Columbia University

Graduate School of Business

Managerial Economics (B6006)

Classes 18-24

Strategic Competition I:
Simultaneous Games and Dominant Strategies

I. Introduction: Game Theory, or The Logic of Strategic Interaction

Up to this point in the course we looked at individuals and firms making various decisions. We
looked for the rules that will lead to the best decision making process (profit maximization), saw
how different market environments are likely to affect outcomes (industry analysis), and
examined the implications for long run profitability. In all the situations and examples that we
studies we took the behavior of other firms and individuals as given. This is not always the
correct way to approach decision making. There are various situations in which the best action to
take depends on what other individuals/firms might do, as well as how others will react to our
actions. Any situation in which what you do depends on what others do (and what others do
depends on what you do, etc., etc.) is called “a game”. And, therefore, we will call all the
participants in such interactions “players”.

The goal of this section is to introduce the language of games for the two usual purposes: to
predict the outcome of strategic competition across firms; and to understand how to formulate
the best course of action for a manager engaged in strategic competition.

Is there a game?
The first and crucial step in applying “game theory” to real life is recognizing that you are
involved in a game. The worst thing that can happen to you is to be involved in a business (or
personal) situation and not recognizing that you are involved in a game. Whether you are in a
game or not will determine what is the best course of action. Think about examples.

Here’s one. You go out for drinks with friends of yours and you know you will share the check equally at the end.
Obviously, the amount of money that you will have to pay depends not only on the number of beers you drink, but
also on those other people in your party will drink.

Another, business related example is the timing of movie releases by the major studios. If two movie distributors are
planning to release a movie they believe has the chance of being a season blockbuster, they would rather coordinate
and make sure that each movie gets out without competitors for the season. The release of two movies at the same
time is likely to affect negatively revenues for each movie, as many movie goers will select only one movie to watch
(say, on Thanksgiving day). Obviously, the release of a movie by a competitor is going to dramatically affect the
revenues and, ultimately, the profits one distributor can make.

The theory of games was first developed in simple settings such as social games (chess, cards,
and so on) because of the simplicity of the situations (rules of the game, number of players,
possible actions are all clearly understood). But the goal of game theory since its beginning was
to predict the outcome of strategic interaction in more complex, real-life situations. In particular,
one of the major ambitions of the founders of game theory –John von Neumann above all, in the
1930s and 1940s– was to be able to represent human behavior through mathematics, at least in
certain specialized, mechanical tasks. A practical application that funded much of the first
studies was robotics. The first applications were supposed to serve the U.S. war efforts during
WWII. The U.S. Navy was trying to substitute machines (robots) for humans in carrying out
high-risk missions such as the short-range torpedoing of enemy ships, and the Office of Naval
Research heavily funded von Neumann and other Princeton researchers in the development of
his ideas. Subsequently, game theory has been successfully applied to several areas of human
behavior, including the cold war arms race, the economics of firms, industrial organization,
encrypting and coding, computer networks architecture, evolutionary biology, business and
more.

Two basic types of interactions.

We will study two basic forms of interactions/games:
1. Simultaneous Games (a.k.a., “normal form games with complete information”)
2. Sequential Games (a.k.a., “Extensive Form Games”)

In both games, there is a finite number of agents participating in the game. We will call them
“players.” The players know the rules of the game. In both forms, players make choices, i.e.,
select actions, and receive prizes. We will call the prizes “payoffs.” The payoff to each player
depends not only on the choices selected by that player, but also by the choices taken by
everybody else participating in the game. This is the key feature that creates a non-trivial
problem of strategic interaction. Players act in order to maximize their payoffs (as firms act in
order to maximize their profits).

1. Simultaneous Games
We study two types of simultaneous games. In the first, the one-shot (simultaneous) games, time
does not play any role. The game lasts one period. In the second type, the repeated games, the
one-shot game is repeated either a finite or an infinite number of times.

1.a Simultaneous Games: One-Shot

Each player can select one out of many possible, alternative actions. Actions are chosen
simultaneously and secretly by each player. That is, when a player selects an action he does not
know and does not observe the choice of the other players. An action profile is a collection of
actions chosen by each player.

As a business example, consider how Treasuries are auctioned off by the U.S. government. The Treasury organizes
an auction where institutional investors (primary dealers) have to bid for the price (and sometimes, the quantity) of
T-bills to be issued on a certain date. All bids have to be submitted in sealed envelopes to the Treasury (in fact, these
operations are done electronically and through the NY Fed). The Treasury then ‘opens the envelopes’ and assigns
the quantities demanded to the bidders in ascending orders, starting from those that have submitted the lowest
interest rate (the highest price).

Immediately after the players chose actions, they receive a payoff (prize, profits….) and the
game ends.

The payoff to each player depends on the chosen action profile, i.e., my payoff depends on the
action chosen by myself as well as by the action chosen by everybody else in the game.
The players know the rules of the game. Thus, they know the relation between action profiles
and the payoff of everybody in the game.

A word of caution on payoffs

It is important to understand the meaning of “payoffs.” Let us pick a player, say player #1, and
two action profiles, say A and B. If the payoff to player #1 associated with profile A is greater
than the payoff associated with B, it means that the first player would rather end up with A than
with B. Thus, it is important to bear in mind that the payoffs in a game take into account all the
aspects that affect a player valuation of an action profile, i.e., money, feelings, happiness,
depression….

Bear this in mind when we start going through the examples.

1.b Simultaneous Games: Repeated

This is a situation where the one-shot game is repeated either a finite or an infinite number of
times. In each period, players choose their actions secretly and simultaneously.

At the end of each period, each player gets a payoff, (instantaneous payoff), that depends on the
action profile chosen in that period. Each player acts as to maximize the (discounted) sum of the
instantaneous payoffs.

Again as a business example, consider the previous U.S. T-bill auction. Now assume you and your homologues with
other primary dealers are submitting bids every two months (or whenever the Treasury decides to issue T-bills).
Your goal is to maximize profits of your trading desk for the whole year, and you have to take into account how
each one of your opponent is going to react to the news of you winning the auction with a very low interest rate (a
very high price), or to the fact that you may be using or not some inside information a friend of yours is providing to
you working at the U.S. Treasury, and so on.

Important remark. When we say ‘one-shot’, we also want to cover games that are played many
times, and not just once. That is, the same situation may arise over and over again, but not with
the same players. Think about when you walk into a diner at a gas station on one of your
vacation trips. You and the waiter are playing a game: he gives you good or bad service, and you
can either tip him or not. While you may go to many such diners, and he may see many
customers like you, only few times you go back to the same place, and only few are the
customers the waiter sees every day. Therefore, each one of you is playing a ‘one-shot’ game.
We will see that the ‘one-shot’ situation is common in certain business situations as well. When
we talk about ‘repeated games’, we refer to games repeated with the same competitors or
players.

2. Sequential Games

Players act (possibly more than once) according to a specified and well understood time
sequence.

When it is the time for a player to act, she can choose one out of many possible, alternative
actions. When a player acts, he knows the “history” of the game, i.e., the sequence of actions
chosen by the players that acted before him.

When the game ends, each player receives a payoff. The payoff to each player depends on the
entire history of the game, i.e., by the entire sequence of actions chosen by all players.

As a business example, think about the airlines industry, as it was presented in class earlier on. In the airline
industry, pricing decisions of one company are going to be affected by the pricing decisions of other airliners. If you
offer a 10% discount on your fares while the others keep their fares unchanged, you are going to attract many more
customers than if other airlines follow suit. Moreover, airlines that want to try fare increase typically announce them
over the weekend, and then wait to see what their competitors are going to do. If the competitors follow suit, they
stick to the price increase. Otherwise, they have the option of reverting to lower fares over the week, since most of
the travel agents are open during the week, and especially business travel plans –which represent the larger fraction
of the market share– are made during the week. Since one company announces fares before the others, this is a
sequential game.

We will put off the discussion of sequential games until later. Now we focus on simultaneous
games.

II. One-Shot Simultaneous Games.

In this section we present a number of “famous” examples of one-shot simultaneous games.

They all have a simple, particular, form called “matrix” game. Here is their description:

There are only 2 players. Each player has finitely many actions.
The game is represented by a matrix (box), with several rows and columns. Each row represents
a possible action for the first player. Similarly, each column represents a possible action for the
second player: the first player chooses a row, while the second chooses a column.

Each cell represents a choice of a row and a column. There are two numbers in each cell. The
first (second) number is the payoff to the first (second) player when the given column and row
have been chosen by the two players.

The examples that follow are chosen to be simple and to clarify the relevant issues. They are not
framed in a managerially relevant way. The relevant applications follow this preliminary
discussion.

Everyone knows how to play simple child games, so to understand what a matrix form is, let’s
start by looking at one of these games, ‘Rock, Paper, Scissors’.

In this game, two players simultaneously have to show each other a sign. They have three
options: flat hand (Paper, P), fist (Rock, R), the victory sign (Scissors, S). The payoffs are, say,
twenty cents for the winner, paid by the player who loses.
The rules of the game are as follows:
a) P beats R
b) R beats S
c) S beats P
d) Two identical signs tie the game, nobody wins, nobody loses.

To cast it in matrix form, first we recognize that there are 2 players. Each player has 3
possible actions: P, R and S. So we construct a matrix with 3 rows and 3 columns.
 Player 2
P R S
P
Player 1 R
S

Now we put a pair of numbers in each cell, where the first number corresponds to the payoff of
player 1, the second is the payoff to player 2. For instance, if (P,R) is played, the number is (.20,-
.20); and so on.

Player 2
P R S
P 0,0 .20,-.20 -.20,.20
Player 1 R -.20,.20 0,0 .20,-.20
S .20,-.20 -.20,.20 0,0

Now that we have written down our first game, it’s time to move on to more business relevant
examples, and also to ask ourselves: what is the outcome of a game? What is the outcome of
strategic competition? To put it in different words, how do we solve a game?

1. Games with Dominant Strategy.

We are going to look first at the simplest possible games, those where players have dominant
strategies.

1.a Time and Newsweek

Suppose that there are currently two major stories the two magazines have covered this week:

1. The impasse in the Senate on the Supreme Court Nominee.

2. The release of a new drug against AIDS.

The editors have to choose which story should make the cover. Their goal is to maximize the
number of readers.
Of newsstands' buyers, 30% are interested in the nominee story & 70% are interested in the
AIDS story. If they both issue their magazine with the same cover story, they split the market
equally.
Let's see the payoff matrix:

NEWSWEEK
AIDS NOMINEE
TIME AIDS 35,35 70,30
NOMINEE 30,70 15,15

Analysis:
Assume that you are the editor of Time. You have to decide what cover to choose without
knowing, in advance, what the editor of Newsweek is going to choose. If she chooses AIDS, you
can make 35 by also choosing AIDS or, alternatively, choose NOMINEE and make only 30.
Therefore, we can state that given Newsweek choice of AIDS your best choice would be also
AIDS. If you follow the same line of reasoning, assuming the Newsweek decides to go with the
NOMINEE story, you will easily see that, again, your best choice remains the same: to go with
AIDS.
Now, assume that you are the editor of Newsweek, and repeat the same exercise. You will see
that the best choice for Newsweek, regardless of what Time does, is to go with the AIDS story.
AIDS is said to be a dominant strategy. In other words, each of the editors, choosing the AIDS
story for the cover page could state:

"I am doing the best I can no matter what you do. You are doing the best you can no matter what
I do."

When all players have an action (in general, not necessarily the same action for all players!) that
will be their best choice regardless of what all other players are doing we have a game with
dominant strategies.

In real life it might take you some time and work to realize that you are involved in an
interaction in which you have a dominant strategy. Having realized that you have a dominant
strategy has the potential of making your life simple. Even though what other competitors do has
an impact on your profits, you don’t have to worry about what the other player(s) have in mind.
When there is a dominant strategy, the best managerial decision can be found ignoring what the
others will do. Therefore, the predictable outcome of strategic competition is for each player to
choose their dominant strategy. In this case, it is: (AIDS, AIDS).

1.b A Dominance Solvable Game.

What if no player has a dominant strategy? We claim the game is still solvable by iterated
deletion of dominated strategies or, simply, it’s dominance solvable.
To see this, let's make a small change in the payoff matrix (by changing the payoff of
Newsweek).

NEWSWEEK
AIDS NOMINEE
TIME AIDS 42,28 70,30
NOMINEE 30,70 15,15

You will observe that AIDS is not a dominant action anymore for Newsweek.

If Newsweek knows that Time selects AIDS, it should select NOMINEE, while it should select AIDS if Time
selects NOMINEE.

Can Newsweek predict what Time is going to select? Or, equivalently, can we predict the
outcome of this game?

It suffices to observe that Time magazine does have a dominant action: AIDS. Independently of
the Newsweek choice, Time maximizes its payoffs by choosing AIDS. Thus, Newsweek (and
anybody else) can safely predict that Time will pick AIDS and, therefore, Newsweek should pick
NOMINEE. The outcome of the game must then be the action profile: (AIDS, NOMINEE).

This type of game is called dominant solvable because the outcome can be found by deleting
dominated actions. A dominated action is an action that will be never selected no matter what is
the choice of the opponents. Equivalently, an action is dominated if for any choice of the
opponents there is some other action available that yields a higher payoff.

NOMINEE is a dominated action in the example above. Delete the action NOMINEE for Time since AIDS
dominates it (in different words, AIDS is a dominant action. Hence Time will never play NOMINEE). After that you
have deleted NOMINEE for Time, delete AIDS for Newsweek. Since Time never plays NOMINEE, Newsweek
never plays AIDS. We are left with one action for each player (AIDS for time and NOMINEE for Newsweek) that
constitutes the outcome of the game.

Below is a more complex example where the same logic applies. Let’s solve it.

Player 2
A B C
A 15, 150 9, 5 2, 8
Player 1 B 7, 5 3, 100 5, 6
C 23, 1 7, 3 12, 10

To find out if there are dominated actions, we put a ^ next to a player 1’s payoff if that player is
choosing the corresponding action under some circumstance (some action of player 2):
 Player 2
A B C
A 15, 150 9 ^, 5 2, 8
Player 1 B 7, 5 3, 100 5, 6
C 23 ^, 1 7, 3 12^, 10

So, for instance, we write (23 ^ , 1) because C is the best response against A.
An action is dominant for player 1 if it has ^ next to all its possible payoffs.
An action is dominated for player 1 if it has no ^ next to any of its possible payoffs.

For Player 1, there are no dominant strategies, but B is dominated. No matter what player 2 does,
player 1 will never select B.

Hence, we safely conclude that the dominated action B will never be chosen by player 1.

Let’s delete it.

We get the following game:

Player 2
A B C
A 15, 150 9, 5 2, 8
Player 1
C 23, 1 7, 3 12, 10

In this new reduced game, let’s find out if there are dominant/dominated actions for player 2, by
putting a * next to a player 2’s payoff if that player is choosing the corresponding action under
some circumstance (some action of player 1).

Player 2
A B C
A 15, 150* 9, 5 2, 8
Player 1
C 23, 1 7, 3 12, 10*

Again, an action is dominant for player 2 if it has * next to all its possible payoffs.
An action is dominated for player 1 if it has no * next to any of its possible payoffs.
B is for player 2 a dominated action. Again, let’s delete it:

Player 2
A C
A 15, 150 2, 8
Player 1
C 23 ^, 1 12 ^, 10
You can immediately observe that C is now a dominant action for player 1 and, hence, that
Player 2, understanding that, will choose the action C.

Thus, (C,C) is the outcome of the game.

We obtained this result by iterated deletion of dominated actions.

Games with “dominant” strategies or dominance solvable games are simple. If we have dominant
strategies, we do not worry about our competitors’ behavior, while if the game is dominance
solvable, the choices of our opponents are quite simple to reconstruct.

Now that we know how to predict the outcomes of these games, let’s study the implications of
strategic behavior for the type of outcomes we get.
Games can have very bad outcomes for managers. Our next section will make this point clear.

1.c. The Prisoners’ Dilemma.

The Conductor, Tchaikovsky & the KGB (Taken from “Thinking Strategically”, by Dixit and
Nalebuff)

This is probably the most famous game. It has been studied extensively and analyzed in all
branches of social science. Here is one version of the story:

The conductor of an orchestra in the Soviet Union (during the Stalin era) was traveling by train
to his next engagement and was looking over the score of the music he was to conduct that night.
Two KGB officers saw what he was reading and, thinking that the musical notation was some
secret code, arrested him as a spy. He protested that it was only Tchaikovsky’s Violin Concerto,
but to no avail. On the second day of his imprisonment, the interrogator walked in smugly and
said, “You’d better tell us all. We have caught your friend Tchaikovsky, and he is already
talking.”

So begins one telling of the prisoners’ dilemma, perhaps the best-known strategic game!

Let us develop the story to its logical conclusion. Suppose the KGB has actually arrested
someone whose only offense is that he is called Tchaikovsky, and are separately subjecting him
to the same kind of interrogation. If the two innocents withstand this treatment, each will be
sentenced to 3 years of imprisonment. If the conductor makes a false confession that implicates
the unknown “collaborator,” while Tchaikovsky holds out, then the conductor will get away with
1 year (and the KGB’s gratitude), while Tchaikovsky gets the harsh sentence of 25 years for his
recalcitrance. Of course, the tables will be turned if the conductor stands firm while Tchaikovsky
gives in and implicates him. If both confess, then both will receive the standard sentence of 10
years.
The Payoff Matrix:

MR. TCHAIKOVSKY
CONDUCTOR CONFESS NO CONFESS
CONFESS -10,-10 -1,-25
NO CONFESS -25,-1 -3,-3

Analysis:
Consider the conductor’s thinking. He knows that Tchaikovsky is either confessing or holding
out. If Tchaikovsky confesses, the conductor gets 25 years by holding out and 10 years by
confessing, so it is better for him to confess. If Tchaikovsky holds out, the conductor gets 3 years
if he holds out, and only 1 if he confesses; again it is better for him to confess. Thus confession is
clearly the conductor’s best action, regardless of what Tchaikovsky does.
In a separate cell in Dzerzhinsky Square, Tchaikovsky is doing a similar mental calculation and
reaching the same conclusion. The result, of course, is that both of them confess.
In other words, both players have a dominant strategy to confess. What is devastating in this
game is that both could be better off by NOT confessing, thus going to jail for three rather than
ten years.
To illustrate the devastating nature of the prisoners’ dilemma, Dixit and Nalebuff add the
following little story:

“Later, when they meet in the Gulag Archipelago, they compare stories and realize that they
have been had. If they both had stood firm, they both would have gotten away with much shorter
sentences.
If only they had had an opportunity to meet and talk things over before they were interrogated,
they could have agreed that neither would give in. But they are quick to realize that in all
probability such an agreement would not have done much good. Once they were separated and
the interrogations began, each person’s private incentive to get a better deal by double-crossing
the other would have been quite powerful. Once again they would have met in the Gulag, there
perhaps to settle the score of the betrayals (not of the concerto). Can the two achieve enough
mutual credibility to reach their jointly preferred solution?”

Business applications
There are numerous situations in business resembling the prisoners’ dilemma, and a lot of efforts
are being made by managers to avoid or get out of a prisoners’ dilemma game.

To illustrate a typical business example of the prisoners’ dilemma, consider the following
example:
Two airlines, EasyJet and RyanAir, are competing on a specific route between two cities. For
simplicity, we assume that they have to decide whether to charge high or low price for the flight.
The following payoff matrix describes the payoffs (this time profits, in 100s of euros) associated
with different decisions.
 RyanAir
LOW PRICE HIGH PRICE
EasyJet LOW PRICE 0,0 1200,-200
HIGH PRICE -200,1200 800,800

Analysis:
Repeating the same type of analysis you can easily confirm that both airlines have a dominant
strategy to charge low prices, thus making no money.

However, if, by some mechanism, they could have agreed (and able to confirm) that both will
charge a high price, they would both be better off and make 800 each.

In the EU and many other countries, such an agreement is illegal. This is to say, the law simply
does not recognize and does not protect such agreements. Without the possibility to take
someone to court if they break the price fixing agreement, the logic of the prisoners’ dilemma
suggests that such fixing scheme will not work. Indeed, companies around the world have found
that once in their market there are competitors that start a price decrease, there is little to stop
prices from falling. Examples are: airline companies after deregulation has opened hubs to
competition from other airliners; banks after deregulation has blurred or eliminated geographical
limits of operations for commercial banks.

Sometimes, though, companies find ways to enforce the collusive outcome even without explicit
agreements, through reputation building, say. For this and other reasons, which we will see later,
the government not only does not protect these agreements under the law, but also actively
punishes cartels or collusive behavior. Companies, therefore, have to look for more imaginative
(and legal) solutions to the prisoners’ dilemma.

What if a game is without dominant or dominated strategies? After all, the presence of such
situations seems quite peculiar, though of great importance as we saw through the prisoner’s
dilemma example. How do we predict the outcome of strategic competition in games which are
not dominance solvable?
 Strategic Competition II:
Nash Equilibrium

I. The Need for Nash Equilibrium.

We use the following example to introduce the notion of Nash equilibrium. Stars and hats are
already put in to signal best actions against the opponent for each player.

Player 2
A B C
A 15, 150 * 9 ^, 5 2, 8
Player 1 B 47 ^, 5 3, 100 * 5, 6
C 23, 1 7, 3 12 ^, 10 *

Notice that there is a * for each row and a ^ for each column. Why? Because, there is (at least) a best response (a
payoff maximizing action) to any action taken by the opponent. Thus, there is a * for each row chosen by player 1
and a ^ for each column chosen by player 2.

As we have done so far, we start by putting ourselves in a situation where it is commonly known
that both players are rational, i.e., payoff maximizing. We want to predict how people would
play this game (the outcome of strategic competition). There is little to predict if players are
irrational in a strong sense, that is, if their behavior does not display any kind of regularity.

Throughout the discussion that follows, bear in mind that if we find out that there is a logical,
natural way to act (i.e., to select an action), player 2 will find that out and vice versa. The goal is
to find out what the two players are going to choose, i.e., to predict the outcome of the game.

Analysis:
In this game, none of the players has a dominant action, and there is no dominated action either,
i.e., the game is not dominance solvable. The difficult question, then, is whether there is a way to
predict how the two players will play such a game. Recall that we are trying to come up with a
theory that will help us predict how rational individuals interact in strategic situations (in
particular, when dominant actions are absent).

Rationality in particular means that any player can draw the * and ^ in the matrix of payoffs.
The collection of the cells with ^ is called reaction function (a.k.a. best reply or best response)
for player 1. The collection is (B,A), (A,B) and (C,C). (B,A) means that if player 2 choose A, B
is the payoff maximizing choice for player 1. Similarly, if player 2 chooses B (C), A (C) is the
best action for player 1. With a similar logic, the collection of the cells with * is called reaction
function (best reply) for player 2.

We can immediately check that (C,C) is the only point with both a * and a ^. Thus the action
profile (C,C) has a magical property:

Given player’s 1 choice of C, player 2 finds in her best interest to choose action C. Vice versa,
given player’s 1 choice of C, player 1 finds in her best interest to choose action C. In different
words, the action profile (C,C) has the property that nobody has an incentive to take unilateral
deviations.

Why should we care about this ‘magical’ property?

If players played this game, at the beginning most likely they would try to figure out, or learn,
the game, its rules, how the opponent plays, and so on. Everything can happen at this stage.
Suppose the game is played over and over. We do not know how and when players will get to a
stable way to play, but for sure if they keep using the same strategies against each other, it’s
because they have no incentive to do otherwise.

Therefore, a minimal natural requirement that a stable and predictable outcome of strategic
competition must have is the ‘magical’ property. This is the most direct way to define a Nash
equilibrium:

II. Nash Equilibrium

A Nash equilibrium is an action profile such that nobody has an incentive to take
unilateral deviations.

Let us try to be very clear. There are three points that we want to make.

1. What does unilateral deviation mean?

For instance, the cell A-A has payoffs 15 and 150. Thus, if both players deviated from the action
profile (C,C) to the action profile (A,A), they would be both better off. However, this is not a
unilateral deviation. To the opposite, to reach (A,A) from (C,C), both players have to change
their choice, i.e., both players have to deviate. However, (A,A) is “better” than (C,C). Why is not
(A,A) a more natural solution than (C,C)? Because, if it were natural, i.e., obvious that player 2
chooses A, then player 1 would choose B, not A. In different words, (A,A) is not a point on the
reaction function of player 1. This implies that at (A,A) player 1 has a unilateral profitable
deviation, i.e., the action B.

2. Why is the proposed solution (the Nash equilibrium) natural or obvious?

There is a sense in which a Nash equilibrium is an obvious outcome and a sense in which it is
not. The obvious part concerns the self-reinforcing aspect of it. Nobody has an incentive to
deviate. If it were obvious that one of the two players were playing C the other should obviously
play C and vice versa. However, why is it obvious for any of the two players to select C? The
answer to this question may seem natural in this example, but it is not natural in general. Here,
the answer is that the natural aspect of (C,C) comes exactly from being the unique action profile
with the “lack of unilateral deviations” property. Later, we see that with other simple games we
might very well run into trouble. You should keep in mind that, once more, while we stress that
if ever a situation of equilibrium is reached, it must satisfy the Nash property, we are silent about
how players will ever get to this equilibrium situation.
3. What is the relation between the solutions of the previous examples and the notion
of Nash equilibrium?
You can easily check that all the solutions previously discussed are Nash equilibria. Evidently, if
a player posses a dominant action, he will always chooses it, that is, his reaction function always
selects the dominant action no matter what is the choice of the other players. Hence, in a game
with dominant actions, they constitute the Nash equilibrium action profile. You can check that if
a game is dominance solvable, then its solution is a Nash equilibrium, but not vice versa.

III. Coordination Games

Here we look at a few examples of games that share a common feature: They all have multiple
Nash equilibria. They are called coordination games, because in order to reach a particular Nash
equilibrium, players have to “coordinate” their choices. In the first game the possibility of
coordination is more than reasonable, while in the second game much less. In the third game
there is not room at all for coordination.

1. Pure Coordination Games

Robin
Opera Zoo
Chris Opera 30 ^, 30 * 0, 0
Zoo 0, 0 3 ^, 3 *

The story behind this game is obvious (and stupid). Hence we skip it. First, you can immediately
find out that there are two Nash equilibria: (Opera, Opera) and (Zoo, Zoo). Let’s confirm, for
example, that (Opera, Opera) is a Nash equilibrium. If it were obvious that Robin goes to the
Opera, then Chris, by going to the Opera, gets a payoff of 30, where the payoff of going to the
Zoo is 0. Thus, Opera is his best choice. The same logic applies to Robin and the same logic
applies to (Zoo, Zoo). The problem is that neither Robin nor Chris knows where the other person
is going to go. However, they can argue that since it is preferable for both of them to go to the
Opera, they will both choose to do it.
Hence, the Nash equilibrium (Opera, Opera) seems to be the natural outcome of the game.
The second version of the same game is more problematic.

Robin
Opera Zoo
Chris Opera 30 ^, 30 * 0, 0
Zoo 0, 0 30 ^, 30 *

Once again, there are two Nash equilibria: (Opera, Opera) and (Zoo, Zoo). However, contrary to
the previous example, it is impossible for, say, Chris to mentally reconstruct what Robin is going
to do. Indeed, payoff wise, the two Nash equilibria are indistinguishable. In the absence of
additional information, it seems equally probable for Robin to go to the Zoo or to the Opera.

How can we pin down a Nash equilibrium? There are two answers to this question. At this point
we discuss the first answer. We will discuss the second answer in the next example. Suppose that
the players can communicate before taking a decision. Then, in this case, we do not run into any
trouble. If Chris, for instance, communicates that he is going to the Zoo, Robin will not have any
problem going to the Zoo herself. A little bit of communication would solve this coordination
game, pinning down one of the two Nash equilibria. You might object by arguing that
communication is not part of the description of the game. Nevertheless, in this case, it dos not
alter the basic strategic interaction and helps reaching the necessary coordination.

2. The Battle of the Sexes

This example makes a clear point. Players may not be able to find out what the opponents are
choosing neither on the basis of logics, nor on the basis of pre-play communication. Something
else is needed to solve this type of strategic situations. Quite frankly, the next example illustrates
the limit of the notion of Nash equilibrium as a predictive devise for the resolution of strategic
interactions. Some games cannot be solved in a vacuum. If you take literarily the story that
players select secretly and simultaneously actions, the notion of Nash equilibrium may be of
little help in some circumstances.
Robin
Football Ballet
Chris Football 30 ^, 2 * 0, 0
Ballet 0, 0 2 ^, 30 *

Other than for the numbers, this game is identical to the previous two games. Hence as before,
there are two Nash equilibria (Football, Football) and (Ballet, Ballet). However, contrary to the
previous situations, the two players have different preferences concerning the different
outcomes: Robin would rather go to Ballet, while Chris prefers the Football game. Any attempt
to logically reconstruct the choice of the opponent will, in absence of additional information, fail.
Unfortunately, a pre-play communication will not help. Chris might very well call Robin and
suggest going to the Football game. However, it is very likely that Robin will react by suggesting
going to the ballet. Threatening unilateral actions will not necessarily help: The sentence “Hello,
this is Chris. I am going to the football game and I do not care where you go” may be very well
followed by “Hello, this is Robin. I am going to the ballet game and I do not care where you go.”

The notion of Nash equilibrium per-se will not help telling us what is going to happen. Worst
than that, if two individuals, who do not know each other, are asked to play the game, the
outcome will not be necessarily be a Nash equilibrium even if we allowed them to have some
sort of pre-play communication.

How do we solve this game then? The answer is rather sketchy and disappointing. Games like
this cannot be played in a vacuum. Sometimes, culture (or institutions) will help pinning down
the solution. In some societies, the Chrises always win, in others, the Robins do. To go deep
inside this issue is outside the scope of the course. However, be aware that this notion of Nash
equilibrium cannot be the solution to all problems.
IV. Games Without Nash Equilibria (In Pure Actions): Mixed Strategies (time
permitting)

1. Matching Pennies

Player 2
H T
Player 1 H -1, 1 * 1^, -1
T 1^, -1 -1, 1 *

Two kids play the following game: Each holds a penny in their hand. They simultaneously show
their pennies to each other. If both pennies show head (or tail), Player 1 pays a $1 to player 2. If
one penny shows head and the other tail, Player 2 pays $1 to Player 1.

What is the Nash Equilibrium of this game? (Notice that the game is identical to another famous
game, “Paper, Scissors and Rock”.) The answer is simple. There is no Nash equilibrium action
profile.

Indeed, there is no cell in the matrix where both a * and a ^ appear. Thus, for any conceivable
action profile, at least one player has a unilateral incentive to deviate. To illustrate the problem in
a different way, say that player 1 consider playing H. To this player 2 reacts optimally by
selecting H. To this, player 1 reacts optimally selecting T. To this, player 2 reacts optimally
selecting T. To this, player 1 reacts optimally by selecting H and, then, we start all over again.

There is a particular aspect differentiating this game from the coordination games analyzed
before. There, players had a mutual benefit from coordination. In all three coordination examples
the lack of coordination generated losses for both players. Put it differently, there is an advantage
in being predictable. The situation in the Matching Pennies is, in some sense, orthogonal. Players
do not want to be predictable. If player 1 were predictable, he would loose for sure one penny. In
a situation like this, it is quite obvious that an action profile that constitute a Nash equilibrium
cannot exist. Its existence would imply total predictability and, hence, a sure loss. (The latter
implies existence of profitable deviations, an obvious contradiction).

However, a Nash equilibrium exists if we are willing to extend the notion of choice (strategy) for
a player. So far players could choose actions, i.e., H or T. However, we can allow players to
select actions randomly or to put it differently, to select the probabilities by which they choose
either H or T. These are called mixed strategies.
Accordingly, a Nash equilibrium in mixed strategies is a mixed strategy profile such that no
player has an incentive for unilateral deviations. With this extension, Nash equilibria always
exist (more precisely, under very general and well understood technical conditions). A treatment
of this problem is beyond the scope of this course.
Here we just point out that a mixed Nash equilibrium for the “Matching Pennies” game is for both players to play H
with probability ½ and T with probability ½. Why? We just want to give a rough intuition of the problem.
Suppose that player 1 is adopting such a mixed strategy. What should player 2 do?
If player 2 chooses T, she will end up in the cell H-T with probability ½ thereby realizing a -$1 payoff and in the
cell T-T with probability ½, thereby realizing a $1 payoff. Thus, if she plays T, her expected payoff is ½*(-1)+½*1
= 0.
If player 2 chooses H, she will end up in the cell H-H with probability ½ thereby realizing a $1 payoff and in the cell
T-T with probability ½, thereby realizing -$1. Thus, if she plays H, her expected payoff is ½*1+½ *(-1) = 0.
Hence, player 2 is indifferent, given the mixed strategy of player 1, whether to play H or T: they give her the same
payoff! Since she is as well indifferent whether to play with probability ½ H, thereby realizing $0 of expected
payoff, or playing with probability ½ T, thereby realizing $0 of expected payoff, the mixed strategy “play T with
probability ½ and H with probability ½ ” generates the same payoffs as any other choice.
A similar argument shows that player 1 is indifferent, given the mixed strategy of player 2 of playing T with
probability ½ , between playing H or T, or in fact any mix of the two, so in particular the ½ probability of T
strategy.
Neither player has an incentive to deviate from this particular random choice, so that this strategy pair is indeed a
Nash equilibrium.
(Test: Why a mixed strategy with probabilities different than ½ cannot be a Nash equilibrium?)

2. Interpreting mixed strategies

The interpretation of mixed strategies as saying that individuals, managers, and so on, really flip
a well-crafted coin to decide what to do is unsatisfactory. True, sometimes individuals make
decisions that look random, to confuse the opponent. However, what our mixed strategies
achieve is to make the opponent totally indifferent among the several courses of action so that it
becomes a best decision for them to mix their choices as well. Many would argue this is
unrealistic.

An alternative way to interpret mixed strategies is to think of the game as repeated several times
the same. Then, mixing corresponds to choosing an action in a percentage of times the game is
played (i.e., ‘T with ½ probability’ is actually ‘play T 50% of the times’). The expected payoffs
should then be seen as average payoff to a player over time.

In some circumstances, when the game is symmetric, another alternative view of games and of mixed strategy can
be applied. A game does not represent what two isolated players do, but what two players out of a large population
do once they randomly meet. Example: think about taking the stairs to or from a subway station in New York City.
Here, players do not make conscious decisions, but are ‘programmed’ to play, or behave instinctively, according to
some rule. Some of them go down the left side of the stairs, some go down the right side. The question we ask is:
what’s going to happen in the population of New Yorkers? Well, if the Nash equilibrium exists in pure strategies,
the answer is: 100% of the population decides to do the same thing. If the equilibrium is in mixed strategies, we
interpret it the probability as the fraction of the population that behaves/plays in a certain way. Symmetric games
arise in many social and business situations. They have been successfully applied to evolutionary biology, and an
evolutionary theory of business competition is now growing.
 Competition Through Quantities and Prices

I. Industry Analysis: Oligopolies.

We apply the theory of simultaneous games to the study of homogeneous oligopoly markets.
These are markets with a homogenous product, few big firms, and extremely high barriers to
entry.

The crude oil industry is a good example of such a market, but examples abound. While many
consumer products are not homogeneous across firms (or at least consumers do not see them as
perfect substitutes), manufacturing products that are used as intermediate goods in other
industries are often considered the same even if produced by different firms. So, crude oil is as
good an example as copper could be, or cement, or steel bars, or foamed polystyrene, and so on.

We are going to look at a simple situation where only two firms are present in the industry
(duopoly). All the conclusions survive generalizations, but require heavy computations. The
important element is that, for our theory of how games should be played to apply, the number of
players should be small and players must be sufficiently sophisticated (like managers in firms).

We want to find out the best managerial strategy as well as predict the outcome of competition in
the industry. We will see that the outcome (in particular, how profitable the market will be for
the firms) will depend on what kind of competition firms are engaged in:
- If competition is through capacity, or quantities produced, then the profits for each firm will
be positive but below the industry maximum level.
- If competition is through prices, things will look grim for the industry, as economic profits
will tend to go down to zero, unless firms operate with capacity constraints.
One key question then naturally arises in oligopoly markets: Are firms then going to collude by
forming cartels (e.g., OPEC), to reduce the negative effects of competition and increase profits?
Within this analysis of the cartel problem, we also introduce and analyze the notion of repeated
simultaneous games.

A Duopoly Example

Consider an industry with two firms. Firms are identical and produce a homogenous product.
Each firm knows its own total cost of production, the total cost of production of the competitor,
and the industry demand.

Data: The following data are known by both firms and describe the industry situation:

1) p = 140 - (Q1+Q2) (industry demand)

2) C(Q1 ) = 20Q1 (total cost of firm 1),
3) C(Q2 ) = 20Q2 (total cost of firm 2).
1. Cournot Competition

From now on, and until the end of this section, we assume that competition in this market takes
the form of quantity competition. Firms select outputs Q1 and Q2 in order to maximize profits,
but sell the product in a market at a common price, determined by the interaction of demand and
supply. (Some authors suggest that a more correct interpretation is to view Q1and Q2 as choice of
production capacities.) This is called a Cournot game of competition. Observe that the industry
price, equation (1), depends on the output of both firms. This feature has two implications:
a) Since the profits of each firm depend on the price, they depend on the quantity choice
made by the competitor (strategic interaction).
b) In order to establish profit maximizing decisions, each firm has to guess what the
competitor will do.

We know that firms with market power maximize profits by selecting a level of output at which
the marginal revenue is equal to the marginal costs. However, contrary to what we have seen up
to now, in a situation of strategic interaction the marginal revenue of a firm depends also on its
competitor’s choices. In particular, after both firms have decided how much to produce, each
will only serve part of the market demand, the part that is not already served by the competitor.
So, in computing its demand or revenues, each firm has to take into account the quantity
produced by the competitor.

We will analyze this market under two different scenarios:

- a one-shot scenario, i.e., the life of the industry lasts one period.
- a repeated scenario, i.e., the life of the industry lasts several periods.

We start first with the analysis of one-shot games, and defer the study of repeated games to the
next section.

In the one-shot case, we compare two situations. In the first, firms compete strategically in the
Cournot fashion. In the second, firms try to collude and coordinate their actions by forming a
cartel.

1.1 Strategic Competition

What game are we playing?

This is a game of simultaneous moves. There are two players. Each has to choose a quantity
between 0 and 140 (the maximum market demand). So quantity Qi is the action for each firm i =
1,2. The payoffs are profits, or revenues minus cost of production, i.e.,

p Qi – C(Qi) = [140 - (Q1+Q2)] Qi - 20Qi

again for each firm i = 1,2.

Finding out the best managerial strategy is then the same as computing the firm’s best response,
a.k.a. the reaction function. To find out the outcome of competition is like asking what the Nash
equilibrium of this game is.
A solution method, i.e., a recipe for Nash equilibrium in Cournot games.

One difficulty arising here is that the number of actions is not finite. So we cannot represent this
game in matrix form (you would need infinitely many rows and columns, one for each possible
Q between 0 and 140). What do we do then?

The logic we followed to find the Nash equilibrium in matrix games was to go and find pairs of
actions for which there is no incentive to deviate (the ‘magical property’). Alternatively, this
property occurred at pairs of actions with both a ^ and a *. That is, to find a NE, we looked at
pairs of actions that were best responses for each player.

This logic can still be applied here.

First, we put the ^ and the * on each firm quantity, taking as given what the other firm might do.
This means finding the best Q1 for any given Q2 , i.e., the response Q1(Q2) (and similarly, the
best Q2 for any given Q1 , i.e., the response Q2(Q1) ).

Second, we look for pairs Q1,Q2 that have each a ^ and a * on them, that is, Q^1 = Q1(Q*2) and
Q*2 = Q2(Q^1). These are the Nash equilibrium output levels, and then we get the price and
profits. It’s our predicted outcome of competition.

Step 1: Compute the firms’ best responses.

Start with firm 1. For any given quantity that firm 2 will produce, Q2 , what is the best
managerial output decision for firm 1? This firm has some market power, and therefore the logic
of profit maximization suggests that the best output level should equate marginal revenue and
marginal cost. However, we need to correctly compute firm 1’s marginal revenue.

The demand for firm 1 is the residual market demand after subtracting Q2 ,

p = 140 - (Q1+Q2) = (140 - Q2 ) - Q1,

Since it is linear, as we did before we can apply the short-cut instead of taking the derivative: the
marginal revenue curve has the same intercept and twice the slope, or

4) MR1 = 140 - Q2 - 2 Q1

Similar reasoning applies for firm 2, which should equate marginal revenue and cost to find its
optimal output level given what firm 1 might do. Then, the marginal revenue for firm 2 is
similarly computed, and MR2 is:

5) MR2 = 140 - Q1 - 2Q2

A more direct mathematical solution (for those who want to take derivatives) is as follows. Total revenues of firm 1,
R1, are:

R1 = pQ1 = (140 - (Q1+Q2))Q1 = 140 Q1 - Q12 - Q2Q1.

The marginal revenue of firm 1, MR1, is just the derivative of the total revenues with respect to Q1. Hence, MR1 =
140 - Q2 - 2 Q1. Similarly, the total revenues of firm 2, R2 are:

R2 = pQ2 = (140 - (Q1+Q2))Q2 = 140 Q1 - Q2 2 - Q2Q1.

Hence, taking the derivative with respect to Q2, the marginal revenue for firm 2, MR2, is MR2 = 140 - Q1 - 2Q2.

As for marginal costs, observe that equations (2) and (3) imply that MC1 = MC2 = 20.

Now all we have to do is to equate marginal revenue and marginal cost for each firm.

Start with firm 1: MR1 = MC1 implies using equation (4):

140 - Q2 - 2 Q1 = 20

or, equivalently

6) Q1 = Q1 (Q2) = (120 - Q2 )/2

Equation (6) is also called a reaction function: it sets the profit maximizing value of firm 1’ s
output, Q1, for any value of the competitor output, Q2. In other words, the profit maximizing
level of output depends on the other firm’s choice of output. The reaction function tells, for any
feasible choice of output by firm 2, what the profit maximizing level of output for firm 1 will be.

Using the same argument, for firm 2, MR2 = MC2 implies using equation (5):

140 - Q1 - 2Q2 = 20

or, equivalently

7) Q2 = Q2 (Q1) = (120 - Q1 )/2

Hence, equation (7) is the reaction function of firm 2. Notice how the reaction functions give you
the best managerial decision of output in this market. The manager correctly takes into account
the fact that the firm’s marginal revenues are lower when increasing output by one unit, as its
demand is NOT the entire market demand, but the demand reduced by the quantity produced and
sold by the competitor.

Step 2: Find the outcome of competition.

For given choices of the competitors, both firms use their reaction functions to set profit
maximizing quantities of their outputs. However, the game is simultaneous and firms do not
observe the actions selected by the competitor. Therefore, how are they going to guess the
competitor output? Each firm knows that everybody in the market will try to maximize profits or,
equivalently, that every firm will use its own reaction function. Both firms can compute the
solution (Q1^, Q2*) to equations (6) and (7). Given Q2*, firm 1 will maximize its profit by
choosing Q1^ and, given Q1^, firm 2 will maximize its profit by choosing Q2*. In other words, the
pair (Q1^, Q2*) constitutes a Nash equilibrium: no firm has an incentive to take unilateral
deviations.
In order to compute the pair (Q1^, Q2*), we need to solve equations (6) and (7). However, a
simple observation will simplify the computations. The two firms are identical and, therefore, it
must be that Q1^ = Q2*.

More precisely, (6) and (7) are linear equations symmetric in the indexes 1 and 2 .

Substituting the equation Q1^ = Q2* into equation (6) (or (7)) we get:

Q1^ = (120 - Q1^ )/2

which together with Q1 ^ = Q2 * implies

Q1 ^ = Q2 * = 40

This is the Nash equilibrium of the Cournot game.

Substituting Q1 ^ = Q2 * = 40 in the demand expression (1), we get

` p* =140 - ( Q1 ^ + Q2 * ) = 60

Profits are:

A1* = p* Q1 ^ - 20 Q1 ^ = 60*40 - 20*40 = 1,600

A2* = p* Q2 * - 20 Q2 * = 60*40 - 20*40 = 1,600

Conclusion : Cournot competition leaves the firms with some positive economic profits.
Are these profits the maximum of what the market (that is, technology and demand) can
generate? If not, why? To see this, we look explicitly at what the maximum industry profits may
be.

1.2 Collusion, or cartel formation

Suppose now that both firms agree to form a cartel. The goal of the cartel is to set the industry
output at a level that maximizes industry profits. A rule governing the cartel behavior specifies
how the industry output and profits must be shared among the cartel members. In our example,
things are very simple. Since firms have the same cost structure, we can safely assume that firms
equally share both profits and production. Let Q denote the industry output. Profits maximization
of the industry implies that the industry marginal revenue must equate the industry marginal cost.
The two firms are just acting as a monopoly!!

Since by (1) demand is p = 140 - Q, the industry marginal revenues are

MR = 140 - 2Q

Since both firms have the same constant marginal (and average) cost, the marginal cost of the
industry, MC, is equal to 20 (by equation (2) and (3)). Therefore, MR = MC implies

140- 2Q = 20
which implies that the profit-maximizing industry output, QC , is

QC = 60

Substituting QC = 60 into the demand (equation 1) we get that

pC = 140 - QC = 140 - 60 = 80.

Each firm will therefore produce (and sell) half of the industry output (i.e., Q1 C = Q2 C = 30)
realizing profits equal to:

A1C = pC Q1 C - 20 Q1 C = 80*30 - 20*30 = 1,800 ,

A2C = pC Q2 C - 20 Q2 C = 80*30 - 20*30 = 1,800.

1.3 Cournot competition, economic profits and the stability of the cartel

When firms compete independently in the Cournot fashion, we predict they will realize profits
equal to $1,600. When they form a cartel, they both realize profits equal to $1,800.

You see what happens when firms compete: under Cournot competition, each firm produces too
much and the overall market price will go down, reducing the firms’ profits.

But why do they produce too much? And since the goal of both firms is profit maximization,
should we conclude that the existence of the cartel is the only sensible prediction of the one shot
situation? The answer to this second question is negative. The cartel is intrinsically unstable:
both firms have a unilateral incentive to produce above the cartel agreement (i.e., above 30). In
different words, the cartel agreement does not constitute a Nash equilibrium and it is, therefore,
vulnerable to unilateral deviations.

To analyze this point more clearly, suppose that firm 2 credibly commits itself to produce,
according to the cartel prescription, an output Q2C = 30. What will be the output choice of firm 1,
given the output choice of firm 2? From the previous analysis, we know that firm 1 computes the
profit maximizing output utilizing its own reaction function. Therefore, the profit maximizing
output, Q1, is obtained by substituting Q2C = 30 into firm 1 reaction function, equation (7):

Q1 = (120 - QC2)/2 = (120 - 30)/2 = 45

The market price is computed by substituting Q1 + Q2 C into the demand equation (equation 1).

p = 140 - (Q2C + Q1) = 140 - (30+45) = 65.

Profits are equal to:

A1 = p Q1 - 20 Q1 = 65*45 - 20*45 = 2,025 ,

A2 = p Q2 C - 20 Q2 C = 65*30 - 20*30 = 1,350.
If firm 1 breaks the cartel by producing 45 instead of 30, it realizes profits equal to $2,025, while
if it stays in the cartel, it realizes profits equal to $1,800. It follows that firm 1 has a unilateral
incentive to break the cartel. Of course this is also true for firm 2. Therefore, cartels will never be
an outcome of this game.

You see why firms produce too much: the firm deviating pays a portion of the cost of a price
reduction due to the increased output, and gets a benefit from increased revenues. The other firm,
by staying put, only pays the cost of sales, and gets no benefit! Now, each thinking what the
other might be doing, they will react optimally, as in (6) or (7).

By simplifying a bit, the situation can be represented as a very simple simultaneous game, where
each firm has just two actions, according to the following table:

Firm1 \ firm2 RF2 Cartel

RF1 1,600 \ 1,600 2,025 \ 1,350
Cartel 1,350 \ 2,025 1,800 \ 1,800

Each firm has two choices: to respect the cartel (“Cartel”) or deviate by following its reaction
function (RF). As usual, the first number represents firm 1 profit and the second firm 2. Firm 1
selects rows, firm 2 columns.

It is immediately apparent that RF is, for both firms, a dominant strategy. Hence, (RF1, RF2) is
the only Nash equilibrium of the game. Notice that this is a Prisoners’ Dilemma type of game.
The pair of actions (Cartel, Cartel) provides higher payoffs (relative to the Nash equilibrium
payoffs). However, both players have an incentive to deviate unilaterally by playing their
dominant strategies.

2. Duopolies and Price Competition

Until now, we have assumed that competition in this market occurs through quantity, not prices.
In many industries firms compete also through the prices they charge to buyers. So, in many
markets firms choose their capacity and then price their product competitively. As an example,
think about two airliners serving the same route, JFK- New York – Boston, the early morning
flight, before 9am. Firms will select the number of flights before 9am, as well as the type of
aircraft to be used, determining capacity. Then, they choose the price at which to sell each seat.

We are going to give just a summary analysis of price competition in this last section. Also, for
the rest of this section, we assume away complications that take us away from the main
conclusion we want to make: price competition dries out the economic profits of the firms.

So we will assume that each company sets a unique price for all its customers (no price
discrimination). Also, we focus on the simpler case in which capacity is not a constraint: the
firms have already allocated enough capacity to satisfy all the demand on the given route at a
particular time slot. Third, we are going to consider the case when firms choose their prices
without knowing what their competitor will do: they choose prices simultaneously. Think of the
choice of prices that have to be posted at the beginning of a season (as we saw, such pricing
decisions many times are made sequentially, but we disregard this for the time being).
Competition through prices of this kind is also referred to as Bertrand competition. Finally, we
consider firms that have identical constant marginal costs (this may well be the case for the
major airline companies in the U.S.).

Observe that, since the product sold is essentially homogeneous (that is, we are saying that
customers regard the two airlines as identical in terms of ground or in-flight services, and so on),
all customers will choose to buy the product from the company that sells it cheapest. When the
companies sell at the same price, consumers will buy randomly from the two companies, and we
assume that the two firms will split the market equally (in fact, for our conclusions it does not
matter how exactly the market is split when the firms quote the same price).

This is a game of simultaneous moves. There are two players. Each has to choose a price at
which to sell their product, so price pi is the action for each firm i = 1,2. The payoffs are profits,
or revenues minus cost of production, i.e., pi Qi – C(Qi), again for each firm i = 1,2. Firms set
prices simultaneously to maximize profits.

To find out the best managerial strategy and the outcome of competition is like asking what the
Nash equilibrium of this game is. Again, one difficulty arising here is that the number of actions
is not finite. So we cannot represent this game in matrix form (you would need infinitely many
rows and columns, one for each possible price p between 0 and the highest reservation price of
consumers). To find the Nash equilibrium, we apply the logic that we used in matrix games and
find pairs of actions for which there is no incentive to deviate (the ‘magical property’).

I claim that p1 = p2 = MC is a Nash equilibrium for the price competition game. In fact, no firm
has an incentive to deviate. If firm 1, say, reduces its price below marginal cost, it will serve the
entire market but at a loss (recall that MC = AC!). Hence, it’d rather keep the price higher and
have no loss (and no economic profits). Raising the price is not going to be profitable either, as it
won’t bring any customers in, and profits will stay zero. The same is true for firm 2. Hence no
firm has an incentive to deviate, and p1 = p2 = MC is indeed a Nash equilibrium. You can now
convince yourself that this is the ONLY Nash equilibrium of this game (just pick any other
profile of strategies and you should be able to find a profitable deviation!).

Bertrand (i.e., price) competition brings economic profits down to zero.

The only way out this disastrous situation is if firms have capacity constraints: at capacity, this
may sustain prices higher since increasing demand is not going to bring any more revenues.

II. Repeated Games.

It has been suggested that the outcome of a game might be different the moment the players are
playing the game more than once. For example, when students are asked to play the Prisoners’
Dilemma game once, students rarely co-operate and, as theory predicts, follow their dominant
strategy. However, when students are told, in advance, that they are going to play 20 rounds of
the same game, there is an increase in the number of cases where students do not follow their
dominant strategy.

Are the conclusions of the one-shot situation going to change once we increase the length of the
industry life to several periods? What is the prediction of the theory about this situation of
strategic interaction? Will cartel prevail as an equilibrium outcome of the repeated game?
These are the questions that we analyze in this section. It turns out that the conclusions depend
on whether we assume or not the existence of a last date for the industry life. Hence, we are
going to look at two scenarios: 1) the industry life lasts for T periods; 2) the industry life lasts
forever.

1. Finite Industry Life

The life of the industry lasts several periods. Each period, the two firms play the simultaneous
one-shot game described by the table at the end of the previous section. Firms know the actions
selected in the past by themselves and the opponent. For sake of simplicity, suppose that the
discount factor is one so that firms wish to maximize the (undiscounted) sum of all periods’
profits. In other words, each period, each firm chooses simultaneously and without observing
the choice of the opponent whether to play RF or Cartel. Then, in each period, the firm collects
profits according to its action and the action of the opponent.

It is natural to conjecture that a situation where firms interact for several periods provides a
powerful incentive to the formation of a Cartel. Unfortunately, this conjecture is not correct (that
is, our theory cannot explain it).
The easiest way to understand the problem (I believe) is to start studying the two-period game.

To construct an equilibrium, we need to introduce a new tool: backward induction. We start

analyzing the game at the last date. What is going to happen in the second period, i.e., the last
period of the game?

Since the game lasts only two periods, both firms are in the last period in the same situation of
the one shot game previously analyzed. We know that the only (Nash equilibrium) outcome in
this case is (RF,RF). Now go back to the first period. Both firms know that, independently of
what they play today, tomorrow they will both select (RF,RF). In other words, the actions
selected today will not affect the decisions of tomorrow. But, then, even in the first period we are
in the same situation of the one-shot game and therefore both firms will choose the dominant
action RF!

Is this conclusion dependent on the assumption that the life of the industry lasts only two
periods? Not at all. Let’s say that the game is repeated T times, for T = 10,000 periods. Again,
proceed by backward induction. What is going to happen in period T? This is the last period, i.e.,
there is no future. Both firms play (RF,RF). Now go to period T-1. Firms’ choices at T-1 will not
affect what is going to happen at T. Therefore, in period T-1 we are in a one-shot game situation.
Therefore, (RF, RF) is the only possible outcome. Now go to period T-2. Both firms know that
independently of what they do at T-2, at periods T and T-1 the outcome will be (RF,RF).
Evidently, we are again in a one-shot situation. Therefore, both firms choose (RF,RF). Proceed
backward until the first period. Conclusion: in each period firms select (RF,RF).

2. Infinite Industry Life

The analysis developed so far is based on the fact that there exists a final period. Suppose that
the life of the industry lasts forever. The backward induction mechanism cannot be applied.
What is going to happen in this scenario?

Consider the following fictitious phone call exchanged by both firms (without being heard by the
Justice Department):
Firm 1: “Hello? This is firm 1.”
Firm 2: “Firm 2 speaking.”
Firm 1: “I have a proposal for you. Let us both play cartel forever. However, if, in any period,
you cheat by selecting the action RF, from the next period on I will always choose RF. What do
you say?”
Firm 2: “It sounds like a good idea. However, be aware that I will use the same strategy. If you
cheat, from the next period on, I will play RF.”

Evidently, if both firms comply with the strategy outlined in the phone call, the cartel will last
forever. Are these strategies a Nash equilibrium? Or, equivalently, do firms have an incentive to
take unilateral deviations?
Consider one firm, say firm 1. Also consider an arbitrary period. Let us analyze a unilateral
deviation taken by firm 1. (Since the deviation is unilateral, firm 2 is following the strategy
outlined in the phone call). If firm 1 plays Cartel realizes $1,800 (firm 2 is playing Cartel), if it
plays RF it realizes $2,025. Therefore, there is a short-term gain of $225. However, after this
deviation, firm 2 will switch to RF (and so will firm 1). As a consequence of the deviation, firm
1 will realize $1,600 loosing each period and forever $200. Evidently, the long-run losses are by
far heavier than the short-term gain. Hence, there is no room for unilateral deviations or,
equivalently, the phone call strategies constitute a Nash equilibrium. Both firms play Cartel in
each and every period of the game.

Remark: Although the phone call strategies are a Nash equilibrium, you might have noticed that
there is an unpleasant aspect that makes them not credible. Say that firm 1 cheats, by selecting
today RF. Firm 2 is supposed to “punish” the cheater by switching, from tomorrow on, to RF.
However, by doing that firm 2 is also punishing itself: it goes from a payoff of 1,800 to a payoff
of 1,600. There are credible and more sophisticated punishments that support the (Cartel, Cartel)
outcome. This discussion, however, is beyond the scope of this course.

Conclusions: You should be very careful in applying theoretical to real oligopoly industries. A
first point to take into account is the meaning of a finite industry life.
Firms are run by managers that have a finite horizon.
If the industry is stable (no entry by outsiders, no appearance of powerful substitute, and no other
changes that affect the nature of the industry) we may as well take the infinite version of the
model to be the relevant one.
However, we need to make one other important qualification. In order to punish the cheaters,
firms have to be able to detect deviations. This is possible only if there are few firms in the
industry and if the demand is fairly stable. Historically, these seem to be the conditions that made
cartel agreement effective in real industries. (Also, do not forget that collusive behavior is illegal
in advanced capitalistic countries.) However, if the industry is unstable or there are indications
that in the near future the industry structure is going to change, we must use the finite version of
the model. Under these circumstances, as in the real world, cartels break.
 Strategic Competition III:
Sequential Games

I. Introduction

Price competition, we saw through industry examples, is likely to occur over time. Delta posts
airfares before other airlines, who then may decide to copy or differentiate their pricing schedule.
CBS may decide to announce its fall season programming schedule for Monday nights before
NBC does (see Extra Problem 6). Auctions many times allow bidders to submit their bids
publicly, so that all other auction participants know what the latest bid is. As an example,
Sotheby’s auctions for artwork are typically are organized as English auctions, where the price is
bid up sequentially by participants.

On the other hand, we saw that simultaneous games often have the unpleasant feature of multiple
Nash equilibria, reducing our ability to predict the outcome of strategic competition, and limiting
the ability to prescribe a clear course of action to managers. We may ask ourselves if, by playing
in sequence, managers can force one Nash equilibrium (the one they like the most).

Finally, in many businesses firms who enjoy market power want to keep it, and keep competitors
out, or want to form alliances with competitors to reduce the negative effects of competition on
their profits. Many times they announce dissuasive strategies. But are these strategies credible?
How do we judge the credibility of a competitive threat?

To study all this, we now analyze situations of strategic interaction where players act
sequentially.
As we have done so far, we look at relatively simple situations called ‘games of complete
information’. In the context of sequential games, the latter also means that players, whenever is
their turn to select an action, know the past history of the game, i.e., they know what are actions
selected by all the players that acted before them.
The games last a finite number of periods. At the end of the game, each player collects a payoff
which is, as usual, dependent on the action chosen by all the participants in the game.
Sequential games can be described as decision problems by means of a tree. However, in a
decision tree the same individual makes choices at all decision nodes. In a sequential game,
different players may have to act at different decision nodes.
We illustrate the structure of the problem with an example. We first tell a simple story and then
we translate it in a sequential trade as illustrated by a “game tree.”

The Story (fake)

Cool Inc. is operating in the US fashion industry. Cool has to take a decision about the new line
to be launched for this coming spring. They are pondering about two alternative lines: Sporty and
Classical. Once the decision is taken, the production process starts rather quickly and Cool’s
choice becomes known in the industry. The top management of Cool has learned that a long time
competitor, Elegant Ltd., is entering the US market in the coming spring. Elegant is present in
Europe and Australia and can select for the US market either the more classical European line or
the more casual Outback line. Given the timing of production and entry, Classical will know the
line adopted by Cool before deciding whether to use the European or the Outback line. This is
the key element of this game. The table below summarizes the profits of the two firms (in
millions of $) for each conceivable line selection profile.
 Elegant
European Outback
Cool Sporty 6, 4 3, 3
Classical 2, 2 4, 6

II. A Tree
This table, however, lacks to represent the key element of this strategic situation: time, and the
fact that Classical will know what Elegant has chosen by the time Classical makes its decision.
To make time explicit, we describe the game as a tree.

Cool (Player 1) moves first by deciding whether to use Classical (C) or Sporty (S). Then, having
learned the choice of the competitor, Elegant (Player 2) picks either European (E) or Outback
(O) . Finally, the game ends and the two firms receive their payoff.

A tree is a collection of nodes and branches.

= Decision Node

= Event Node

⊳ = Terminal Node

Decision Node: A point in which the decision maker has to select one out of several possible
decisions. All possible decisions at this point are described as branches coming out of the
squared node. For example:

Sporty

Cool Inc.

Classical

Note that along the tree different decision nodes may be assigned to different players.

Event Node: A point in which several possible outcomes/states/events can take place/happen: an
event node is just a decision node for ‘Mother Nature’ ©. All possible states of nature are
described as branches coming out of the circled node, so that each of the branches describes a
possible state of nature. For example, suppose that the firms’ profits depend on an unknown level
of market demand, which could be high or low depending on random factors (such as the
performance of the overall economy which, eventually, is related to some uncertainty over
technological innovation, availability of resources, and so on).
Since each of these possible states is uncertain, we assign to each of them a probability. If we
believe that the chance of a high demand is 60% then we will have the following event node
description:

High .6

Demand

Low .4

Notice that the sum of the probabilities of all states of nature is 1 (100%). This means that all
states of nature are mutually exclusive and all possible states are covered. However, we will not
pursue the description of uncertainty at this stage for our strategic competition example.

Terminal Node: A point in which the decision process ends and payoffs (profits) are realized.
To each terminal node there is a number attached, the payoff realized when that terminal node is
reached. Since there are no decisions to take at terminal node, there are no branches emanating
from it.

At the four terminal nodes there are two numbers: the first number corresponds to the profits of
Player 1, Cool, the second to those of Player 2, Elegant.

E (6,4)

S Elegant (1)

O (3,3)
Cool
E (2,2)
C
Elegant (2)

O (4,6)

III. Actions and Strategies

There is a first important difference between simultaneous and sequential games. As we have
already seen, in a simultaneous game, players select an action once and for all. In a sequential
game, a player might have to make decisions at different times and this depending on what the
other players have played, i.e., at different decision nodes. Observe:
Actions chosen at different nodes can potentially be different.

After all, this is the advantage of knowing in advance what others have done. Based on the
information on what Cool has played, Elegant can decide to act on this information. That is,
Elegant can act differently depending on whether Cool has chosen the Sporty or Classical
clothing line. Thus, in sequential games, we cannot describe the behavior of a player by
indicating the chosen action, but rather we have to indicate the collection of actions chosen at
each node at which the player might be operating. This is called a strategy.

A strategy for a player is a list specifying the action chosen at each node where the player might
be operating.

Let’s go back to the previous example. What are the actions of the players?
Cool: Sporty or Classical.
Elegant: European or Outback.
What are the strategies of the players?
Cool has as many strategies as actions, namely, Classical and Sporty, since Cool does not get any
information on what the other is going to do before it is its turn to play.
However, Elegant gets information on Cool’s action, and has two decision nodes. Thus, a
strategy for Elegant specifies the choice of an action at each of the two nodes. For example,
(Outback, European) is a strategy indicating that Elegant selects Outback at the first decision
node (i.e., if Cool selects Sporty) and European at the second decision node (i.e., if Cool selects
Classical).
Hence, Elegant has four strategies:
(Outback, European); (European, Outback); (European, European), (Outback, Outback).

Playing the Sequential Game Simultaneously, Through Strategies

Here it’s time for an important and may be not so obvious observation. Strategies are contingent
plans for action. If in a sequential game players choose strategies, they can decide how to play
the game before the game is actually played, by thinking ahead and preparing a plan for action
(that’s what a strategy is). Therefore, the game is played sequentially in actions, but
simultaneously in strategies!!

To help you fix the ideas, imagine you are the top manager at Elegant. You study the situation
and come up with a plan for action, or a strategy. Let’s say you think it’s best for you to follow
(European, Outback) as a plan (whether that really makes sense or not for profit maximization,
we will see below). Now, it turns out that this year you will be away for family reasons in Down
Under, and won’t be able to follow the U.S. market personally. So, you will not be around after
Cool will have announced and selected the clothing line for the spring. Does it mean you cannot
effectively compete with Cool?
No!
In fact, you can tell your assistant beforehand what to do. You just write on a piece of paper:
“Dear Alex, I will be away for the rest of the year. I won’t be around before Cool makes a
decision on which clothing line they will make next year. However, if Cool selects Classical, we
will go ahead and select Outback. If Cool selects Sporty, we will go ahead and introduce our
European line.”
The assistant, with this set of instructions carefully filed, is going to confidently wait for Cool to
make its announcement for the spring season. Of course, our top manager at Elegant does not
observe NOW what Cool will do, but will observe it when in due time, and this is all that matters
to be able to make contingent plans. Now you see that Cool and Elegant are again choosing
strategies before knowing what strategy the other has chosen. So, the game is played
simultaneously in strategies.

IV. (Nash) Equilibria, Credibility, and Backward Induction.

We start investigating what is the right solution concept for sequential games. In other words, we
would like to come up with a theory that predicts how rational players are going to play any
given sequential game.

Nash Equilibria
Since we said that sequential games are actually simultaneous games in strategies, it seems quite
natural to extend the definition of Nash equilibrium used in simultaneous games to the setting of
sequential games. Thus, naturally and mechanically we are led to the following definition:

A Nash equilibrium of the sequential game is a strategy profile (i.e., a choice of a strategy for
each player) such that no player has an incentive to take unilateral deviations.

We are going to argue that this is not a satisfactory theory to predict the behavior of rational
players interacting in a sequential game. However, first let us indicate and explain what the Nash
equilibria of this game are.

First the game in matrix form is as follows.

Elegant
European, European, Outback, Outback,
European Outback European Outback
Cool Sporty 6,4 6,4 3,3 3, 3
Classical 2, 2 4, 6 2, 2 4, 6

Repeat: (European, Outback) means “Elegant chooses European at decision node 1, and Outback
at decision node 2”. So (European, European) means “Elegant chooses European no matter what
Cool does”; and so on. So if Cool chooses “Sporty”, it does not matter for payoffs what Elegant
does at decision node 2. For payoffs. However, it DOES matter for the final outcome of the
game, as we will see (Intuitively, even if we do not end up at decision node 2, I can scare you off
from being there through my plan of action for that contingency.)
Start by putting ^ and *.

Elegant
European, European, Outback, Outback,
European Outback European Outback
Cool Sporty 6^,4* 6^,4* 3^,3 3, 3
Classical 2, 2 4, 6* 2, 2 4^, 6*

Observe that Elegant, e.g., is indifferent between (European, European) and (European, Outback), so we put a star
on each.

We see that there are 3 Nash equilibria:

1) Cool’s strategy: Sporty; Elegant ‘s strategy (European, Outback);

2) Cool’s strategy: Sporty; Elegant’s strategy (European, European)
3) Cool’s strategy: Classical; Elegant’s strategy (Outback, Outback);

Why is {Cool’s strategy: Sporty; Elegant ‘s strategy (European, Outback)} a Nash equilibrium?
Given the fact that Elegant adopts the strategy (European, Outback), Cool knows that by choosing Sporty is going
to end up at the first terminal node realizing a profit of 6, while choosing Classical is going to end up at third
terminal node realizing a profit of 4. Thus, Cool’s best choice is Sporty or, equivalently, Cool does not have
unilateral and profitable deviations.
Let us now think about Elegant. Since Cool picks Sporty, Elegant is operating at the first decision node. From
Elegant’s point of view (but not from Cool’s) the choice at the second decision node is irrelevant. What matters is
Elegant’s choice at the actual decision node that Elegant is, i.e., the first. If Elegant selects European, Elegant
realizes profits of $4m, while the profits drop to $3m under the alternative choice, Outback. Thus, given that Cool
has chosen Sporty, Elegant profit maximizing choice is to select European at the first (and actually reached) decision
node, while its choice at the second (and unreached decision) node is quite irrelevant. Thus, Elegant does not have
unilateral and profitable deviations. Hence, the strategy profile {Cool’s strategy: Sporty; Elegant ‘s strategy
(European, Outback)} is a Nash Equilibrium of the game.

The third Nash equilibrium profile is conceptually more complicated. It shows that the notion of
Nash equilibrium is for sequential games problematic. Thus, we take our time to explain it.

Let’s start with the obvious part of the 3rd equilibrium. Given that Cool plays Classical, the
second decision node is reached and, there, the profit maximizing choice for Elegant is Outback.
As already explained, for Elegant, its choice at the first (and unreached ) node is irrelevant.
Elegant is not there. Hence, we can again conclude that at the strategy profile:
Cool’s strategy: Classical; Elegant’s strategy: (Outback, Outback);
Elegant does not have a profitable and unilateral deviation.

Can we reach the same conclusion for Cool? Here we see what is wrong with the notion of Nash
equilibrium.
By definition of Nash equilibrium, each player takes as given the strategy adopted by the other
player.
Hence, in that spirit, Cool takes as given Elegant’s strategy (Outback, Outback). If that is the
way Elegant is going to play the game, and if Cool plays Sporty, Elegant plays Outback and
Cools realizes 3m in profits, while if Cool Plays Classical, Elegant plays Outback and Cool
realizes 4m in profits. Thus, Cool should play indeed European or, equivalently, the 3rd strategy
profile is a Nash equilibrium.

However, there is a clear problem with this line of argument. Elegant can change action in the
course of the game.

Should Cool then really take as given the strategy of Elegant? It seems quite reasonable to argue
that it should not. If Elegant is rational and if Cool plays Sporty the choice of Outback is not
credible. The profit maximizing choice for elegant is to select at the first decision node
European, not Outback. In different words, the idea of taking as given the strategy of the other
player makes little sense in a sequential game.

How should we think then? We can actually argue that sequential games are easier to analyze
then simultaneous games provided that we make explicit the idea that Cool should not take as
given any “crazy”, or not credible, strategy of Elegant. Cool should consider that only credible
strategies for Elegant are going to be played.
What’s a crazy strategy?

Backward Induction and Credibility

Imagine to be the first player, Cool, and think about what your rational opponent could do at the
different nodes that can be reached in the game. Thus, imagine that Elegant finds itself at the first
node, i.e., that Cool selected Sporty. What should Elegant select at this point?

Cool should assume that Elegant is a rational manager, i.e., wants to maximize profits. We say
an action is crazy or not credible if it does not maximize profits at that point of play.
A strategy is credible if it does not involve playing any action which is not the best response at a
given node where it’s played.

Obviously, in this case the profit maximizing action is European. Imagine now that Elegant is at
the second node, i.e., that Cool selected Classical. What should a rational Elegant select at this
point? The profit maximizing action is Outback.

Thus, to reconstruct credible strategies we have to start from the end of the game, and look for
profit-maximizing actions at each decision node reached before the end. By arguing by backward
induction, we then exclude strategies that use actions which are not best responses (i.e., profit
maximizing) at any decision node. The “credible” strategy of Elegant is (European, Outback).
Since this is the strategy chosen by Elegant, Cool knows that by selecting Sporty it realizes $6m
of profits, while the profits generated by the choice of Classical are only $4m. Thus, Sporty is the
profit maximizing strategy for Cool. In the picture below, the darker branches indicate the
“credible” strategies for both players.
 E (6,4)

S Elegant (1)

O (3,3)
Cool
E (2,2)
C
Elegant (2)

O (4,6)

By arguing by backward induction we have eliminated 2 of the three Nash equilibria that we
found in the previous section. This game has a unique “sensible” outcome. It is the Nash
equilibrium number 1) and it is obtained through a backward induction argument. (This
equilibrium is called “subgame perfect” in game theory). The logic is simple. By proceeding
backward we select at each node the profit maximizing action of the players thereby eliminating
any Nash equilibrium based on non-credible threats. Let us repeat the same concepts with
different words:
Elegant wants to threat Cool. It goes around screaming that no matter what it will adopt the
Outback choice. Ex-ante, Elegant would love to commit to this choice. Indeed if it were
believable, outcome 3) would be an equilibrium and Elegant would realize a profit of $6m.
However, this is not credible. Once we actually play, if the first node is reached, Elegant will
adopt European. The ex-post incentives make (Outback, Outback) a non credible strategy or,
equivalently, a strategy based on a non credible threat.

Back to Playing Strategies Simultaneously: What Should the Assistant Do?

Recall that we said that the sequential game (in actions) can be played simultaneously in
strategies. So, what went wrong with that? The point is that not all strategies can be played when
the game is to play them as a sequence of actions that can be changed while we play. Suppose
our assistant received the instructions “Choose Outback no matter what”. Our poor Alex would
be totally lost if ever Cool were to play Sporty: he’d see that the manager told him to choose
Outback, but that makes no sense at this point! He’d think the manager made a mistake in typing
the instructions, since obviously playing European would help Elegant Ltd. make higher
profits…and for sure the manager will not be happy if Alex did not always try to maximize
profits…
You see that playing “Outback no matter what” cannot be a stable situation. Taking this into
account, Elegant’s top manager will leave sets of instructions that are always consistent with
profit maximization. As we saw, they are easily found looking at the game in its tree form and
using backward induction.
Unless, somehow, Elegant can credibly commit.
 Threats, Credibility and Commitment

I. “Burning Bridges”a and Other Commitment Devices

The discussion we had so far points out the main problem revealed by sequential games. What is
credible is what is in the interest of player when it comes the time for the choice.
Let us reiterate the problem with an additional example (in some sense, similar to the one
analyzed before). The game that follows is called the trust game:
Player I can take one out of two alternative actions (do not trust), (trust). With the first action the
game ends and payoffs are (0,0). With the second action, the game reaches a decision node of
Player II, where she can take one of two possible actions, Fair or Unfair. The sequential game is
represented below.

Fair (1,1)

Trust II

Unfair (-1,2)
I

Do not
(0,0)

We can immediately argue by backward induction and find out what are the equilibrium
strategies. If the second player is at her decision node she will realize a payoff of 2 by selecting
Unfair and of 1 by selecting Fair. Quite obviously, Unfair is the rational choice. Player I
understands that and, thus, he picks the action (not Trust). Thus, the only credible Nash
equilibrium is Not Trust, for Player I, and Unfair for Player II. The dark branches denote the
backward induction strategies, i.e., the “credible” Nash equilibrium of this game.
Three observations are in order:
1) The choice Trust, for Player I, Fair, for Player II, delivers a payoff of (1,1) that dominates
the equilibrium payoff of (0,0). Thus, Player II has an incentive to promise that she will
pick Fair. However, this promise is not credible. Ex-ante, Player II would love to commit
herself to the choice of Trust. Ex-post the incentives go in the opposite direction.

a
The expression “burning bridges” refers to an old practice of army generals. In order to make the troops fully
committed to the fight, they would order the destruction of the bridges behind them to make retreat impossible. This
of course would signal to the other army the determination to aggressively fight.
 2) This game is very similar to the game between Cool and Elegant. There, Elegant was
threatening the choice of Outback no matter what. Ex-ante that’s what Elegant wanted to
commit to. However, ex-post that threat was not credible. The ex-post incentives were for
Elegant to select European when Cool had previously selected Sporty.
3) You might find confusing and even annoying the adopted definition of equilibrium. Why
do we need to specify what the second player is doing? After all, the first player picks
(Not Trust) and, hence, the second player will never have the opportunity of playing. The
answer is obvious. The first player selects (Not Trust) exactly because he is anticipating
that the second player would otherwise select Unfair. Without specifying this choice, the
behavior would not be well defined.

This discussion opens up a conceptual problem with important practical applications. There are
circumstances where players make threat or promises that are not credible. The players would
love to commit to them, but ex-post cannot deliver them. How can a rational player transform a
not credible promise or threat in a credible one? In different words, how can a player change the
game to her advantage?
The general idea is the following:

Take unilateral and costly actions that align your ex-ante with your ex-post incentives.

There are a lot of stories that make this point in the business world. Using a third party as a
commitment device, you reduce your set of choices in the future: send an attorney to
negotiations; use a human resources manager to implement layoffs in your division; and so on.
Alcoa in the aluminum industry is also an example of such strategy of ‘tying yourself to the
mast’, which itself is similar to the old military school “burning bridges” adagio; same for what
Polaroid did in its successful battle against Kodak in the 70s and 80s. You can also increase
your opponent’s choices to your advantage, to weaken its ability to commit, or your ability to
deviate from cooperation. This is what Intel did when selling the processor 086. Many more
examples of this logic are found in the practice of strategic competition in the business world.

We tell the basic idea by writing down the very famous entry game. The second player is a
famous firm operating into a market (Alcoa in the aluminum industry). The second player is
aware that a potential competitor, Player I, is thinking of entering the market. Thus, Player I has
two choices: Enter, or stay out. However, if the first player enters the market, the second can
either accommodate the entry, or take an aggressive action and fight (for instance lowering the
price). The picture below explains the situation.
 Accommodate (2,2)

Enter II

Fight (-1,-1)
I

Do not
(0,6)

Once again, by backward induction, the only credible equilibrium is: Entry, for Player I;
Accommodate, for Player II. Again, Player II may very well threaten Player II that he will fight
upon entry. Ex-ante, that’s what he would like to do. Ex-post, this is not credible. The incentives
are not there.
Now we follow up the story by introducing an additional choice that Player II can take today,
before the entry choice of Player I takes place. Player II has to decide whether to invest in an
optimal technology or to invest resources in a technology with high capacity (and high fixed
costs). To be precise, optimal technology means that the profits to Player II are higher with the
optimal technology irrespectively of the action taken by the competitor. The game tree below
represents the situation:

Accommodate
(2,2)

In II (2)

Fight (-1,-1)
Opt. techno. I (1)

Out (0,6)
II (1) Accommodate (2,2)
In II (3)
High capacity
I (2) Fight
(-1.5,2.5)

Out (0,3)
Let us stress the main point: the high capacity technology is suboptimal. The profits of the
second player are uniformly lowered by its adoption. However, the high capacity technology has
a great feature: in the case of entry, it makes Fight more attractive then Accommodate. Let us
characterize by backward induction the only “credible” equilibrium. This is illustrated below.

Accommodate
(2,2)

In II (2)

Fight (-1,-1)
Opt. techno. I (1)

Out (0,6)
II (1) Accommodate (-2,2)
In II (3)
High capacity
I (2) Fight
(-1.5,2.5)

Out (0,3)

The “credible” equilibrium is characterized by the darker branches in the picture above. The
equilibrium is:
{(High Capacity, Accommodate, Fight), for Player II; (In, Out) for Player I}.
(Remember that an equilibrium must specify the choice of an action at each decision node included those that are
never reached.)

What we are saying, in the present context, is that the second player adopts the High Capacity
technology and that, as a consequence of that, the first player does not enter the market. Thus,
although the second player is apparently adopting a “bad” technology (cost inefficient), it is
because of this that he is able to make the Fight threat credible, thereby gaining (compared to the
previous situation) $1m.

Lesson: do not judge actions (adoptions of technology) without understanding how their choice
alters the strategic interactions with the other players. The High Capacity technology is bad. If
we do not look at how it affects the strategic balance we should never adopt it. However, when
we look at the problem anticipating the reaction of our opponent, it is the right technology. It
makes Fight credible. That is what we need to do in order to keep the competitor out.

Promises.
The previous example has shown that threats can be made credible by costly actions. Alcoa
adopted this strategy in the aluminum industry. The next example shows that the same principle
holds true for “promises”. The story is taken from the textbook of D. Kreps. We are in 1982.
Intel is trying to convince computer makers to build machines around its new processor, Intel
086. The computer makers are skeptical about this idea. Why? They are afraid that by
constructing these machines they will become prisoners of Intel. The day after (and indefinitely,
in the future) Intel can increase the price of 086 and eat up the Computer makers’ profits. This is
the way the computer makers see the game:

High price (0,17)

Adopt Intel

Low price (10,10)

Computer
maker
Do not
(1,0)

The computers makers are right to be skeptical. Today Intel has an incentive to promise Low
prices. Tomorrow, its incentives will be different. The only “credible” equilibrium is: (Not
adopt) for the computer makers, High (price) for Intel. How can Intel make sure that the low
price promise is credible?

This is what they did it: Intel licensed the production of the microprocessor 086 to several other
competitors, retaining only 30% of the market share. If Intel tries to increase the price of the 086,
the other suppliers will not and Intel will lose a lot of business. The low price strategy is now
credible.
This is how Intel modified the game (the dark branches indicate the equilibrium strategies; Intel
is Player II).
 High price
(8, 5)

Adopt Intel (2)

Low price (10, 7)

CM (1)
License to others
Do not (1,0)
Intel (1) High price (0, 17)
Adopt Intel (3)
Exclusivity
CM (2) Low price (10, 10)

Do not (1,0)

Once again, considered in isolation, “License” is a bad choice. It lowers the profit of Intel for any
conceivable choice of the competitors. In the strategic context, License is a great choice. It alters
the incentives and it makes Low (price) a “credible “promise (thereby increasing the profits of
Intel by 7m).

II. A Warning: The Limits of Backward Induction

The arguments we have used so far are heavily based on backward induction. Here, we want to
suggest that some caution should be used in taking for granted that actual players act and think
according to backward induction.

The first observation is clear, but with little practical content: Games can be complicated. For
instance, the chess game (adopting the international rules) is a finite game. Thus, it is very
similar to our sequential games. It is well known that the chess game has a solution. Namely, it is
well known that there exists an optimal strategy according to which either the White wins, or the
Black wins or there is a tie. The optimal strategy could be in principle computed by backward
induction. However, the game tree associated to Chess is too complicated. We do not know what
the optimal strategy is (and we do not even know what the optimal strategy yields, White or
Black victory, or a tie). By the way, this is why we still play chess.

The second and may be more convincing objection comes from an observation we made before.
When there is a large number of players and we are not sure whether they are rational, then the
premises of backward induction fall apart. We saw this already in the ‘beauty contest’ game we
played in class (“pick a number between 0 and 100…”). Playing a strategy based on backward
induction is like playing based on repeated guesses of the rationality of a large number of
players. It is likely that too many rounds of “I know that you know that I know that you
know…that we are rational” are beyond the logical ability of players. Not to mention that since I
can never observe what you would think or do in situations that will never occur leaves us free to
make whatever guess we want, and perhaps some of these guesses will turn out to be self-
fulfilling! Then, our certainty about how a player will play down the tree vanishes. And then we
might as well deviate from such a ‘logical way’ to play!

One Final Word On Sequential Games: First Mover Advantage.

Let’s go back to an earlier example. Extra Problem 6 said that CBS and NBC were playing a
game of TV programming, in which two were the possible Nash equilibria of the simultaneous
move game (after deletion of dominated strategies):

Payoffs are (CBS, NBC)

Comedy Movie
Comedy 50, 129 110^, 130*
Movie 90^, 190* 80, 90

However, which equilibrium will occur we cannot say. One idea for you as the manager at CBS
could be the following: if we announce our programming schedule first, since changing it is
going to be costly, NBC will realize we will stick to it. Now, observing our choice as given, they
will have only one possibility, which is to follow us in whatever the strategy which is sustainable
for both of us (i.e., profit maximizing). If CBS moves first, contrary to what you may at first
think, this does not give up an informational advantage to the rival. On the contrary, it is going to
give CBS the first mover advantage. CBS is going to corner NBC to their best response.

Here is the sequential game, with the backward induction solution.

C (50,129)

C NBC (1)

M (110,130)
CBS
C (90,190)
M
NBC (2)

M (80,90)

Of course, NBC realizing this will try to move first as well. Who wins this race to be first will be
the one most effective in arranging a programming schedule in advance. In this example this
works and provides a dynamic explanation of what Nash equilibrium is played in the industry
and why. In other situations, unfortunately, you will have to look for other clues to what players
might play: as there are no obstacles to rushing to be first, everyone will be pulling back the
move to be first (you may have already experienced this in a summer-job interview rush...)
 Economics of Auctions I:
Private Values

I. Introduction.

Auctions, the act of selling an object through a bidding process, have been a common practice
for hundreds, if not thousands, of years. Yet, in recent years there has been a surge in the use of
auctions, especially on the Internet. Millions of sellers and buyers are interacting every day using
various web sites devoted to auctions, most notably E-Bay.

As a manager of a firm, or as an investor, chances are high that you will find yourself
participating an auction or setting up one: auctions are many times used to award contracts for
service provision to the local or Federal governments; real estate public property for commercial
use is also frequently auctioned off; stocks are now offered through auctions in initial public
offerings, to cite a few examples.

In this class we discuss various forms of auctions. The study of auctions is a wide discipline
these days, and we will be able to illustrate only some basic principles that are useful to
managers participating in auctions. We will compare types of auctions and evaluate them from
the viewpoint of the buyer (or bidder). What is the best bidding strategy at an auction? We will
also briefly comment on the type of auctions a seller should choose.

We will see that auctions are just another instance of games, and we will use game theory to
understand how they work.

II. Types of auctions.

In order to understand how to bid in auctions we need to distinguish among different types of
auctions. Auctions can be of a wide variety, and we will focus on the most popular examples. We
group auctions along two lines: according to how the bidding is organized; and according to how
bidders value the object.

1. How the bidding is organized

1.1 Open cry auctions

(auctioneer in the Tokyo fish market)

In an open cry auction the participants act publicly by openly bidding for the item. The two most
important and popular forms of open cry auctions are English and Dutch auctions.

English (or ascending) auction:

The English auction is the typical auction used by the British auction houses such as Sotheby’s
and Christie’s. The auctioneer starts at a low price and calls out progressively higher prices
waiting for a bid before proceeding. If no bidder is willing to pay the last called price, the item
goes to the last, highest bid.

Dutch (or descending) auction:

Dutch auction is named so because of its use at auctions for selling tulips in The Netherlands.
The price starts very high and is descending at every tick of the clock. The item goes to the first
bidder that stops the clock. The winner pays the price displayed on the clock at the moment she
stopped it.
Dutch auctions are commonly used when multiple units of the same products are being sold (for
example Google’s IPO (initial public offering) on the internet in August 2004). As the price goes
down, bidders declare the number of units they want to buy at any given price. The moment the
price reaches a point where all units are sold, a market clearing price, the auction stops and all
buyers are getting the number of units they asked for, paying this lowest price. In a common
variant on the Internet, called Yankee auction, bidders are ranked according to the value of the
purchases they declared with their bid, and each bidder pays the price they bid during the
auction, starting with the one with the highest purchase value.

2.2 Sealed-bid auctions

In a sealed-bid auction the bidders submit their bids secretly and simultaneously. The two most
important and popular forms of sealed-bid auctions are the first-price and the second-price
auction.

First-price auction
In a first-price auction each bidder submits their bid in a sealed envelope and the item goes to the
highest bidder, who pays a price equal to the bid.

Second-price auction
In a second-price auction the process is similar to the first-price, and the item goes to the highest
bid. However, the winner pays a price equal to the second highest bid.
This type of auction is also called a Vickrey auction after the Columbia University economist
who first introduced it and studied it in the 60s. At first, such an auction seems crazy. Imagine
that you put your house for sale, using a sealed bid auction. The highest bid is $1 million, while
the second bid is only $800,000. If you use a second price sealed bid auction you will get for the
house only $800,000 (from a person that was willing to pay $1M!). A more careful analysis,
however, will show that things are not that simple and there is a good reason for using such an
auction.

2. How bidders value the object

It is important to distinguish and analyze auctions not only according to how they are organized,
but also according to how bidders value the object. The two polar cases are represented by
common value and private value objects.

Private value
Here bidders have their own private and subjective valuationb of the object sold. The typical
examples of private value objects are art work, antiques or collectibles, objects that have
different sentimental value to bidders. When a house is for sale and buyers value it differently
depending on their preferences for features and amenities of the house, especially non-typical
characteristics (e.g., a million-dollar recording studio), again we say that it is a private value
object. The private valuation of the object by one bidder is usually unknown to the seller, as well
as to other bidders, and is likely to vary substantially across bidders.

Common value
The item has the same value to the each potential buyer (e.g., treasury bills or oil drilling rights).
In general, however, the exact value of the item is unknown. The participants in the auction have
typically different estimates based on their private information. For instance, a house is a
common value object to developers or real estate brokers who want to purchase the house to
resell it later on a same given market and make a profit. Usually, if there is a resale market for
the object, then the value of the object sold at an auction will be the market price at the second
sale, and the object will have a value common to all bidders, though unknown.

III. Bidding strategies: Private value auctions.

We start our analysis of bidding strategies by looking at private value auctions. We first study
the most intriguing case, second-price sealed-bid auctions. Later we argue that the same analysis
applies to English auctions.

1. Bidding at second-price auctions

Imagine yourself participating in a second-price sealed-bid auction for a case of 6 bottles of
Chateau Margaux 1964, an item that you value, let’s say, at $ 10,000.

The first step in deciding how to bid is to recognize that an auction is a game. A second-price (as
well as a first-price) sealed-bid auction is a simultaneous game. The players are the bidders. The
actions of each player are the possible bids. When you submit your own bid, you do not know
what other auction participants will bid. If you win in a second-price auction, your payoff is
equal to the difference between your valuation and the price you pay, namely the second highest
bid. If you lose, your payoff is zero. As in any game, the payoff to a player depends on her
chosen action as well as on the action profile chosen by the opponents.

b
We define “value” (or “valuation”) as the highest price the bidder is willing to pay for an object.
The second thing you should realize is that uncertainty and information play an important role in
auctions. Not only you don’t know what other bidders will bid, but you do not know for sure
how other bidders value the case of wine!
In fact, if everyone knew everyone’s valuations, the auction would be dull: assuming this is a unique item, the seller
has complete bargaining power and would ask a price equal to the highest valuation among the participants; the
wine case would go to the bidder who values it the most, and the seller would extract the entire surplus.

If you go to wine auctions, most likely you are an expert of wine, and even if you do not know
about the exact valuation your competing bidders have, you may know the distribution of values
from reading specialized magazines, or having gone to other auctions of the same type. In other
words, you have some probabilistic information over the other bidders’ valuations.

The lack of knowledge of other players’ valuations as well as the high number of available
actions (bids) gives the wrong impression that this is a difficult game to analyze: when it comes
to second-price auctions, the opposite is true. We will show that each bidder has a dominant
strategy: to bid her own valuation. Hence, you do not need to guess other bidders’ bids or
valuations, or their probability. You just bid your own value.

Bidding your valuation is a dominant strategy

Recall that an action is dominant if it maximizes the player’s payoff no matter what the other
players do.

Why is it a dominant strategy to bid your own valuation v = 10, 000 rather then a smaller
amount? We will show it by computing, for each conceivable bidding profile of your opponents,
your payoff when you bid your own valuation. Then, we will show that you could not achieve a
higher payoff by changing action.
Let’s assume an arbitrary bidding profile of your opponents. All that matters is actually the
highest of such bids, which we denote by r (the rival bid).

There are two possibilities: a) r is below your valuation v (for example r = $8,000), or b) r is
above your valuation v (for example r = $13,000).
a) If r is below your valuation v there are two key observations: 1) if you win, your payoff is
positive since it is given by v – r, the difference between your valuation and the highest bid of
your opponents (for example, when r = $8,000: v – r = 10,000 – 8,000 = 2,000); and 2) if you
win the wine case your payoff does not depend on your bid since it is equal to v - r !
By bidding your valuation, $10,000, you get the wine case and realize a payoff equal to $2,000.
Bidding less than r results in no win, and your payoff is 0.
Bidding more than r but different than $10,000 still results in a win but the payoff is still of
$2,000.
Conclusion: no bid delivers a payoff above the one generated by bidding your valuation.

b) If r is above your valuation v, then you win only if you bid an amount above r, hence above v.
Your payoff from winning is again v – r, but now this is negative! (For instance, if r = 13,000: v
– r = 10,000 – 13,000 = -3,000). Then, losing the wine case is preferable to winning it.
By bidding your valuation, $10,000, you lose and realize a payoff of 0.
By bidding above r, you win the wine case, but realize a negative payoff, say -$3,000.
By bidding anything else below r you also lose and realize a payoff of 0.
Conclusion: no bid delivers a payoff above the one generated by bidding your valuation.
Therefore, no matter what r is, bidding your valuation v = 10, 000 is a dominant strategy, as you
achieve the maximum payoff for any conceivable bidding profile of the other bidders.

Hence we get the following complete characterization of a second-price auction for the private
value case:
i) Rational bidders bid their valuation.
ii) The winner is the player with the highest valuation.
iii) The price paid by the winner equals the second highest valuation. This is the amount
of money collected by the seller of the item.

It is remarkable that everyone tells the truth in a second-price auction, when it is clear that each
bidder, if they could, would hide their true valuation from the seller (typically pretending that the
wine case is not worth that much to them).

2. Bidding at English auctions

Suppose now that our case of Chateau Margaux 1964 is sold through an English auction. Start at
any moment of the English auction. If the current bid r is above your valuation (say £13,000),
your best action is to stop bidding, leave the room, and realize a zero payoff. If the bid r is below
your valuation (say, £4,000), you can raise your hand (and stay in the auction). At this point, the
bid r would be raised, say by the smallest increment allowed by the house. Let’s assume that the
increments used in this auction are of £100. So you’d bid r + 100 (say, £4,100).
If the auction ended at this point, you would realize a payoff v – (r + 100), which is always
nonnegative (if r = 4,000, then it is 10,000 – 4,100 = 5,900; as long as r < 9,900, it will be strictly
positive).
Clearly, bidding more than r + 100 amounts to taking a useless risk of reducing your final payoff
if you win, and does not give you any benefit, since you can always raise your bid later, while
perhaps nobody valued the wine case more than r + 100. So even if you had the option to raise
the bid by more than the minimum £100, you should not do it!
If the auction does not end, it’s because someone else counters your bid with a higher one. Then,
you repeat the reasoning above, and stay in the auction only if the rival bid is below your
valuation.
Again, you see that we can describe your optimal bidding strategy without guessing how the
other bidders evaluate the wine case. Also, this is a sequential game, so you do not have to guess
the way they bid: you can see it! So the strategy is:
a) bid r + minimum increment if r is below your valuation;
b) exit otherwise.

Who will win the auction? What price will be paid by the winner?
Assume for example that you are the one with the highest valuation, while the bidder closest to
you in valuing the object, Mr. Bond, has a valuation of £4,500. As long as the going price is
below £4,500, both you and Mr. Bond will raise their hand. At exactly £4,500 Mr. Bond will
quit, leaving you to win the wine case.

As the above example shows, the winner in an English auction is the bidder with the highest
valuation and the price she pays equals the valuation of the second-highest bidder, plus the
increment used by the auction house. Given that standard increments are small relative to the
price of the object, we can see the similarity between English auctions and second-price sealed
bid auctions. In both cases the winner is the bidder with the highest valuation and the price paid
for the object (or the seller’s revenue) equals the valuation of the second-highest bidder.

3. Bidding at Dutch auctions

The analysis of Dutch auctions is much more complicated. While in the previous cases we saw
that bidders do not have to worry about other bidders’ valuations when deciding how to bid, here
such knowledge will affect the bidding strategies of all participants. Given the uncertain nature
of such information makes the analysis complicated.

We just give the intuition of what bidding involves. The complete description of the bidding
strategy requires more calculus than we muster in this course.

Think of yourself bidding for the wine case in a Dutch auction. As long as the going price is
above your valuation you have nothing to think about: you stay put and wait. If someone claims
the wine case at the going price, you are better off leaving it to them, because you think it is too
expensive.

First conclusion: in a Dutch auction, never stop the clock if the displayed price is above your
valuation.

As soon as the price reaches your valuation v you have a dilemma: when should you raise your
hand? The sooner you do so, the more likely you are to win, but your gain is lower.

Therefore you sit there and consider the cost and benefit of bidding at below v (raising your hand
later): the benefit is that you increase your profit if you win; the cost is that you risk losing, since
someone else may claim the wine case before you (someone else has a valuation just below
yours).

However, it’s clear that bidding your valuation gives you 0 for sure, but that you can expect not
to do worse by bidding below v (you have a chance of winning at a lower price).

Second conclusion: an optimal bidding strategy requires shading that is, bidding below valuation.

The exact amount of shading requires computing the expected cost and benefit of reducing your
bid. Any knowledge of the valuations of other bidders could give you an advantage. For
example, if you knew that the bidder with the valuation closest to you has a valuation of $8,000,
you could stay put until the price goes down to $8,010 (price decreases by $10 each second of
the auction).
To complicate things further, notice however that it’s not clear that you should raise your hand at
$8,010, because the other guy might also consider shading, something you must take into
account. If everyone engages in shading, even if you knew that the second-highest valuation
were $8,000, when the price reaches $8,010, you could only…start getting nervous again!

Third conclusion: in a Dutch auction, you need to guess the other bidders valuations as well as
their bidding strategies, in order to choose your own exact optimal bid. The optimal equilibrium
shading equates the expected marginal cost of shading to its expected marginal benefit.
In the game theory language, we are looking for a Nash equilibrium, but of a game complicated
by the presence of uncertainty (that is, bidders do not know each others’ payoffs).

4. Bidding at first-price auctions

We will show that first-price sealed-bid and Dutch auctions have identical outcomes. As with
Dutch auctions, a precise analysis of the bidding strategy and the Nash equilibrium of the first-
price auction is complicated. As with Dutch auctions, it is nevertheless possible to convey the
main ideas of the optimal bidding strategies. Remember that with the first-price sealed-bid
auction we pay what we bid. Again, we get an immediate first observation:

First conclusion: in a first-price auction, never bid above your valuation.

Should we bid our valuation in a first-price sealed-bid auction? Again, same reasoning as for
Dutch auctions suggests that you cannot get less, and you may get more if you bid below your
valuation. In a first-price auction we may or may not win if we bid exactly our valuation
(depending on the bids of the opponents), but in either case we will realize for sure a payoff of
zero!

Second conclusion: an optimal bidding strategy requires shading that is, bidding below valuation.

As before, the exact optimal strategy can be derived with calculus. By shading the bid, we
increase the probability that the bid of someone else will be above ours, but we increase our gain
in case of win. Thus

Third conclusion: the equilibrium bidding strategy requires the bidder to equate the marginal cost
of shading to the marginal gain.

In fact, one can show that the Nash equilibrium bid b(v) when:
„ your valuation is v ;
„ every bidder’s valuation is independent and identically distributed;
„ every bidder is risk neutral, is

b(v) = E ( r | r < v),

that is, bid the expected value of the second-highest value r computed as if your valuation turns
out to be the highest.

Example: if everyone’s value for the wine case can be between 0 and $15,000, with equal
probability, your valuation is v = $10,000, and there are 10 bidders. Your equilibrium bid then is

b(10,000) = E ( r | r < 10,000) = (n-1) v / n = (9/10) (10,000) = 9,000.

You can see that for first-price auctions on average the item goes to the bidder with the highest
valuation who pays the second-highest valuation in the auction. Hence, on average the seller
collects the second-highest valuation in the auction.
5. Issues in bidding: collusion

There is of course an incentive for bidders to come up with a strategy that defeats the seller’s
efforts to pitch bidders against each other: collusion.

With few bidders, a second-price sealed-bid auction can be defeated by submitting a high bid,
and a low-ball second-highest bid (around the minimum allowed). Such a bidding strategy in
first-price auctions is more difficult to sustain as every bidder has an incentive to deviate, pretty
much as in the prisoner’s dilemma kind of situation. Collusion can then be sustained if bidders
find themselves participating at many auctions.

In open cry English auctions collusion can be obtained through communication at first rounds of
bidding. In 1997, the FCC auctioned U.S. airwaves for personal cellular service, and some firms
found a way to communicate which airwaves each wanted to get by using code-bidding: the last
three digits of a bid were FCC codes or telephone area codes for specific regions a bidder was
interested in…sometimes colluding bidders represented touch-tone keypad equivalents of their
names!

On the seller side, and in open cry versions more than in other types of auctions, sometimes
sellers can use an undercover bidder to plant false bids and create ‘frenzy’ during an auction, in
the hope to raise the final revenue. Such a practice is known as shilling.

Needless to say, all such behavior is illegal: but you should be warned of the existence of such
schemes and protect yourself from that.
 Economics of Auctions II:
Common Values and Seller’s Strategies

I. Introduction

We continue the description of bidding strategies in auctions, by tackling the common value
case. We then briefly explore the implications for the optimal selling mechanism for the seller.
This short guide through auctions should alert you as a manager of the issues involved in bidding
or setting up an auction, and you should have a qualitative understanding of the problems
involved. We will not pursue a formal analysis of common value auctions, though.

II. Bidding Strategies: Common Value Auctions.

In common value auctions all participants attach the same value to the item. However, since the
precise value of the item is unknown, each bidder forms their own subjective estimate of that
value. For instance when auctioning drilling rights for crude oil, each oil company has some idea
about the value of tract. These estimates vary across companies depending on the quantity and
quality of their private information, e.g., how close they operate to the tract, the results of the
sample extractions and so on. It is immediately apparent that one important feature affecting the
analysis of any type of common value problem is the nature of the information across
participants.

The basic idea when you bid in common value auctions is the following: since all bidders receive
information about the common value of the item auctioned off, the information other bidders
have is valuable to every bidder.

The problem is, of course, that this information is private to each bidder. However, each bidder
reveals her information while bidding. In particular, when the outcome of the auction is
announced, so that for instance we have won the item, this reveals to us something about the
information others have!

We should therefore keep this in mind when bidding in the first place: the fact that we win is in
itself useful information, and should be used to establish what value the item for sale has.

In this course we will not go into the details of this problem, but rather we give a general idea for
the simplest conceivable case, i.e., for a situation where the estimate of each bidder is an
unbiased, identically, and independently distributed random variable.

Naïve bidding and the winner’s curse.

Suppose that we are participating in a first-price sealed bid auction for oil-drilling rights over a
given tract of sea. This is a classical common value object. Let’s say that the true value of the
lease is $1 billion (although this is unknown to everyone, including the seller).

Of 10 oil companies bidding, each company gets an independent estimate of the value of the
underground oil field by forming a scientific team, headed by one of the top geologists in the
world as exclusive consultant. Each company’s team is trained the same way and uses the same
testing methods. From previous experience, each estimate has a $100 million maximum error, so
that all numbers between $.9 and $1.1 billions are equally likely to occur. Notice that each
company’s average estimate will be equal to $1 billion, exactly the true value of the lease. This
means that on average the expert geologist each bidder consults is 100% accurate: it is unbiased.
Moreover, with a large number of firms, the average of the actual estimates across all teams is
going to be equal to the true value.

Suppose we prescribe to each bidder to bid her own estimate of the item’s value. We call this
bidding strategy naïve. The reason will become apparent in a moment.
Q: Who is going to win the auction?
A: The bidder with the highest estimate.
Q: How much is she going to pay?
A: Her own estimate.
Q: What will be the payoff of the winner?
A: It will be negative!

The reason is simple. The average across participants is equal to the true value of the object. The
winner has the highest estimate which is therefore greater than the true value. Hence, her payoffs
are negative. This phenomenon is called the winner’s curse.

Inexperienced bidders tend to bid their valuation and, therefore, they acquire the object when
typically its value is below the price that they pay.

To reiterate the explanation of the winner curse, we ask the following question. Why is the
valuation (estimate) of each bidder correct on average, while, when everybody bids their
estimate, the winner of the auction (which is one of the participants) always overestimates the
item’s true value? The reason is simple. The auction selects as a winner the bidder with the
highest estimate. Because the average across participants of the estimate coincides with the item
value, the highest estimate across participants is by definition above the average estimate and,
therefore, above the item value. Thus, the naïve winner always overestimates the item’s value.
The conclusion holds in second-price auctions, in fact, in all types of auctions we have seen.

To put it in statistical language, each bidder tries to estimate the unknown value v of the object
sold when the bidder wins. The correct estimate is not the expected value of v given just the
bidder’s own information from testing, x,

E( v | x )

but it is the expected value given that information x plus the fact that the bidder wins, which
happens because she has the highest bid, b > max bj, where the maximum is taken across all
other bidders’ bids,

E( v | x, b > max bj )

To sum up, in common value auctions, the optimal bidding strategy always requires shading the
bid below the bidder’s valuation (estimate). The more bidders there are, the smaller the
adjustment has to be. The specific form depends on the probabilistic laws governing the value of
the item and the estimates of the participants, and also on how we expect other bidders will bid,
relative to their estimates: we are looking for a Nash equilibrium. The computation of Nash
equilibrium bidding formulas goes beyond the scope of the course.

III. Auctions on the Internet.

More than $60 billion worth of sales are made each year on the web these days. Auction sites
such as eBay, Amazon and many others make money on listing fees, seller commissions,
advertising revenues. What kind of auction types do we see on the Internet?
It turns out that many items are sold online, each single item sometimes through different
methods, ranging from English to Dutch auctions, and more.

When English auctions are offered (say, on eBay) the bidder is offered the possibility to place a
direct bid, or to use a proxy. Since the auction is open continuously for such a long time, buyers
often find convenient to use such a proxy system. If a proxy is created, the bidder enters a
maximum price that she is willing to pay for the object, her bid. Then, the computer at eBay
hides the bid and keeps placing a bid equal to the current rival bid plus the minimum increment
for the item, as long as the current bid is below the hidden bid.

You see that this is the kind of strategy we invoked as optimal in English auctions. Online
auctions however are modified versions of the standard live ones mainly because the live
versions have an open end, while these online versions have a deadline, typically within 7 days
from the posting of the item for sale.

When deadlines are present, bidders sometimes use a bidding system called sniping: they wait
until the very last seconds the auction is open and then submit their direct bid. In this case, the
item is awarded to the highest such bid, at the second-highest price. This makes such auctions
effectively second-price.

If you snipe, your optimal strategy is then to bid your valuation (when the object is private value,
such as stamps, antiques, collectibles, and so on).

Sniping gives the advantage of calming the auction, and avoids feeding the frenzy sometimes
occurs in online auctions of the English type. For common value objects, it hides from other
bidders information the bidder may have regarding the object. Sniping is not without cost: you
may be too late, you may also be stolen your ID while you bid through a sniping server…!

IV. Selling Objects Through Auctions.

Now that we have a better understanding of how to bid in auctions, we turn to briefly asking
what the seller of an object should do to maximize her profits from the sale. Since in many cases
of interest the production cost is already entirely sunk, this amounts to asking what mechanism a
seller should use to maximize revenues. Again, we establish general principles, but do not
provide the details.

The first thing to ask is: why an auction? When should a seller use an auction?

We already saw that if we have an apartment and we want to put it up for sale, if there is no
uncertainty over the demand, what to do is trivial: sell it to Mr. Rich, a C.E.O., the customer with
the highest reservation price, and charge him his reservation price. Obviously, this will work if
the apartment is somewhat unique, and we are ‘a monopolist’ which perfectly price
discriminates. Since we have only one unit to sell, we only sell to the customer with the highest
reservation price. We are able to charge their reservation price if we have complete bargaining
power over them (this is the only apartment they are considering for purchase, they need an
apartment to move in a.s.a.p., the apartment has a unique location, and Mr. Rich’s wife adores
the living room with bay windows and a XIX century fireplace…). Otherwise, we can at least
ask for the second-highest price, threatening the buyer to sell the house to Mr. Jones, a young
and upcoming manager.

Things become more difficult if we do not know the demand, both in terms of number of
customers out there and their reservation prices (how much they are willing to pay for the
house). Then an auction allows finding out what the market price is for the house.

Of course, this will come at a price, since now buyers have an incentive to hide their own
valuation from the seller. We already saw that if a €2 million apartment is sold through a second-
price auction, and we are in a private value situation, the seller would end up getting only €1.8
million (if this is the second-highest valuation, say). This looks like a steep price to pay for
getting buyers to honestly reveal their valuation! The seller cannot do better but corner the
highest-valuation buyer to the second-highest valuation price.

In order to lure buyers to be honest and reveal their valuation to us (through their bidding), an
auction pitches buyers against each other. Another solution would be negotiating with each buyer
independently. It turns out that, as intuition suggests, competition among bidders increases the
potential revenue for the seller, relative to independent and separate bargaining with each buyer.
The cost of extracting the information from the buyers is minimal when an auction is used.

If auctions are best suited to extract information from buyers, which auction format does the job
best? Which one maximizes the revenues of the seller?

In private value environments, we already saw that second-price auctions achieve truthful
revelation of the buyers’ valuations through bidding, since each buyer bids her valuation. Using
an English auction would not change the average revenues of the seller, since also in that case
the winner pays the second-highest bid valuation. It turns out that, when bidders are risk neutral,
any of the 4 formats we saw (English, second-price, Dutch, first-price) give rise to the same
revenues for the seller on average: this is known as the revenue equivalence theorem.

Experimental and statistical evidence shows overbidding in second-price auctions (indicating

that people fall prey to the winner’s curse), but not in English auctions. Online Dutch auctions
seem to fetch 30% more revenues than first-price sealed-bid ones.

What situations could make a difference? Here we list a couple, without getting into the details.

Risk aversion

We have talked about risk neutral bidders, but sometimes (think about bidding for a firm your
company wants to buy) bidders can be averse to risk. They may be more concerned with the loss
due to underbidding than with the cost of winning the object at somewhat a higher price. Then,
risk averse bidders bid more aggressively when it pays. This is the case of first-price, or Dutch
auctions: they will reduce shading, and bid closer to their valuation.
With risk aversion, on average revenues are higher with a first-price than with a second-price
auction.

Common value

When estimates are correlated, and we are auctioning off a common value item (such as U.S.
Treasuries, stocks, commercial property), the revenue equivalence across auction types falls
apart.

First of all, there is more information disclosure in an English auction than in any other. The
more information a bidder has, the more accurate her estimate. The final result on bids depends
on the bidding strategies and on the correlation existing across estimates.

For example, if optimal equilibrium bids are increasing with the estimates, and estimates are
positively correlated, more information implies more aggressive bids.

An English auction will provide higher average revenues than a second-price auction (which is
sealed bid, and therefore less information disclosure through bids makes bidders less aggressive),
and this in turn will provide more revenue than a first-price or Dutch auction.

English auctions tend to display hyperactivity when there is good news, and bidding behavior is
feistier than in Dutch auctions (where nothing may happen for long periods of time).

[I don’t expect you to find this obvious for many reasons. One is that when estimates are
positively correlated, and I observe a low estimate, say $.9 billion in the previous oil drilling
example, I consider more likely that everyone else also got a low estimate, and that the value is
low, and therefore I increase shading; but if I receive a high estimate, say $1.1, I consider more
likely that everyone got a high estimate and will shade less! It turns out that the total effect on
bidding is to make it more conservative when the outlook is gloomy than more aggressive when
it is bully.]

In such common value cases, the number of bidders also plays a role in deciding which type of
auction to use. When the number of bidders is small, say 2, the result of an English auction can
be totally different if valuations are positive or negative correlated: if both bidders have received
positive news on an acquisition target, they will push up the price very high; if one received good
news, but the other received bad news, the seller will end up giving the item away to the
optimistic guy at the pessimistic guy’s price!