must understand the context in which a game would otherwise
take place. The game in gure makes this pointI hopein rela-
tively stark fashion.
. Conclusion
This talk sketches out some of the basic ideas of game theory.
There is a standard language for representing situations, giving rise
to the normal form and extensive form games, and ways to discuss
solutions, such as dominant strategy solutions, Nash equilibria,
backwards induction, and forward induction. I hope that I have
suggested a number of ways in which these ideas help us generate
counterintuitive insights about legal problems. The central lesson of
the torts example is that a legal rule brings about changes through
the consequences it attaches to behavior that never happens either
when the legal rule is in place or when it is not. I found that surpris-
ing. I found even more surprising the notion that having a
Prisoners Dilemma handy might actually help solve collective action
problems, rather than create them, and that this should make us
cautious in relying on the Prisoners Dilemma to justify legal inter-
vention. I would have found it dicult to reach either of these
points without using game theory, though there very well may be
other routes.
I return to where I started. The bar stool test demands simplic-
ity. The work of Ronald Coase, and a lecture worthy of his name,
demands both simplicity and depth. I hope that the ideas set forth
here at least come close on both scores. Nonetheless, if I have failed,
I accept no blame and instead place it squarely on the shoulders of
Dean Georey Stone. Any failings must reect the fact that I spent
too little time in bars in preparing this talk and that in turn can be
attributed to the measly research budget for it. Notwithstanding
this, I am prepared to move forward and undertake more research
and we can begin at the reception that immediately follows.

See Randal C. Picker, Security I nterests, Misbehavior, and Common

Pools, U. Chi. L. Rev. 6 (.a).
Some of the material in part an of this essay has been published be-
fore in the Law School Alumni Magazine of the University of
Chicago. See Randal C. Picker, Law and Economics I I : The
Sequel, U. Chi. L. School Record .c (Spring .). Much of the
analysis contained herein is taken from selected chapters of Douglas
G. Baird, Robert H. Gertner, and Randal C. Picker, Game Theory
and theLaw (forthcoming, Harvard Univ. Press, .). A version
almost identical to this essay will appear as the law and economics
selection for a symposium on interdisciplinary approaches to the law
in the Loyola (Los Angeles) University Law Review.
Readers with comments should address them to:
Randal C. Picker
Professor of Law
The Law School
The University of Chicago
.... East 6cth Street
Chicago, I llinois 6c6,
Cn:caco Wovx:Nc Pavvvs :N Law aNo EcoNox:cs
(SvcoNo Svv:vs)
.. William M. Landes, Copyright Protection of Letters, Diaries
and Other Unpublished Works: An Economic Approach (July
..).
a. Richard A. Epstein, The Path to TheT. J. Hooper: The Theory
and History of Custom in the Law of Tort (August ..).
. Cass R. Sunstein, On Property and Constitutionalism
(September ..).
. Richard A. Posner, Blackmail, Privacy, and Freedom of Con-
tract (February .a).
. Randal C. Picker, Security I nterests, Misbehavior, and
Common Pools (February .a).
6. Tomas J. Philipson & Richard A. Posner, Optimal Regulation
of AI DS (April .a).
,. Douglas G. Baird, Revisiting Auctions in Chapter .. (April
.a).
. William M. Landes, Sequential versus Unitary Trials: An Eco-
nomic Analysis (July .a).
. William M. Landes & Richard A. Posner, The I nfluence of
Economics on Law: A Quantitative Study (August .a).
.c. Alan O. Sykes, The Welfare Economics of I mmigration Law:
A Theoretical Survey With An Analysis of U.S. Policy
(September .a).
... Douglas G. Baird, .a Katz Lecture: Reconstructing Contracts
(November .a).
.a. Gary S. Becker, The Economic Way of Looking at Life
(January .).
.. J. Mark Ramseyer, Credibly Committing to Efficiency Wages:
Cotton Spinning Cartels in I mperial Japan (March .).
.. Cass R. Sunstein, Endogenous Preferences, Environmental
Law (April .).
.. Richard A. Posner, What Do Judges and Justices Maximize?
(The Same Thing Everyone Else Does) (April .).
.6. Lucian Arye Bebchuk and Randal C. Picker, Bankruptcy Rules,
Managerial Entrenchment, and Firm-Specific Human Capital
(August .).
.,. J. Mark Ramseyer, Explicit Reasons for I mplicit Contracts:
The Legal Logic to the Japanese Main Bank System (August
.).
.. William M. Landes and Richard A. Posner, The Economics of
Anticipatory Adjudication (September .).
.. Kenneth W. Dam, The Economic Underpinnings of Patent
Law (September .).
ac. Alan O. Sykes, An I ntroduction to Regression Analysis
(October .).
a.. Richard A. Epstein, The Ubiquity of the Benefit Principle
(March .).
aa. Randal C. Picker, An I ntroduction to Game Theory and the
Law, (June .).