Professional Documents
Culture Documents
2
Tools I: An Introduction
to Game Theory
In football, everything is complicated by the presence of the opposite team. (French original: Au football,
tout est complique par la presence de lequipe adverse.)1
O U T L I N E
We are faced with a situation that seems to reveals a whole new layer in the social system
be both simple, almost trivial, and difficult to that is investigatedthe layer of strategies and
resolve, thus complex. There are only two indi- their consequences. This chapter will give
viduals involved (A and B), they can choose a basic introduction to the modeling with stra-
between only two options (to wait or to enter tegic games, an introduction to game theory.
the doorway), their objective is the same: to More formal concepts and advanced methods
resolve the situation and pass the doorway will be discussed in Chapter 8.
either A first or B firstit does not even mat-
ter who passes first.
When faced with the situation it may seem
like a game: choose a strategy and hope for 2.2 UNDERSTANDING
some luck. Depending on what the other per- A STRATEGIC GAME
son does you win (i.e., you are able to pass) or
you lose. And this idea is not at all wrong. To distinguish our model games from the
Many interactive situations, in fact, most of the common sense term for recreational games,
social sphere may be conceptualized as one in we may specify our models as strategic games.
a family of strategic games, each with its own A strategic game is characterized by a set of
rules. There are games in which the players participants, a set of behavior options for each of
share a common objective like in the present the participants, and a set of rules as well as the
example and games where they try to exploit information available to each of the participants.
each other. There are games that offer an obvi- The participants in a strategic game are com-
ous best choice option to all those involved monly called agents; the options among which
and games where the resolution is tricky and an agent chooses are referred to as her strategies.
full of uncertainty (like in the present case). There are many possible forms of games;
There are games with only two agents and some of the more prominent ones, including
only two options per agent and those with mil- repeated games and sequential games, will be
lions and millions of participants and an infi- explained in Chapter 8. One form, however,
nite continuum of possible strategies. There is as universally applicable and powerful as
are games which may be played repeatedly a model as it is simple and easy to use as
and games that only offer one single try. a theoretical concept: simultaneous 2-person
It is obvious that in an interactive world, it 2-strategy normal-form games. Common to this
is always a simplification to reduce a social sit- group of strategic games is that the game
uation to one game. The game can merely be a involves two agents each of which chooses
model; it is a way to analyze the structure that between only two strategies and the choices are
lies beneath the social interaction, its possibili- to be made simultaneously without any possi-
ties and opportunities, the development paths bility for collusion between the two agents.
of the interaction, less likely and more likely
outcomes. A word of warning should not be Player B
foregone: Oversimplification results in inaccu- Pass Wait
rate models; assuming a strategic game as a 0 1
Pass
model (or as part of a model) implies a num- 0 1
Player A
ber of other assumptions (such as rational Wait
1 0
decision making) that are not always justified. 1 0
The skilled and careful use of strategic games FIGURE 2.1 Anti-coordination game; payoffs of the
in models of social interactions, however, row player (A) highlighted in bold face.
BOX 2.1
The game introduced in Figure 2.1 is a 2.3 THE INVISIBLE HAND AND
member of this group. We can write it as THE FALLACY OF AGGREGATION
shown in Figure 2.1. The strategies of player A AGAIN
are given in the rows, those of player B in the
columns; for 2 3 2 strategies there are four pos- In Chapter 1, we introduced two general
sible outcomes for which the payoffs of both concepts of economic modeling, the invisible
players are given (see Box 2.1 for details on the hand and the fallacy of aggregation. Both tell
concepts utility and payoff in game theory). illustrative stories of the direct interaction of
Payoffs of the row player (A) are given in the rational agents. The invisible hand describes a
lower left (bold), those of the column player situation where every agent chooses the
(B) are given in the upper right of the respec- socially optimal option out of her incentive to
tive field. This matrix notation is commonly maximize individually. Say, every agent has
used to write strategic games. an option to contribute (costs: 1) or not to con-
The view obtained from writing the game in tribute (zero cost) to a public good and
matrix notation (Figure 2.1) is the following: receives three times her own contribution (ben-
As said, both agents try to pass through the door, efit: 2 or 0 respectively) and twice the contribu-
one first, then the other. There are two solutions tion of the other player(s) (2 or 0 respectively);
that allow to accomplish this end: A waits and B see Figure 2.2.
passes or B waits and A passes. There are also The concept of optimality in interactive
two less fortunate outcomes namely one where situations, specifically the Pareto optimum, has
both agents attempt to pass and another one been introduced in Chapter 1. It is easy to see
with both agents waiting for the other one to that the game will arrive at the socially opti-
pass and neither of them actually attempting to mal (and single Pareto-optimal) point (4,4).
move through the doorway. Consequently, this model justifies ignoring the
Game theory offers a number of approaches micro-level of direct interactions (since every-
to solve strategic games and make predictions one chooses equivalently and optimally). A
regarding outcomes. Before these methods are different story is told by the fallacy of
discussed, however, the following section will aggregation.
illustrate that strategic games may be used to Say, as before, the agents choose whether to
model a broad range of concepts and phenom- contribute or not; but this time, they only
ena in interactive economics. receive twice the payoff of their opponents.
1 X
Y
0 X
X
1 X
W
2
2 1 0 1 2 3
Payoff player A
Player B Player B
Strategy 1 Strategy 2 Strategy 1 Strategy 2
1 2 0 1
Strategy 1 Strategy 1
1 1 0 1
Player A Player A
1 0 1 0
Strategy 2 Strategy 2
2 0 1 0
FIGURE 2.5 Pareto optima in a prisoners dilemma FIGURE 2.6 Best answer strategies (payoffs underlined)
game. (Crossed payoffs indicate Pareto dominated outcomes; in an anti-coordination game.
all other outcomes are Pareto optima.)
Hence, the concept of dominance is not helpful
Obviously in the prisoners dilemma game in predicting outcomes.
above, neither player would be inclined to play Still, it is clear which outcomes are prefera-
their second strategy. Any possible payoff they ble and thus to decide which strategy the
could obtain with this strategy is less than the agents choose given the choice of the other
alternative given the other players choice. Such player. This concept is termed the best answer
a strategy is called a strictly dominated strategy. strategies. If a strategy 1 of player A leads
No rational agent will ever make this choice. player B to choose her strategy 2 then strategy
A strategy for which every payoff (given the 2 of B is called the best answer (or a best
choice of the opponent) is better than all alter- answer since there may be several strategies
natives is called a strictly dominant strategy. with equal payoffs) to strategy 1 of A. Every
player has at least one best answer strategy
corresponding to every choice of the other
player(s). It is helpful to identify all best
2.5.2 Best Answers answer strategies to all the strategies of all the
Returning to the anti-coordination game players in a game: Underline the payoff the
(Figure 2.1), we find that there are no domi- player hopes to achieve as a best response to
nant or dominated strategies in this game. the choice(s) of the opponent(s) (Figure 2.6).
2.5.3 Nash Equilibria more complex strategies since they may con-
tain instructions for more than one period
An outcome that is achieved as a result of and since they may also take the opponents
every player playing her best answer strategy is actions and reactions into consideration.
called a Nash equilibrium named in honor of Repeated games may be repeated for a certain
game theorist and Nobel memorial laureate John time or indefinitely; they are also called super-
Nash. Every involved strategy is the best answer games, particularly in case of indefinite repe-
to the strategies of the other player(s), thus no tition. These different types of games also
player has an incentive to unilaterally deviate require refinements of some of the game-
from the Nash equilibrium. It follows that theoretic concepts; these will be detailed in
a Nash equilibrium isfor rational players Chapter 8.
an absorbing state, an outcome that is dynami-
cally stable (if there were the option that one or
more players revise their choices), an outcome 2.7 SUMMARY
that is more probable than others.
To obtain the set of Nash equilibria in a This chapter offered a basic understanding
strategic game, underline the best answer of what a strategic game is, how it is used in
payoffs and identify the outcomes for which modeling interactive situations, and how the
all payoffs for all players are underlined. central concepts of basic game theorybest
Figure 2.6 does this for the anti-coordination answers, dominance, Nash equilibria, and
game discussed earlier in this chapter: the Pareto optimalityare applied. Game theory
Nash equilibria are the lower left and upper is only one in a number of methods that may be
right fields, the outcomes in which the players used to analyze and understand economic
choose different strategies. interactions and their consequences. Until very
recently, a large part of the profession of
economics chose to forego using game theory
2.6 HOW MANY though this is now slowly but steadily chang-
GAMES CAN WE PLAY? ing. This was because direct interactions are
very difficult to fit into general equilibrium
So far we restricted the considerations in models which in turn made it possible for the
this chapter to a specific type of games: Games first time to analyze the economy as a whole
that are limited to exactly one time period and the nontrivial interdependence between
in which all agents act simultaneously. These different sectors, still one of the most impor-
games are referred to as normal-form games tant accomplishments in the history of eco-
(because they are conveniently represented in nomics. However, this came at the cost of
the matrix form, also called normal-form, as sacrificing heterogeneity and direct interaction
shown above). However, many different types and reducing the social sphere to an agglomer-
of games are conceivable. The agents may ation of homogeneous agents. The elaborate
choose strategies sequentiallyin this case, the models of perfect markets in effect shifted the
second player knows the decision of the first attention away from strategic interactions to
one when she chooses her strategy (sequential another part of economic reality. The market
games). Normal-form games may be played perspective will be explained in Chapters 5
repeatedly, which means that the agents through 7 while Chapter 8 and the following
remember choices and outcomes from previ- chapters explain more advanced methods for
ous periods (repeated games). This allows the analysis of interactive economies.
A broad perspective on interactive econom- (Figure 2.3) and the social optimum
ics using and applying the methods explained game (Figure 2.2)
in this chapter will be detailed in Chapters 3 c. whether there are strictly dominant
and 4. strategies in the social optimum game
(Figure 2.2)?
Chapter References 2. Consider the following game and identify
Pareto optima, best answer strategies, and
Hargreaves Heap, S.P., Varoufakis, Y., 2004. Game Theory:
Nash equilibria. Are there strictly dominant
A Critical Introduction. Routledge, London, New York.
strategies?
1. Identify: Player B
a. the Pareto optima in the anti- Strategy 1 Strategy 2
coordination game (Figure 2.1) and in 0 0
Strategy 1
the social optimum game (Figure 2.2) 0 0
Player A
b. the best answer strategies and Nash Strategy 2
0 0
equilibria in the prisoners dilemma 0 0