Static Games of Complete Information

Raphael Boleslavsky

1/10

1
1 Static Games of Complete Information
In game theory the goal is to develop a theory of decision making in situations where final
outcomes are influenced by the actions of more than a single individual. The best decision
in such a setting depends on a decision maker’s forecast of what other decision makers will
do. The other decision makers’ best decisions are also influenced by their forecasts of what
others will do, and so on. The goal of the study of normal form games is to determine
what types of outcomes are possible when these types of decisions are made by groups of
”intensely rational” and ”self-interested” decision makers.

1.1 Definition
This definition is important. It will serve as the basic concept that we will build on in the
course. A game is a combination of three things:

1. A set of players N . Think of these as decision makers: firms, buyers, sellers, politi-
cal parties, bidders in an auction etc. This set may be finite, infinite, countable, or
uncountable, but there should be more than one player.

2. A set of strategies (or decisions, or actions) for each player, for each i ∈ N , there is
a set of strategies Si . Throughout the course, as we look at different issues in these
kinds of decision situations, we will think of the set of strategies available to players
in different (and more elaborate) ways, but for now, let’s just think of an action or
strategy si as a choice from a specific set, Si .

An outcome of the game is just a combination of strategies, (or actions) one for each
player: (s1 , s2 , ..., sN ) where si ∈ Si . To write an outcome concisely, I sometimes write
S−i to represent the choices of all players besides player i. Thus outcome (s1 , s2 , ..., sN )
can be written concisely as (si , S−i ).

3. A payoff function ui (·) for each player, which assigns a numerical payoff to each possible
outcome. Higher payoffs are better than low.

Fundamentally, our goal is to represent strategic situations using such frameworks. Obvi-
ously, whenever we go from a complex, multi-faceted situation into a mathematical repre-
sentation, some aspects of the real-world situation will be lost. The goal is not to capture
every detail of the real situation. Instead, we need to capture just enough of the interesting
features of the strategic setting to generate relevant insights.

Example: Competing for Grades.

1. A set of players N = {1,..., 10}. Represents students in class.

2
 2. A set of strategies (or decisions, or actions) for each player. A student’s strategy is a
choice of ei ∈ Ei = [0, 1]. A student’s decision represents how much time or effort to
devote to studying.

An outcome of the game is just a combination of strategies, (or actions) one for each
player. In this case, an outcome is just a list of the students’ efforts, (e1 , e2 , ...e10 ).

3. A payoff function for each player, which assigns a numerical payoff to each possible
outcome. Let’s imagine that the professor assigns ‘A‘ to the top 5 students in class,
and an ‘A’ is worth 1 to a student. The cost of effort is equal to the effort chosen.
Therefore if ei is in the top 5, then ui = 1 − ei , and ui = −ei otherwise.

Example: Bargain Game

1. A set of players N = {Seller, Buyer}

2. A set of strategies (or decisions, or actions) for each player, for each i ∈ N . The seller’s
strategy is a positive number representing a price, p∈ [0, ∞), the least he will accept
for the item. The buyer’s strategy is also a price p, the most the buyer will pay.

3. A payoff function for each player, which assigns a numerical payoff to each possible
outcome. In this case the payoff function is

p+p
if p ≥ p uS p, p = 2

p+p
uB p, p = v − 2


p<p uS p, p
= 0
uB p, p = 0

Example: Matrix Game

Two player games in which each player’s action set is finite have a convenient representation
as a table.

1. A set of players N = {1,2}. Think of player 1 as a celebrity and player 2 as an admirer.

2. A set of strategies (or decisions, or actions) for each player, for each i ∈ N.. Each player
decides whether to go to China Grill or Tap Tap for dinner; thus S1 = S2 = {C, T }

3. A payoff function for each player, which assigns a numerical payoff to each possible
outcome. In this case, let’s imagine that the admirer wants to go wherever the celebrity
goes, and if he is able to, he gets a payoff of 1, but if he is not able to he is disappointed
and gets a payoff of -1; the celebrity, on the other hand wants to avoid the admirer,

3
 and if he ends up at the same place as the admirer the celebrity gets a payoff of -1,
while if he ends up at a different place, he gets a payoff of 1. Thus

u1 (C, C) = u1 (T, T ) = −1
u1 (C, T ) = u1 (T, C) = 1

u2 (C, C) = u2 (T, T ) = 1
u2 (C, T ) = u2 (T, C) = −1

In the case of the above matrix game, consider the following simple representation

Player 2
C T
Player 1 C −1, 1 1, −1
T 1, −1 −1, 1

In this case, player 1’s choice of strategy is represents a row of the table, and player 2’s
strategy represents a column of the table. An outcome is thus a cell of the table, each of
which contains two numbers. The first number of each cell is player 1’s payoff associated
with that outcome, and the second number is player 2’s payoff associated with that outcome.

1.2 Dominant Strategy

Consider the following (famous) game, known as the prisoner’s dilemma (developed in the
1950’s as a model of an arms race between the US and the Soviet Union)

Player 2
C S
Player 1 C −5, −5 0, −12
S −12, 0 −1, −1

The story here is that two individuals are arrested (from an anonymous tip) on their way to
rob a bank. Weapons were found in the car. The district attorney isolates each prisoner and
offers each of them an option to confess (C) or stay silent (S). Neither one will know what
the other has chosen before choosing himself. The deal is that if both prisoners stay silent,
the DA will only be able to arrest them for gun possession, which carries a sentence of 1 year.
However, if one prisoner stays silent and the other confesses, the one who confesses goes free,
while the other goes to jail for 12 years (conspiracy and gun possession). If both confess,
then both go to jail, but because they cooperated, the DA promises leniency, a sentence of
5 years instead of 12.
What will happen in this situation? Many people’s first instinct is to say that both players
should stay silent, that way they only spend a total of 2 years in jail instead of (at least)
10. Remember, however, that players are intensely rational and self interested; if a player
anticipates that the other player will stay silent, would he prefer to stay silent or confess? Of

4
course, he prefers to confess, thereby going free instead of spending a year in jail. Suppose
furthermore that a player anticipate that the other player will confess... is it better to confess
or stay silent? Confessing is again better, leaving a player with a 5 year prison sentence
instead of 12. In this example, no matter what you anticipate the other player doing, it is
better for you to confess than to stay silent. Such a strategy is called dominant.A strategy
is dominant if it is the best thing you can do, no matter what the other players are doing.
Thus, your best choice of action doesn’t really depend on your forecast of the actions of the
other players.

Definition: Dominance. Strategy s1 dominates s2 for player i if for all possible choices of
the other players in the game, strategy s1 delivers a strictly higher payoff than s2 :
ui (s1 , S−i ) > ui (s2 , S−i ) , for all possible S−i .

If the statement is true only with a weak inequality, then the s1 weakly dominates s2 .

Definition: Dominant Strategy. Strategy s is a dominant strategy for player i if it

dominates all of player i’s other strategies. .

When a player has a strictly dominant strategy it is clear that the player will use it. In
the prisoner’s dilemma each player’s dominant strategy is to confess, so the only outcome
that we could anticipate in this game is (C, C) and yet (S, S) is better for both players than
(C, C). This is the hallmark feature of the Prisoner’s Dilemma.

1.3 Nash Equilibrium

Not every game has a player with a dominant strategy. When players do not have dominant
strategies, their choice of action depends on the actions they anticipate from the other
players. Consider the following example, known as the game of Chicken, based on the 1950’s
game played by teenagers.
Player 2
D V
Player 1 D −100, −100 10, −10
V −10, 10 −1, −1

If you anticipate that the other player will play D, what should you play? Obviously, you
should play V , giving you a payoff of -10 instead of -100; similarly if you anticipate that the
other player will play V , you should drive, giving you 10 instead of -1. We say that player
i’s action is a best response to the actions of the other players if, given what the others are
doing, there is no action for player i that brings him a higher payoff.

Definition: Best Response. Strategy s∗i is player i’s Best Response to the decisions of all
other players, S−i = (s1 , ..., si−1 , si+1 , ..., sN ), if
ui (s∗i , S−i ) ≥ ui (si , S−i ), for anysi ∈ Si .

5
Equivalently, s∗i solves
max ui (s1 , s2 , ..., si−1 , si , si+1 , ..., sN )
si ∈Si

The best response is not necessarily unique.

Given that a player’s forecast of the actions of the other players is correct, the only strategy
that a rational self-interested player would select is his best response (or one of them) to the
actions of all of the other players. If all players are rational, self-interested, and correctly
anticipate the behavior of others, then the only outcomes that would ever occur have the
property that every action is a best response to the actions of all other players. This idea
motivates the concept of Nash Equilibrium.

Definition: Nash Equilibrium A Nash Equilibrium is an outcome (s∗1 , s∗2 , ..., s∗N ) in which
every action is a best response to all others.

Here are three ways to think about Nash Equilibrium.

1. A Nash Equilibrium is an outcome from which there are no profitable unilateral de-
viations. Given the decisions of the other players, no player could have done better
deviating on his or her own. (It may be possible to do better by deviating as a coalition,
but this is not the focus of the course)

2. A Nash Equilibrium is a self-enforcing agreement. If an “agreement,” about how to

play the game is made (never mind how), then the agreement enforces itself: given that
each player anticipates that the others will stick to the agreement he or she also prefers
to stick to the agreement. Moreover, because this is also true of all other players, the
players beliefs that the others will stick to the agreement is also rational.

3. A process of forecasting.

• Players form some forecasts of what other players will do.

• Given her forecast about what others will do, each player does the best she can.
• All player forecasts turn out to be correct.

The first two points of this interpretation are intuitive, but the last point may seem
a bit strange. In the next set of notes we will introduce randomized strategies, which
allows us to relax the third point. Players forecasts will not be perfectly correct every
time, but they will be correct on average.

The Chicken Game of this example thus has two Nash Equilibria (in pure strategies, the
type we have been looking at) (D, V ) and (V, D).

6
2 Simple Strategic Environments
In this section we will advance our understanding of simple strategic environments by think-
ing broadly about the substantive varieties of 2x2 games. For simplicity, we will assume that,
for a given strategy of player j, player i has only a single best response. This assumption
rules out certain “special cases” that do not really add much to our understanding of these
environments.
We have seen a few different examples of 2x2 games already. To organize our thinking, I
group these games according to the number of Nash equilibria they admit: either 0, 1, or 2.1

No Nash Equilibrium. Consider the game that was introduced in Example 1, between the
celebrity and the paparazzi.
Player 2
C T
∗ ∗
Player 1 C −1, 1 1 , −1
T 1∗ , −1 −1, 1∗
This game is called “Matching Pennies.” The essence of this game one player wants to
“chase” the other, while the other wants to avoid getting “caught,” which results in “cycli-
cal” best responses that do not result in any Nash equilibrium. The idea here is that the
players mutual interests are opposed—each player benefits by correctly predicting the other’s
behavior, while being unpredictable himself. This is a fundamental aspect of many strategic
interactions: espionage, military strikes, detection of criminal behavior, passing vs. running
in American football, penalty kicks in everyone else’s football, serves in tennis, entering a
new market and many more. Because the unpredictability is essential in this strategic envi-
ronment, to understand what is going on here we must allow for randomization on the part
of the players. We take up this issue in the next set of notes.
Though I have presented a specific example of matching pennies, it is easy to see that any
2x2 game without a Nash equilibrium must have the same cyclical pattern of best responses.
Though the payoffs may be different, it is, in essence, equivalent to matching pennies.

One Nash Equilibrium. There are two substantive types of games that have one Nash
equilibrium. We have already seen the first one: the Prisoner’s Dilemma.
Strictly speaking, in the Prisoner’s Dilemma, each player has a dominant strategy, and the
resulting equilibrium is strictly worse for both players than some other outcome of the game.
In 2x2 games, we can extend this concept slightly to include all games with a single Nash
equilibrium, even if (one of) the players does not have a dominant strategy to select it. Thus,
an “extended” Prisoner’s dilemma has a single Nash equilibrium that is simultaneously worse
for both players than some other outcome of the game.
The other type of game with one Nash equilibrium is “boring.” The single Nash equilib-
rium is “good” in the sense that there is no other outcome that is better for both players
1
If a 2x2 game has more than 2 Nash equilibria, it is inherently “special” because at least one player must
be indifferent between strategies for this to occur.

7
simultaneously.
An example of the “extended” Prisoner’s dilemma:

Player 2
C S
∗ ∗ ∗
Player 1 C −5 , −5 0 , −12
S −12, 0 −1, 1∗

Note that player 1 has a dominant strategy, while 2 does not. In general, in a 2x2 game
with a single Nash equilibrium, at least one player has a dominant strategy. By playing with
the best responses, it is easy to convince yourself that this is the case. This observation will
appear in an example later in the class.
The Prisoner’s dilemma represents situations in which strategic considerations lead to an
undesirable outcome. In the original application, we would all be better off without nu-
clear weapons, but the unique Nash equilibrium (in dominant strategies) has both countries
developing them.
An example of the “boring” game:

Player 2
C S
Player 1 C −5∗ , −5∗ 0∗ , −12
S −12, 0∗ −10, −10

This game has a unique equilibrium in dominant strategies, and no other outcome is better
for both players. Note (C, S) is better for Player 1, but worse for Player 2.

Two Nash Equilibria. A minute of experimenting with the best responses makes clear that,
given our assumption on the payoffs, a 2x2 game with two Nash equilibrium can have only
one pattern of best responses, with the Nash equilibria “on a diagonal.” Although the struc-
ture of best responses is always the same in such games, there are actually two different
strategic environments that can be represented by such games.
Consider the following game, sometimes called the “stag hunt” or a “coordination” game.

Player 2
S H
∗ ∗
Player 1 S 5 ,5 0, 1
H 1, 0 1∗ , 1∗

In this game, the players choose whether to hunt Stag (S) or hare (H). If both players hunt
stag, then they catch it, and both receive a high payoff. But, if only one hunts stag and the
other hunts hare, the one who went for stag gets a very low payoff. Meanwhile, hunting hare
gives a medium payoff, regardless of the choice of the other player. Thus, the game has two
Nash equilibria: in one both players hunt stag, in the other both hunt hare.
The two Nash equilibria of the Stag hunt have the feature that one of them, (S,S), is clearly
better for both players than the other equilibrium (H,H), and yet, (H,H) itself is also a

8
legitimate Nash equilibrium. The “hare” equilibrium comes from a “bad belief” about the
actions of the other player (that he intends to hunt hare) that motivates me to hunt hare
and becomes self-enforcing. For this reason, the (H,H) equilibrium is sometimes referred to
as a “coordination failure.”
The main feature of a stag hunt is that the game has two Nash equilibria that can be ranked:
one is better for both players simultaneously than the other.
Intuitively, this game is about cooperation. Each player prefers cooperation (S,S), but each
player also doesn’t want to be the only one to cooperate by choosing S. Another way to
think about this game is that the best responses reveal a desire to coordinate—if the other
player plays S (cooperates) then I would like to play S. But if the other player plays H,
then I would also like to play H. Thus, each player would like to coordinate (match) his or
her choice to that of the other player.
In the other type of 2x2 game with two Nash equilibria, the equilibria cannot be ranked in
this way. For example, the Chicken game has two Nash equilibria (D,V) and (V,D), but
each of these equilibria is good for one of the players and bad for the other. It is not the
case that one equilibrium is better for both than the other. While the Stag Hunt is about
cooperation—both players choosing to hunt stag—the chicken game has both cooperative
and competitive aspects. On one hand, the player who is supposed to veer would rather stick
to the equilibrium than deviate by driving, which would result in a crash and a large negative
payoff. Thus, once the agreement is formed complying with it is a form of cooperation. On
the other hand, each player would prefer the “self-enforcing agreement” in which he or she is
the one to drive, which introduces an element of competition into this strategic environment.

3 Applications and Examples

In this section we will use the tools we have developed so far, Normal Form games, and Nash
Equilibrium to analyze some real world situations in industrial organization, politics, and
social issues. First turn to the most straightforward application for economics, industrial
organization.

3.1 Models of Firm Competition

In this section I will present and analyze several important models of firm competition. The
first model is the Cournot Model of firm competition.

Cournot Competition: Does competition favor consumers?

Imagine a farmer’s market to which two farmers bring their cucumbers. Cucumbers from
different farmers are indistinguishable, so that the market clearing price for a cucumber
will depend only on the total number of cucumbers that have been brought to the market.
Suppose that p = 1 − Q where Q = q1 + q2 . Because the market price is the price at which all
of the cucumbers will sell, each firm will sell all of its cucumbers, so that each firm’s profit

9
is given by
ui (q1 , q2 ) = (1 − q1 − q2 ) qi − cqi

where c represents the cost of growing and transporting each cucumber.To find farmer one’s
best reply to the other’s quantity solve

max u1 (q1 , q2 )
q1

max (1 − q1 − q2 ) q1 − cq1
q1

Simply rewriting this equation tells us immediately that

u1 (q1 , q2 ) = (1 − q2 − c) q1 − q12

so that farmer 1’s payoff is a quadratic function, with horns pointing down, with one root
at zero and the other root at (1 − q2 − c). The maximum should therefore occur halfway
between the roots at (1 − q2 − c). Indeed this is borne out by the first order conditions

(1 − q2 − c) − 2q1∗ = 0
1 − q2 − c
q1∗ =
2
Because the problem is identical for farmer 2, you can see immediately that farmer 2’s best
reply is symmetric so that
1 − q1 − c
q2∗ =
2
To solve for the Nash Equilibrium quantities, we need to find a pair of quantities (q1 , q2 ) for
which
1 − q2 − c
q1 =
2
1 − q1 − c
q2 =
2
Since these linear equations are symmetric, you can suppose that there is a single quantity
that satisfies them

q1 = q2 = qb

1 − qb − c
qb =
2
q = 1−c
3b
1−c
qb =
3
2
Q = 2b
q= (1 − c)
3

10
In this example, if a large corporation were to come in and buy both farmers, the corporation
would produce the monopoly output, 12 < 32 , the Cournot quantity. A higher quantity
produced leads to a lower price and more consumers purchasing cucumbers. In this model
competition benefits consumers and hurts firms; allowing a corporation to purchase the farms
is a bad idea from the perspective of consumers.
Conceptually, the Nash Equilibrium is the intersection of the best responses.
More generally, with N firms, the Nash Equilibrium quantity will be
1−c
q=
N +1
:
leading to a total industry output of
N
Q = Nq = (1 − c)
N +1
obviously as N → ∞ this approaches the perfectly competitive outcome.

Differentiated Duopolists: Does competition always favor consumers?

Suppose that the day after the farmer’s market, there will be a contest for the best pie in the
county. Consumers have come to the farmer market to purchase fruit for the filling of their
pies. We are interested in strategic interactions between a strawberry farmer and a blueberry
farmer at the farmer market. Unlike the case of cucumbers, which were indistinguishable,
strawberries and blueberries are certainly distinct. Furthermore, an increase in the price of
strawberries can lead to an increase in the amount of blueberries that consumers want to
buy. After all, a consumer that originally planned to make a strawberry pie may decide to
switch to blueberries if the price of strawberries goes up. More interestingly, an increase in
the price of strawberries can also cause the demand for blueberries to drop, if say, a consumer
who was planning to make a mixed berry pie decides to make, say a peach pie instead. To
capture this notion,. let’s say that the demand for each farmer’s berry, depends on the price
he posts and the price chosen by the other farmer
qS (pS , pB ) = 1 − pS + αpB
qB (pS , pB ) = 1 − pB + αpS
−1 < α < 1

From these formulas you can see that demand for each farmer’s berry goes down when his
price increases, but can either go up or down (depending on the sign of α) when the other
farmer’s price increases. Of course the influence of the farmer’s price on the quantity he sells
is bigger than the influence of the other farmer’s price, so that |α| < 1. For simplicity let’s
ignore the farmers’ production costs, so that each farmer’s profit is given by

uS (pS , pB ) = (1 − pS + αpB ) pS
uB (pS , pB ) = (1 − pB + αpS ) pB

11
Let’s solve for the Nash Equilibrium prices. To find the strawberry farmer’s best response
to pB

max (1 − pS + αpB ) pS
pS

Similarly to the case above,

(1 − pS + αpB ) pS = (1 + αpB ) pS − p2S

→
1 + αpB
p∗S =
2

Again, the blueberry farmer’s best response is just the same, but reversed

1 + αpS
p∗B =
2
To solve for the Nash Equilibrium prices,
1 + αpB
pS =
2
1 + αpS
pB =
2

because the system is symmetric, it is reasonable to conjecture that in Nash Equilibrium,

pS = pB = p
1 + αp
p =
2
1
p =
2−α
1
so that pS = pB = 2−α is the Nash Equilibrium of this game. Now let’s imagine that a
corporation is interested in purchasing the blueberry and strawberry farms. Would allowing
the corporation to purchase the farms benefit consumers? To answer this question, first let’s
figure out what the prices the corporation would set if it owned both farms. If a single firm
owned both farms, it would set prices to make as much money in total as possible,

max (1 − pS + αpB ) pS + (1 − pB + αpS ) pB

pB ,pS

The first order conditions for this problem are

d
((1 − pS + αpB ) pS + (1 − pB + αpS ) pB ) = 2αpB − 2pS + 1 = 0
dpS
d
((1 − pS + αpB ) pS + (1 − pB + αpS ) pB ) = 2αpS − 2pB + 1 = 0
dpB

12
Because these equations are obviously symmetric, it is easy to see that pB = pS = p so that

2αp − 2p + 1 = 0
1
p=
2 − 2α

so that under corporate ownership the price of strawberries and blueberries will be the same,
1
and will equal 2−2α
In the following graph, the red curve represents the price of each berry under a competitive
(Nash Equilibrium) situation, while the black line represents the price of each berry under
corporate ownership. So we see right away that when α > 0 (the goods are substitutes)
keeping the market competitive benefits consumers, but when α < 0, so that the berries are
complementary (an increase in the price of the other berry leads to a drop in demand for
my berry), corporate (centralized) ownership leads to lower prices and benefits consumers
relative to competition.

Spatial Competition:
Have you ever noticed that fast food restaurants, drugstores, and fast food
restaurants appear in clusters?
The town of Surf City, NC is on a barrier island, and has just one street a mile long, which
we will represent by the line interval [0, 1]. A person lives at each point of the interval. Two
drugstores are planning to locate themselves in this town, and can choose any location in the
town, represented by a point x ∈ [0, 1]. The drugstores are identical in every respect, except
in their location. The citizens of the town, therefore will visit the drugstore that is closer
whenever they need to buy anything. If the drugstores locate at the exact same spot, then

13
each consumer has to spend the same effort to reach either drugstore, and chooses randomly,
so they split the market equally, whenever they choose the same spot.
It becomes clear that in any Nash Equilibrium the firms must locate at the same point.
Whenever the firms locate away from each other, say at points, a, b, with a < b, the firm at
a captures everything to the left of a and the firm at b captures everything to the right of b,
but they split the consumers between a and b down the middle. Given that the other firm is
at b, the firm at a could benefit by moving toward b; in doing so it would grow the size of the
market to its left that it captures for sure, and shrink the market in the middle that it splits,
improving its profits in both ways. Therefore, any time that the firms locate at different
positions, there are profitable deviations. It must be then that in the Nash Equilibrium, the
firms locate at the same point.
It also becomes clear that the only point at which both firms could locate that is a Nash
Equilibrium is 12 . Suppose both firms located at some other point x > 21 . By locating at the
same point the firms split the market equally, but by deviating slightly toward the center
(by ε), one firm could have a market on its left of size x − ε. Since x > 12 , there is always an
ε for which x − ε > 12 . The only possible point that could be a nash Equilibrium for both
firms to locate is 21 , and it is easy to see that there is no profitable deviation from 12 .
Suppose you were the mayor of the town and could dictate to the drugstores where they
should locate, in order to reduce the total travel time for your citizens. Where would you
locate the drugstores? Suppose you decide to locate the drugstores at points a, b with a < b.
The travel costs for the citizenry can be computed in four pieces
Ra
1. Consumers to the left of a 0
(a − x) dx = 12 a2
R a+b
2. Consumers between the firms, closer to a a 2 (x − a) dx = 18 (a − b)2
Rb
3. Consumers between the firms, closer to b a+b (b − x) dx = 18 (a − b)2
R1 2
4. Consumers to the right of b b
(x − b) dx = 21 (1 − b)2

leading to a total transport cost of

 
1 2 2 1 2
a + (1 − b) + (a − b)
2 2

The first order conditions for minimizing this are:

1
(3a − b) = 0
2
1
(3b − a − 2) = 0
2

It is easy to verify then that

1 3
a∗ = , b∗ =
4 4
will minimize the total travel cost. Of course, it’s not surprising that the mayor would want
the firms to be spread out some; when both firms locate at the center, from the point of view

14
of travel cost it’s as if there were only one firm! Intuitively, firms want to be close together
to avoid splitting the market with the other firm; by moving closer together, each firm can
capture some consumers that were previously going to the other firm. The mayor, on the
other hand, wants some spread; by spreading the firms out some he is able to reduce the
travel costs for the citizens at the extremes. Interestingly, the firms do as well under the
mayor’s plan as in the competitive situation. If the mayor could really enforce his proposal
legally, the drugstores would be perfectly willing to go along with it, but If the mayor could
not enforce the proposal legally, and could only make a non-binding proposal, then each
drugstore would cheat if it believed that the other firm would follow the arrangement.

Consumer Privacy: Whom does it help?

Consider the description of Surf City (above). Imagine two stores are located at the endpoints
of Surf City, one at location zero and the other at location one. These stores carry the same
inventories, but may charge different prices for an item. In order to travel to one of the stores,
a consumer must expend effort, equal to the distance that he has to travel. All consumers
shop at one of the two stores. Each consumer will choose to shop at the store that offers her
a better deal, accounting for the travel cost and prices. A consumer located at location x
will spend p0 + x to shop at store zero, and p1 + (1 − x) to shop at store two.
Privacy. Suppose that this market operates under a regime of privacy, so that firms must
simultaneously post a single price, (p0 , p1 ) for the good and any consumer can purchase the
good at this price if he or she wishes. Given prices (p0 , p1 ), the consumer location x that is
indifferent between the stores can be found by solving
p0 + x = p1 + (1 − x)

1 p0 − p1
x = −
2 2
All consumers to the left of x shop at store zero, all to the right shop at store one. Therefore
the profit function for firms is
 
1 p0 − p1
u0 (p0 , p1 ) = p0 −
2 2
 
1 p1 − p0
u1 (p0 , p1 ) = p1 −
2 2
To find the best response of firm zero, maximize firm zero payoff, keeping p1 fixed.
  
d 1 p0 − p1 1
p0 − = (1 − 2p0 + p1 ) = 0
dp0 2 2 2
1 1
p0 = + p1
2 2
To find a symmetric equilibrium, set p0 = p1 = p and solve:
1 1
p = + p
2 2

p = 1

15
Each consumer purchases from whichever store is closer. Consumers with x < 1/2 spend a
total cost (1 + x) shopping, while consumers with x > 12 spend a total cost 1 + (1 − x) =
(2 − x) shopping. The worst-off consumer is located at x = 21 , who has to travel far to arrive
at either store. Each firm’s profit is 12 .

In this figure the solid red lines represents the total shopping cost for the consumer located
at x. It is composed of two pieces: the price, p = 1, represented by the horizontal brown
line, and the travel cost: x for consumers at locations less than 21 and 1 − x for consumers
at locations greater than 12 . Firm 0’s profit is the area bounded by the horizontal brown line
(representing price) and the vertical brown line (representing the indifferent consumer—firm
0’s market share is to the left).
Non-Privacy. Now suppose that the market operates without privacy, so that the stores can
observe a customer’s address x and make individualized price offers to customers based on
this information. Imagine that each customer receives a coupon in the mail from each store,
offering a certain price for the good, but only the named customer can use the coupon. If
firms make offers (px0 , px1 ) to customer x, he or she will purchase from whichever firm offers
a better deal.
px0 + x vs. px1 + (1 − x) ,
that is, the consumer buys from

Firm 0 px0 − px1 < 1 − 2x

Firm 1 px0 − px1 > 1 − 2x
Either px0 − px1 = 1 − 2x.

16
The easiest way to proceed is to consider the possible cases systematically. Suppose x <
1/2—the reverse case is similar.
(1) Could we have an equilibrium in which px0 − px1 > 1 − 2x? In this case the consumer buys
from firm 1. Firm 1 can increase its price a little and still attract the consumer, thereby
increasing its profit.
(2) Could we have an equilibrium in which px0 − px1 < 1 − 2x? In this case the consumer buys
from firm 0. Firm 0 can increase its price a little and still attract the consumer, thereby
increasing its profit.
(3) Could we have an equilibrium in which px0 − px1 = 1 − 2x, and px1 > 0? In this case, the
consumer is indifferent about where to shop. If the consumer shops at firm 0, then firm 1
can drop its price a little (because px1 > 0) and get the consumer to shop at firm 1, thereby
increasing profit. If the consumer shops at firm 1, then firm 0 can drop its price a little
(because px0 > px1 + 1 − 2x > 0) and attract the consumer, thereby increasing its profit.
(4) Could we have an equilibrium in which px0 − px1 = 1 − 2x, and px1 = 0? In this case, the
consumer is indifferent about where to shop. If the consumer shops at firm 1, then firm 0 can
drop its price a little (because px0 = 1 − 2x > 0) and attract the consumer, thereby increasing
its profit. But if the consumer shops at firm 0, then firm 1 cannot drop its price to attract
the consumer. Furthermore, by increasing its price, firm 1 does not improve, because the
consumer would continue to shop at firm 0. Similarly, firm 0 cannot benefit from increasing
its price, because then the consumer would switch to firm 1. Firm 0 cannot benefit from
dropping its price, since by doing so it would simply reduce the amount that the consumer
pays.
Thus, the unique Nash equilibrium is

px0 = 1 − 2x px1 = 0 for x < 1/2

px0 = 0 px1 = 2x − 1 for x > 1/2

As with privacy, each consumer buys from the closer firm. However, the price depends on
the consumer’s location. It is equal to the consumer’s ”cost of switching” to the other firm,
assuming her or she could get the good for free (x < 21 , then the cost of shopping from 0 is x
and from 1 is 1 − x, so cost of switching is 1 − x − x = 1 − 2x). Consumers that are close to
a firm have a high cost of switching, and therefore they pay a high price to the nearby firm.
Consumers that are in the middle can switch relatively easily from one firm to the other,
and so they pay lower prices.
Contrast. In the following figure the blue lines represent the price paid by consumer x under
transparency. Each firm’s profit with transparency is the area under the green line, which
equals 14 . The solid blue lines represent the consumer’s total shopping costs: consumers with
x < 12 have a total shopping cost of (1 − 2x) + x = (1 − x) while consumers with x > 12
have total shopping cost (2x − 1) + (1 − x) = x. The red and brown lines represent the total
shopping cost and the price under privacy, respectively.

17
A few things jump out. Shopping costs are reflections of each other through the line y = 1.
Obviously total shopping cost is higher with privacy. At the same time, the distribution of
shopping costs is reversed: under privacy consumers far from the stores (in the middle of
town) lose, while consumers close to stores gain. Without privacy, consumers in the middle of
town have the smallest shopping costs because competition for them is most-fierce. Privacy
benefits consumers on the ends of the town, while non-privacy benefits those in the middle.
Clearly, firm profits are higher under privacy 21 (area in the dashed box) versus 14 (area in the
dashed triangle). Under privacy, firms can maintain high prices for all consumers, but under
non-privacy, the firms must compete for consumers who are in the middle of the town by
offering them lower prices. What stops both firms from not collecting consumer information
and implementing the outcome that they both prefer? Without doing any analysis, we can
conclude that if one firm doesn’t collect information, then the other firm will benefit by
doing so (if it is not too costly). After all, collecting information allows the firm to make
custom price offers to individual consumers. If the firm wanted to, it could simply make all
consumers the same offer, which would allow it to generate the same payoff it would have
obtained without collecting anything. Thus, collecting information cannot reduce the firm’s
profit. Furthermore, collecting information allows the firm to make custom offers, increasing
its market share. In particular, note that with a fixed price (no information), each firm will
have some consumers. But if the informed firm reduces its price slightly for the consumers
who are almost indifferent, it can steal them away. Therefore, both firms might wish that
they could agree not to collect information, but collecting information is better than not
collecting it—a prisoner’s dilemma.

18
3.2 Political Campaign Advertising
Imagine that two politicians (or political parties) are running for office. Each politician
views being in office as a prize worth 100 (million), and is willing to devote money and other
resources toward being elected. In this example, simply devoting a large amount of resources
to the campaign will not guarantee that you will be elected, but the more resource you devote
relative to your rival, the greater the chance you will be elected. For simplicity, let’s say that
if politician X devotes x resources and politician Y devotes y resources, then the probability
x
of X winning the election is x+y , so that a politicians share of the total resources spent is
also his probability of winning. The payoff functions are therefore
x
uX (x, y) = 100 −x
x+y
y
uY (x, y) = 100 −y
x+y

Consider politician y best response to x

y
max 100 −y
y≥0 x+y

The first order condition for this problem is given by

x
100 −1=0
(x + y)2
√
y ∗ = 10 x − x

Similarly
√
x∗ = 10 y − y

19
It is easy to see from the figure, and not difficult to solve to find x∗ = y ∗ = 25.
√
x = 10 y − y
√
y = 10 x − x

x = y=s

√
s = 10 s − s
√
2s = 10 s
4s2 = 100s
4s = 100
x = y = 25

The Nash Equilibrium expenditure levels for both candidates is the same, and is 25(million),
or one fourth the total value of the prize. Because the expenditure level for each candidate
is the same, each candidate is equally likely to win the election, despite the fact that each
is committing a fairly significant portion of the total reward of winning toward doing so.
Candidates are therefore caught in a type of prisoner’s dilemma, if they could both somehow
commit to spend significantly less, say 21 million, they would save themselves (and society
as a whole) 24.5 million without changing the outcome of the election; however, because the
incentives to increase spending when the opponent spends so little are so great, enforcing
this type of spending limit would require significant monitoring and oversight. In this simple
model, in fact, the best thing from society’s perspective that results in the same outcome
as the competitive situation, would be to avoid the election at all and just flip a fair coin to
decide the winner.
It is also interesting to observe that the equilibrium expenditure on the election, 25 is the
maximum of the best response function. In other words no matter how much one candidate
anticipated that the other candidate would spend, the most the candidate would spend to win
the election is 25. The equilibrium therefore involves the largest amount of expenditure on
the election that could ever be consistent with rational self-interested candidates; competition
in this model is the most intense that it could be.

3.3 Externalities, Freerider Problems, and Public Goods

There is a basic (intuitive) principle in economics which says that whenever an individual
does not bear the entire cost of his actions, the individual will engage in the action more
than he would if he bore the full cost of his action, and conversely, whenever an individual
does not appropriate the full benefit of his actions, he engages in these actions less than he
would otherwise. These unborne costs or unappropriated benefits are sometimes referred to
as externalities. We will return to this idea several times throughout the course, but for now
we will see this idea at work in some simple settings.

Example: Shall We Split the Check?

20
In Amish country, they have a particular kind of buffet known as a Smorgasbord. When
you go to the Smorgasbord, they weigh you going in, then on the way out, and charge you
a certain price per pound that your weight increased. Let’s say that two friends visit a
Smorgasbord√ together. Each friend enjoys eating, so that his payoff from eating x pounds
of food is x. This payoff is increasing in the amount of food the individual eats, but is
increasing at a decreasing rate. Each pound of food costs 1 so that if each individual pays
for his own meal his net payoff is √
u (x) = x − x

In this case each friend will choose a quantity of food to maximize his total payoff
√
max x − x
x

1
√ −1=0
2 x
1
x∗ =
  4
1 1
u =
4 4

When each friend pays for his own meal, he eats a 14 pound of food. Suppose now that the
friends agree ahead of time to split the check. In this case, each individual’s bill will depend
on the amount of food consumed by the other fellow, so that
√
x+y
uX (x, y) = x−
2
√ x+y
uY (x, y) = y−
2

Consider X’s best response to a quantity y :

√ x+y
max x −
x 2

1 1
√ − =0
2 x 2
x∗ = 1
u (1) = 0

There are a couple of interesting things here: first, the best response for player x actually
doesn’t depend on the quantity chosen by y. This means that X has a strategy.
Second, each player is choosing to eat four times the amount of food as when he was paying
for his own meal. Indeed, because he only bears half of his own cost, and the other half is
borne by the other player, each player eats much more than they would have had they borne
the full cost of their meals. When they decide how much to eat, they only account for their

21
own eating, because the amount the other player eats is not under their control, but when
it comes time to pay, they have to pay, not only for their own increased eating, but also for
the increased eating of the other player. In the end, both players are worse off by agreeing
to split the check beforehand.

Example: Cleaning the Apartment

In the previous example, we saw an example in which individuals did not fully bear the costs
of their actions. In this example, individuals appropriate the benefit of each others’ efforts.
Two roommates, Jules and Vincent , share an apartment. Each Saturday morning, Jules
and Vincent each devote efforts eJ ,eV ∈ [0, ∞) to cleaning their apartment. The total effort
spent is thus eJ + eV and the ”cleanliness” associated with a given level of total effort is
ln (eV + eJ ). Jules and Vince value cleanliness in different ways, however, so that

uV (eV , eJ ) = ln (eV + eJ ) − eV
uJ (eV , eJ ) = 2 ln (eV + eJ ) − eJ

so that cleanliness is twice as important to Jules as it is to Vincent. To find the best responses
we thus have to solve the following problem

max a ln (e + x) − e
e≥0

where x is the other player’s effort and a = 1 for Vince and 2 for Jules. The FOC for this
problem is
a
−1=0
e+x

e+x=a
e∗ = a − x

This best response effort level would be fine, except that when x > a, the best response
effort level is negative, which is not allowed. In this case, the best effort is just 0. Hence, we
can write the best responses in the following way

e∗V = 1 − eJ if eJ ≤ 1
else 0
e∗J = 2 − eV if eV ≤ 2
else 0

The easiest way to find the Nash Equilibrium in this case is just to draw the best response
functions and find an intersection:

22
The only Nash Equilibrium is for Jules to clean the apartment as much as he sees fit with
no help from Vince.

Example: The (classic) Tragedy of the Commons

The classic tragedy of the commons occurs when some resource is shared in common, and
there is some form of interference or crowding between different users of the resource, so
that increased usage by one individual diminishes the effectiveness of the resource per unit
for everyone. By slightly increasing his usage of the resource a particular user diminishes
its effectiveness, not only for himself, but for everyone, however, since any particular user
only experiences this effect on the amount of the resource he uses, and not on the usage
of everyone together, any given user does not account for the total cost of usage, and in
aggregate, the resource is overused. Lets make this notion more concrete by looking at the
following example.
A university has an internet network, which is used by N students to surf the internet or
stream music, movies, etc. Because of the way information is transmitted through computer
networks, increased usage by an individual slows down the rate of transmission for all of the
users—this is the congestion mentioned above. Suppose that each student simultaneously
chooses a network usage intensity, xi ∈ [0, ∞), which can be interpreted as the size of the
“request” generated in a unit of time by the student on the network (a request is infinitely
divisible). Each student likes to use the network, but each student also dislikes waiting for
his request to be filled. The time required to complete a student’s request depends on the

23
total size of requests made on the network. In particular, when the total size of requests is
X, the time to complete a request of size 1 is t(X) = X. Thus, wait time to complete a
request of size x is just xt(X) = xX. Let X−i = x1 + +...xi−1 + xi+1 + ...xN be the size of
requests made by all students except i. Student i’s payoff is therefore

ui (x1 , ., , , xN ) = xi − xi (xi + X−i ) = xi (1 − X−i ) − x2i .

Note that by using the network, the student increases the wait time for all requests, but he
or she only cares about how the slowdown affects his or her own requests.
To find the student’s best response,

max xi (1 − X−i ) − x2i

xi

which gives first order condition,

1 − X−i
1 − X−i − 2xi = 0 ⇒ xi = .
2
Let’s look for a Nash Equilibrium in which all students use the network equally, xi = x∗ .
Such a level of usage must be a best response to itself. Recalling that X−i is the total usage
of all students besides i, we find that X−i = (N − 1)x∗ , and hence,

∗ 1 − (N − 1)x∗ 1
x = ⇒ x∗ = .
2 N +1
Total usage is therefore N x∗ = N/(N + 1), which is close to 1 as N gets large.
Next, suppose that the university could mandate that every student uses the network at
the same fixed level, x, to maximize each student’s well-being. In this case, the university
recognizes that allowing a student more usage slows down the network for everyone. The
university’s problem is therefore

max x(1 − (N − 1)x) − x2 = x − N x2

x

which gives first order condition,

1
1 − 2N x = 0 ⇒ x = ,
2N
for a total usage of N x = 1/2. Thus, the individual students (drastically) overuse the network
because they do not account for the slowdowns that their usage generates for others.
To mitigate this problem, the university could try to implement a firm quota of 1/(2N ) per
student, but such a policy is problematic. After all, a student could argue that internet access
is essential for doing schoolwork in the modern world, and that limiting access is hurting their
academic performance. Instead the university could implement a simple payment scheme,
whereby it imposes a fee for network usage in order to discourage the students from overusing
the network. If it then rebates all of the money it collects back to students, then the students
pay nothing (on net) but benefit from the reduced usage. Suppose then that the university

24
imposes a fee of px for a request of size x, while giving each student a credit of some amount
M . In this case, a student’s best response solves

max xi (1 − X−i − p) − x2i + M.

xi

A student’s best response is therefore

1 − X−i − p
xi = ,
2
and thus the Nash Equilibrium is xi = (1 − p)/(N + 1). Thus, by setting a fee satisfying

1−p 1 N −1
= ⇒p= ,
N +1 2N 2N
the university ensures that the Nash Equilibrium and the welfare-maximizing usage levels
coincide.
To interpret the tax rate intuitively, let us compare the student’s first order condition with
the University’s. The student’s first order condition is

1 − (N − 1)xi − p − 2xi = 0

while the University’s is

1 − 2N xi = 0 ⇒ 1 − 2(N − 1)xi − 2xi = 0

This discrepancy arises because an individual student does not account for the way that a
marginal adjustment in his or her own network usage affects the payoffs of other students.
Thus, he or she underestimates the total cost of slowing down the network by (N − 1)xi ,
(the per-unit download time increases by xi for N − 1 other students). Notice that if we set
p = (N −1)x̄, that is, we set the tax equal to the marginal damage that an individual student
imposes on others at the University’s preferred usage level, then the solution of the student’s
first order condition will agree with the University’s. In essence, this tax rate forces the
student to account fully for the damage that his network usage imposes on others, aligning
the student’s individual objective with the University’s.

25