0% found this document useful (0 votes)
16 views28 pages

11 October 2024 Game Theory-2

The document contains study notes on game theory, covering basic definitions, examples like the Prisoner's Dilemma and Battle of the Sexes, and concepts such as dominated strategies and Nash equilibrium. It provides normal form representations of various games, illustrating player strategies and pay-offs. The notes aim to help understand the structure and analysis of strategic interactions in games.

Uploaded by

Jordy Kenfack
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
16 views28 pages

11 October 2024 Game Theory-2

The document contains study notes on game theory, covering basic definitions, examples like the Prisoner's Dilemma and Battle of the Sexes, and concepts such as dominated strategies and Nash equilibrium. It provides normal form representations of various games, illustrating player strategies and pay-offs. The notes aim to help understand the structure and analysis of strategic interactions in games.

Uploaded by

Jordy Kenfack
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 28

Game Theory Study Notes

Safak Ozden

November 4, 2024
Contents

1 Basic Definitions 2
1.1 Prisoners Dilemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Normal form representation of Prisioner’s Dilemma . . . . . . . . . . . . . . . 3
1.2 Battle of the Sexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Normal form representation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 A game without a story . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Rotten father with two pennies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Example 5: Stag Hunt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5.1 Exercise 6: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Stupid or Chicken . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Dominated Strategies 11
2.1 Iterated Elimination of Dominated Strategies . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Procedure for IEDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Implications of IEDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Nash Equilibrium 16
3.1 Best Response Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Nash Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Cournot Duopoly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1
Chapter 1

Basic Definitions

Last week we have talked about representing games in normal form and converted some informal
explanations into normal form games. We have also studied ways of comparing strategies for a given
player. Let us recall these and then start learning ways of solving games.

Remark 1. Note: For simplicity, I’ll consider games with 3 players: n = 2, 3. This is to make the
notation accessible to those who are not familiar with mathematical formalism.

Definition 1. A normal form game with 3 players consists of the following data:

1. Players: Throughout the course, we will denote them with P1 , P2 , P3 .

2. Strategies/Actions: For each player Pi , there is a set S (i) called the set of strategies. Elements
of S (i) will be denoted as s(i) .

3. Pay-offs: For each player, there is a utility/pay-off function ui , which assigns a real number
for each combination s(1) , s(2) , s(3) .

Remark 2. The last item in the definition is tantamount to say, for every combination s(1) , s(2) , s(3)


of actions/strategies of the three players we have a number:

u1 (s(1) , s(2) , s(3) ) called payoff/score of player P1 ,

u2 (s(1) , s(2) , s(3) )) called payoff/score of player P2 ,


u3 (s(1) , s(2) , s(3) ) called payoff/score of player P3 .

1.1 Prisoners Dilemma


We first explain the scenario.

1. Scenario: Two individuals are arrested for a crime. Each prisoner is offered a deal:

- If only one of them betrays the other, the one who betrays go free while the other gets
a full sentence of 20 years.
- If both betray, they receive medium sentences, both 5 years.

2
- If both stay silent, they receive light sentences, which is 1 year for each.

2. Preferences: Each prisoner wants to receive the shortest sentence possible and does not care
about how long the other prisoner is sentenced for. This means, if possible, each of them
wants to walk away, and if not, serve only 1 year, etc.

1.1.1 Normal form representation of Prisioner’s Dilemma


Let us write Prisoner’s Dilemma in normal form, for which. For such a task, one should start with
naming the players, and then the possible moves that they can select from are same and they are:

1. The players: We denote them with P1 and P2 .

2. The strategies: Each of them has only two options. Either Betray or silent. This means, for
each of the players, we have

S (1) = {Betray, Silent}


S (2) = {Betray, Silent}

3. Pay-offs / Scores Now let us talk about their scores (or pay-offs). The definition tells us that,
we should consider all possible combinations of both players like:

First possible combination = (Betray, Betray)


Second possible combination = (Betray, Silent)
Third possible combination = (Silent, Betray)
Fourth possible combination = (Silent, Silent)

Again, according to the definition we should assign values/scores to each possible combination
listed above, one for each player. Since they want to minimize their sentences, it is reasonable
to give the punishment time as the score to each player. More precisely, we set the scores of
the first player depending on the combination of the choices of the individuals as follow:

u1 (First possible combination)) = u1 (Betray, Betray) = −5


u1 (Betray, Silent) = 0
u1 (Silent, Betray) = −20
u1 (Silent, Silent) = −1

For example, if you want to express the second equality in English but use mathematical
formulation, you should say

The pay-off u1 of P1 is −20, if P1 selects Silent and P2 selects Betray, (1.1)

but it is in fact saying that

If the first suspect stays silent but his friend snitches, the first suspect will be sen-
tenced to 20 years.

Exercise 1. Rewrite the previous sentence, using the expressions:

3
1. P1 ,

2. −20,

3. selects Silent,

4. P1 ,

5. pay-off u1 ,

6. selects Betray,

7. P2 .

Exercise 2. Convince yourself that the pay-off function u2 of the second person is as follows:

u2 (First possible combination)) = u2 (Betray, Betray) = −5


u2 (Betray, Silent) = −20
u2 (Silent, Betray) = 0
u2 (Silent, Silent) = −1

We end the first example by putting everything into a compact form via the table below.

P1 /P2 Betray Silent


Betray (-5,-5) (0,-20)
Silent (-20,0) (-1,-1)

Table 1.1: Prisoner’s Dilemma in table form

1.2 Battle of the Sexes


This game is about a couple, deciding on what to do on special night.

1. Scenario: A couple is decided to spend Friday night together and go to an event. Unfortu-
nately, they both forgot their decision on which event they will attend, and has to buy the
ticket alone. So, each of them will buy a ticket either to Football match or to a Concert.

2. Preferences: Each partner wants to attend their preferred event but would rather go to the
same event together than attend separate events. That means, for each of them going to an
event is better than going alone, but it is even better to go to the event of their own desire.
Suppose they buy tickets separately, without knowing the other’s choice.

1.2.1 Normal form representation:


1. The players: We denote the first partner with P1 and the second one with P2 .

4
2. The strategies: Each of them has only two options for the event that they go. Either F ootball
or Concert. This means, for each of the players, we have

S (1) = {F ootball, Concert}


S (2) = {F ootball, Concert}

3. Pay-offs / Scores Now let us talk about their scores (or pay-offs). As before, we should
consider all possible combinations of both players like:

First possible combination = (F ootball, F ootball)


Second possible combination = (F ootball, Concert)
Third possible combination = (Concert, F ootball)
Fourth possible combination = (Concert, F ootball)

Again, we should assign values/scores to each possible combination, one for each player. But,
contrary to the previous game, now we are not given any sort of numbers that we can use to
decide what the score of each player will be. However, we are now given players’ preferences
over the combination of tickets. So, we will use these preferences to set made-up numbers,
but keeping in mind the fact that, there numbers should reflect the preferences of the people
in question. We are told that, for one player, we have the following preferences:

(F ootball, F ootball) > (Concert, Concert) > (F ootball, Concert) > (Concert, Concert)

The numbers that we will assign to the possible combinations should reflect the above in-
equalities. This means the numbers we need should satisfy the following inequalities.

u1 (F ootball, F ootball) > u1 (Concert, Concert) > u1 (F ootball, Concert) > u1 (Concert, Concert)

For the choice of numbers, we are free, considering the above inequalities. So, let us set

u1 (First possible combination)) = u1 (F ootball, F ootball) = 4


u1 (Concert, Concert) = 3
u1 (F ootball, Concert) = 2
u1 (Concert, F ootball) = 1

But of course, we could have set

u1 (First possible combination)) = u1 (F ootball, F ootball) = 10


u1 (Concert, Concert) = 9
u1 (F ootball, Concert) = 2
u1 (Concert, F ootball) = 1

which would be equally meaningful. But of course, we would like to put an emphasis on the
relative values, not just the orders. For example, if the player one tells us (or if we believe)
that going together is much more important than going to football, then the second assigning
would be more realistic.

5
P1 /P2 Football Concert
Football (10, ) (2, )
Concert (1, ) (9, )

Table 1.2: Battle of Sexes

Exercise 3. Complete the following table, with your desired assignments for the second player
pay-offs.
For example, if you want to express the second equality in English but use a mathematical
formulation, you should say
The pay-off, of the player P1 when P1 selects F ootbal and P2 selects Concert, is 2, (1.2)
but it is in fact saying that
If the first person goes to the football match but the second goes to the concert, the first
person satisfaction level is 2.
In case you want to be extremely formal, then you should write
u1 (Football, Concert) = 2 (1.3)
Exercise 4. Convince yourself that the pay-off function u2 that I present below reflects the prefer-
ences of the second player, and table that follows is the corresponding table:

u2 (First possible combination)) = u2 ((Football, (Football) = 5


u2 ((Football, (Concert) = 4
u2 ((Concert, (Football) = 0
u2 (Concert, Concert) = 100

P1 /P2 Football Concert


Football (10,5) (2,4)
Concert (1,0) (9,100)

Table 1.3: Battle of Sexes in table form

1.3 A game without a story


In the lecture, we have mentioned that, there are two parts of game theory. One is to turn real life
situations into games. The other one is, solving, analysing game, meaning that finding successful
ways to play games. We also covered different stories where they induced the same table. This
means, when we want to analyse a game, we don’t really need to care about the underlying story,
as long as we are given the table.

6
Don’t
P1 /P2 Go left Go right
move
Go up (5,5) (3,7) (1,4)
Go down (2,6) (1,1) (0,1)

Table 1.4: Where is the story :)

1.4 Rotten father with two pennies


In this example, we consider a criminal.

1. Scenario: A rotten father is torturing his two sons in the following way. He first forces his
sons to flip two coins, and punishes them then.

2. Pay-off ’s:

(a) If both coins have the same results, the father punches the older child, and let the younger
one untouched.
- In the other case, the father punches the younger child, and let the older one untouched.

Exercise 5. Assume that both children want to avoid their lunatic father, and skips the punch.

1. Write down the set of possible outcomes

2. Write down the preference relationship of the brothers on the set of results of coin flips.

3. For each brother, assign a numeric value to each possible outcome that reflects the preference
of the brother.

4. Write down the table of the game, if this is a game at all :)

Solution:

1. The possible outcomes of the above scenario are:

(__, __), (__, __), (__, __), (__, __). (1.4)

2. The older brother’s preference relationship is:

(__, __)__(__, __)__(__, __)__(__, __) (1.5)

and younger brother’s preference relationship is:

(__, __)__(__, __)__(__, __)__(__, __) (1.6)

7
Solution continued:

3. The older brother’s utility function can be:

u1 (__, __) = __
u1 (__, __) = __
u1 (__, __) = __
u1 (__, __) = __

while the younger brother’s utility function can be:

u2 (__, __) = __
u2 (__, __) = __
u2 (__, __) = __
u2 (__, __) = __

4. Write down the table of the game, if this is a game at all :)

old/young __ __
__ (_,_) (_,_)
__ (_,_) (_,_)
Horrible father!

Exercise 6.
Now consider that horrible father and now assume that the younger brother is of a naive altru-
istic, and prefers to get slapped and saves his brother, while the brother is his father’s son. Repeat
the previous question using this new info.

Solution:

8
1.5 Example 5: Stag Hunt
Two hunters have the choice to either hunt a stag together, which requires cooperation, or each
individually hunt a hare. Hunting the stag provides more food, but both hunters need to cooperate.
Hunting a hare provides less food but can be done individually. Each hunter prefers the larger
reward of hunting the stag but does not want to go after the stag unless the other hunter also
commits. If they don’t cooperate, they prefer to hunt a hare for the smaller but guaranteed reward.

1.5.1 Exercise 6:
Using the above info, write down the question by providing scenario, explain the pay-offs as done
in the previous exercise. Then, solve the question you have written.

9
1.6 Stupid or Chicken
Two drivers approach each other at an intersection where there are no rules on who should go first.
Both want to avoid a crash but would prefer to keep driving straight without yielding. However,
each driver would prefer yielding over a crash. Each driver prefers to keep driving straight but will
yield if it helps avoid a crash. Neither wants to crash.

Exercise 7. Repeat the previous exercise.

10
Chapter 2

Dominated Strategies

Now that we are accustomed to the formalism of defining games, we can switch our attention to solv-
ing games. To this end, let us consider a general normal form game G = (S (1) , . . . , S (n) ; u1 , . . . , un )
with n players, where, as usual, the players are denoted with P1 , . . . , Pn . The first definition that
we are going to present below gives us a checklist process which helps us to pinpoint strategies that
one should never do.
Definition 2. A strategy si ∈ S (i) for a player Pi is said to be dominated if there exists another
strategy s′i ∈ S (i) such that, for all possible strategies of the other players, s′i yields a higher payoff
for Pi than si . In other words,

ui (s1 , . . . , si−1 , s′i , si+1 , . . . , sn ) > ui (s1 , . . . , si−1 , si , si+1 , . . . , sn ) for all s−i ∈ S (−i) ,

for every possible s1 ∈ S (1) , . . . , si−1 ∈ S (i−1) , si+1 ∈ S (i+1) , . . . , sn ∈ S (n) . A dominated strategy is
considered irrational, as the player can always achieve a better outcome by switching to s′i .
What does this definition tell? Let us investigate it, in a way that you should do all the time
when someone gives you a definition:
1. It (the definition) tells us, under which conditions we can call a strategy si is a dominated
strategy.

2. It says that there should be a second strategy s′i , and this second strategy should dominate
the first one. But how?

3. It says that, whatever other players play, the pay-off of the i-th player is more, if s/he plays
s′i , instead of si .
So, if we want to check whether a strategy is dominated, we should evaluate it against a second
strategy, and evaluate their payoff s considering all possible moves of other player. Let us exemplify
what we have said, by considering dilemma of poor prisoners. First, let us recall the dilemma of
the prisoners.
Let us ask if the first prisoner’s strategy s1 =Silent is dominated by the strategy s′1 =Betray.
For this, we should consider, all moves that the second prisoner can play. They are Silent and
Betray. Now, we should check if

u1 (s1 , Silent) < u1 (s′1 , Silent) (2.1)

11
P1 /P2 Betray Silent
Betray (-5,-5) (0,-20)
Silent (-20,0) (-1,-1)

Table 2.1: Prisoner’s Dilemma in table form

as well as

u1 (s1 , Betray) < u1 (s′1 , Betray) (2.2)

We can do this, by using the table. The first line becomes

−1 = u1 (Silent, Silent) < u1 (Betray, Silent) = 0 (2.3)

This means, for the first prisoner, choosing Betray is better than choosing Silent, if the second
prisoner chooses Silent. Nice, the first inequality that we need to check is validated. When we use
the information in table, we see that

−20 = u1 (Silent, Betray) < u1 (Betray, Betray) = −5 (2.4)

Now that we know how to check if a strategy is dominated by another strategy, the next step is
using this knowledge to solve games. To facilitate the idea, recall that we have seen that whether
the second prisoner chooses Betray or Silent, it is always better for the first prisoner to choose
Betray. So, the first prisoner has no reason to choose Silent. In short, since Silent is dominated
by Betray, we can (also the second player can) assume that, the first player will always choose
Betray. Knowing this, the situation changes. In reality, the situation is as follows:

P1 /P2 Betray Silent


Betray (-5,-5) (0,-20)

Table 2.2: Prisoner’s Dilemma new reality

We now know, that the first prisoner will do. But this helps us to determine, what the second
prisoner will do as well. It is clear that, in this new situation, the second prisoner’s strategy Silent
is dominated by the strategy Betray. By reasoning in the same way, we see that, the second
prisoner will not choose Silent ever. As a result, there is only one possible option for both of them,
namely Betray. Thus, when we remove the dominated strategies for both players, the table of what
is left is depicted below:

P1 /P2 Betray
Betray (-5,-5)

Table 2.3: Prisoner’s Dilemma, inevitable result

Now we know why this game is called prisoner’s dilemma. Let us generalize what we have just
presented.

12
2.1 Iterated Elimination of Dominated Strategies
In many games, dominated strategies can be systematically removed to simplify the analysis, a
process known as iterated elimination of dominated strategies (IEDS). This procedure can
reveal essential aspects of the game’s structure and, in some cases, lead to a single remaining strategy
profile, which may be a solution to the game.

2.1.1 Procedure for IEDS


The steps of IEDS are as follows:

1. Identify Dominated Strategies: Begin by examining each player’s strategies to identify


those that are strictly dominated by another strategy.

2. Eliminate Dominated Strategies: Remove dominated strategies from the set of available
strategies for each player.

3. Reassess Dominated Strategies: After eliminating certain strategies, other strategies may
become dominated under the updated strategy set. Repeat the identification and elimination
process.

4. Repeat Until No Dominated Strategies Remain: Continue the process iteratively until
no dominated strategies are left for any player.

2.1.2 Implications of IEDS


IEDS can often reduce complex games to simpler versions, making it easier to analyze player behav-
ior. However, it is essential to note that IEDS is most effective when there are strictly dominated
strategies. In cases where there are only weakly dominated strategies (i.e., strategies that yield
payoffs at least as good as other strategies in some cases), IEDS may not yield a unique solution or
may not be applicable without additional assumptions.

Exercise 8. In this exercise, we will apply IEDS to the game presented in the third example,
namely:

P1 /P2 A B C
X (5,5) (3,7) (1,4)
Y (2,6) (1,1) (2,0)

Table 2.4: Where is the story :)

1. We start with a warm-up, and try to decide if Y is dominated by X. Recall that we should
check, if the following inequality
u1 (Y, s) < u1 (X, s)

13
holds for each strategy s available to the second player. More precisely, we should check, if

u1 (Y, A) < u1 (X, A)


u1 (Y, B) < u1 (X, B)
u1 (Y, C) < u1 (X, C)

Now, it is your turn. Use the table and check if these inequalities hold. Recall that, in order
to be dominated, all of these inequalities should hold. If you have one example that the
inequality fails, than BINGO, it is not dominated. Yeah, because of that failed inequality.
So, either show that all inequalities hold, or just write down your reason for not believing the
dominance. Write your answer to the space below.

Solution:

2. Write down the inequalities that you need to check, if you are asked to show C is dominated
by A.

Solution:

3. If you found that C is dominated by A, remove the column that corresponds to C draw the
table.

Solution:

14
4. Check, which one is dominated, A or B. If neither, provide the inequality that breaks the
dominance in each case. If one is dominated, remove the it from the table.

Solution:

5. Redo the first part. Check if Y is dominated by X. If your answer is changed, explain the
reason. If Y is dominated, update the table by removing the corresponding part from the
game.

Solution:

6. Redo the 4th part.

In the exercises above, using IEDS, we have successfully predicted the outcome of the games.
However, there are many cases, IEDS does not yield a precise solution. In other words, we cannot
eliminate all but a single pair of strategies (s1 , s2 ). The next exercise is an example of such a
situation.

Exercise 9. Consider the game whose table form is given below. First, show that, neither X

P1 /P2 A B C
X (2,4) (4,0) (5,3)
Y (4,2) (0,4) (6,3)
Z (3,5) (3,5) (9,7)

dominates Y , nor Y dominates X. Then continue by considering all meaningful tuples, and conclude
that, IEDS does not remove any row or column from the presented game table.

15
Chapter 3

Nash Equilibrium

In the last exercise of the previous chapter, we have seen that iterated elimination of dominated
strategies does not help us for all games. So, in our pursuit of finding ways of predicting how games
will unfold, we should introduce other ways of comparing the desires of the players. The stones of
our sought way are paved by J. Nash, through the concept of Nash Equilibrium. However, before
presenting the definition, it would be better to introduce a helper notion, so called the best response.

3.1 Best Response Function


Let us consider an abstract game G with n players, where, as always, strategies available to players
are denoted by S1 , S2 , . . . , Sn and the utility functions are denoted by u1 , u2 , . . . , un , respectively.
Let Pl be a player that we are focusing on, and assume that all other players have picked their
strategies. That is, si is selected by Pi , for all players other than the l-th player. Then, in this
scenario, Pl can decide what to play by considering how his payoff changes depending on his available
actions. Of course, as he knows what other players play, he should pick the strategy that brings
him the highest payoff.
For example, assume that Pl has only 3 strategies, namely A, B, and C. For the sake of
argument, assume that we are focusing on the second player, just to make the notation easier to
follow. So the second player knows that P1 played s1 , P3 played s3 , etc. So, he just needs to
compare the values

u2 (s1 , a, s3 , . . . , sn )
u2 (s1 , b, s3 , . . . , sn )
u2 (s1 , c, s3 , . . . , sn )

Since P2 wants to maximize his payoff, then his best behavior, knowing other players played as
above, is to choose the strategy among A, B, and C which brings the highest payoff. Again, for the
sake of argument, let us assume that

u2 (s1 , a, s3 , . . . , sn ) = 5
u2 (s1 , b, s3 , . . . , sn ) = 8
u2 (s1 , c, s3 , . . . , sn ) = 4

16
So, P2 would pick B. I think you would not feel uncomfortable if I summarize all this discussion by
just saying

The best response of P2 to the combination of strategies s1 ∈ S1 , s3 ∈ S3 , . . ., sn ∈ Sn


is B.

Let us illustrate the discussion with a normal form game with 3 players. Assume that all players
have the same set of strategies consisting of two options, and the game is presented as below:

Strategies Player 1’s Payoff Player 2’s Payoff Player 3’s Payoff
X, K, A 4 4 4
X, K, B 1 3 5
X, L, A 2 5 1
X, L, B 2 2 4
Y, K, A 5 1 3
Y, K, B 2 4 2
Y, L, A 3 2 5
Y, L, B 1 1 1

Table 3.1: Payoff Table for Players 1, 2, and 3

Let us focus on the third player P3 , and find his best response in various situations.

1. The case where P1 plays X and P2 plays K. In this case, the best response of P3 is B, because

u3 (X, K, A) = 4
u3 (X, K, B) = 5

Since 5 > 4, P3 prefers B.

2. The case where P1 plays Y and P2 plays L. In this case, the best response of P3 is A, as

u3 (Y, L, A) = 5
u3 (Y, L, B) = 1

Since 5 > 1, P3 prefers A.

3. The case where P1 plays X and P2 plays L. In this case, the best response of P3 is B, as

u3 (X, L, A) = 1
u3 (X, L, B) = 4

Since 4 > 1, P3 prefers B.

17
The best response idea is simple and something we are familiar with from life. Now, we will create a
function using the idea. To do so, first notice that the above discussion can be pictured as follows:

Input −→ Output
Some combination of strategies −→ Another strategy
A combination (s1 , s2 ) of strategies of P1 and P2 −→ Best response of P3 to (s1 , s2 )
(s1 , s2 ) = (X, K) −→ Best response: B
(s1 , s2 ) = (Y, L) −→ Best response: A
(s1 , s2 ) = (X, L) −→ Best response: B

If the input is (X, K), then the output is B. But when the input is (Y, L), then the output is A. Yes,
because the best response changes depending on what others do. This means the best response of
P3 is a function. It takes two strategies, one from P1 , one from P2 , and gives back another strategy
of P3 . So in fact, the best response is a function. Let us denote this function with B3 , indicating
that it tells the best response of the third player. We can now abbreviate everything, simply by
writing

B3 (X, K) = B
B3 (Y, L) = A
B3 (X, L) = B

Before proceeding further, we will introduce some notation to simplify our work. As you noticed
above, when focusing on a specific player, we need to carefully denote the strategies of the other
players. For example, in a game with n players, when focusing on the second player, we would write
(s1 , s3 , . . . , sn ), intentionally omitting s2 . However, this makes things really long. Therefore, we
want to make things shorter. We focus on the second player, and s2 is short enough. So, we should
trim the other notation. Instead of writing out all other players’ strategies as (s1 , s3 , . . . , sn ), we
define s−2 to represent the vector of strategies of all players except player 2. That is:

s−2 = (s1 , s3 , . . . , sn )

Now let us show how we use the new notation. Recall our earlier example:

u2 (s1 , a, s3 , . . . , sn ) = 5
u2 (s1 , b, s3 , . . . , sn ) = 8
u2 (s1 , c, s3 , . . . , sn ) = 4

Instead, we could write

u2 (a, s−2 ) = 5
u2 (b, s−2 ) = 8
u2 (c, s−2 ) = 4

where s−2 = (s1 , s3 , . . . , sn ). Observe that we have written the strategy of the second player as the
first argument, contrary to our usual notation. This is because we are now focusing on P2 and want

18
to highlight how their utility depends on their own strategy s2 , given the strategies s−2 of the other
players. Let us be even more specific:

u3 (A, s−3 ) = 1
u3 (B, s−3 ) = 4

where s−3 = (X, L), i.e., s1 = X, s2 = L.


Finally, the translation of
uk (sk , s−k )
into English is

The payoff of player k, when player chosen the strategy sk and rest of them chose the
strategies indicated by s−2 , whatever they are.

We are ready to define the best response function in a formal way.

Definition 3. Let G be a game with n players and let Pl be a player we focus on. The best
response function Bl of Pl is a function that takes the strategies of all other players, s−l , and
produces a strategy s = Bl (s−l ), which satisfies the following:

ul (s, s−l ) > ul (s′ , s−l ) (3.1)

for every other s′ ̸= s that is available to Pl .

Exercise 10. The idea of this exercise is to get accustomed to the definition and notational usage
of the best response function. Throughout this exercise, you should refer to the table of the previous
3 player game.

1. Find the best response of P2 , in case P1 plays X and P3 plays B.

Solution: We are asked to find B2 (s−2 ), where B2 denotes the best response function,
and s−2 denotes the strategies chosen by players other than P2 , so in this particular case
s−2 = (X, B). Now, in order to find the best response of the second player, we should
consider various pay-offs that P2 to can respond to the given combination of moves. By
referring to the table, we see that

u2 (K, s−2 ) = 3
u2 (L, s−2 ) = 2

Thus, we conclude that


B2 (s−2 ) = B2 (X, B) = K

2. Assume that P2 played K and P3 played B. In this scenario, which one of the following is
meaningful: s−1 , s−2 , or s−3 ? Explain your answer and then, determine using the table, find
the best response of the first player.

19
Solution:

3.2 Nash Equilibrium


The next step is to define the Nash Equilibrium using the best response idea. Before delving into
a formal presentation, we will first consider one of our previous examples, which defied the iterated
elimination of dominated strategies. So, let us consider game presented in the following table form.
We have seen that, using IEDS, we were able to predict the inevitable outcome of certain games, like

P1 /P2 A B C
X (2,4) (4,0) (5,3)
Y (4,2) (0,4) (6,3)
Z (3,5) (3,5) (9,7)

Prisoner’s Dilemma. Now, by armed with the best function idea, we will try to predict the outcome
of this stubborn game. Recall that, in our exploration of the Prisoner’s Dilemma, we have said that
(Betray, Betray) is inevitable. So, let us start our exploration, by first considering a combination
of strategies, and ask whether that combination could be inevitable.
We start with considering (Y, A), and ask the opinions of the players about this combination.
What would first player think, if they are forced to play Y and A respectively. He would be happy,
certainly, because, when the second player is forced to play A, the strategy Y is not bad at all!
u1 (Y, A) = 4 > u1 (Z, A) = 3 > u1 (X, A) = 2
In other words, Y is first player’s best response to A:
B1 (A) = Y
Now, let us wear the shoes of the second player and ask whether he would be happy or not, if they
are forced to play Y and A? He would certainly be unhappy, because, knowing that P1 is forced to
play Y , he could very well play B and get 4 points, instead of 0. Because
u2 (Y, B) = 4 > u2 (Y, C) = 3 > u2 (Y, A) = 2
In other words, he would unhappy, because his best response to Y is not A but B, in symbols
B2 (X) ̸= A but B2 (X) = B (3.2)
So, the second player would say that he is going to deviate from picking A to B, if he knows P1
will play Y , as a result, (Y, A) will not be played. We can summarize as follows: if the players are
told to play (Y, A):

20
1. Knowing that the second player will choose A, the first player would fine with playing Y ,
because his best response is to A is X.

2. Knowing that the first player will choose Y , the second player would like to deviate from A,
because his best response is to Y is not B.
We can use the reasoning just presented to make our decisions faster. For example, what would
happen if the players are told to play (X, A)? It won’t be played, because the X is not the best
response of the first player to the move A:

B1 (A) ̸= Y

What would happen if the players are told to play (X, B)? It won’t be played, because

B2 (X) ̸= B (3.3)

We can make our conclusion:


Let (s1 , s2 ) be a combination of strategies of P1 and P2 . Assume that an observer wants
to see the outcome (s1 , s2 ), and tries to manipulate the players by telling the first player
that P2 will play s2 , and telling the second player that P1 will play s1 . Then assuming
that players believe what the other player will play, the outcome (s1 , s2 ) will be realized,
only if s1 is the first player’s best response to s2 , and s2 is the second player’s best
response to s1 .
Exercise 11. What will happen if the first player is told that the second player will play Z and
the second player is told that the first player will play C?

Solution: The moves (C, Z) will be realized because

B1 (Z) = C

and
B2 (C) = Z

We are now ready to define the Nash Equilibrium concept. Informally, Nash Equilibrium seeks
combinations of strategies, where, no player would like to deviate from their corresponding strategy
if they believe that others will not deviate from their corresponding strategy. An
example is (C, Z). Let us give the formal definition now.
Definition 4. A strategy profile (s1 , s2 , . . . , sn ) is a Nash Equilibrium (NE) if no player can
improve their payoff by unilaterally deviating from their chosen strategy. In formal terms, for each
player Pi ,
ui (si , s−i ) ≥ ui (s′i , s−i ) for all s′i ∈ Si ,
where s−i represents the strategies of all players other than Pi in the equilibrium strategy profile.
This condition implies that each player’s strategy is a best response to the strategies chosen by
other players.
Exercise 12. Show that in the game below, (s1 , s2 , s3 ) = (Y, L, A) is a Nash Equilibrium.

21
Strategies Player 1’s Payoff Player 2’s Payoff Player 3’s Payoff
X, K, A 4 4 4
X, K, B 1 3 5
X, L, A 2 5 1
X, L, B 2 2 4
Y, K, A 5 1 3
Y, K, B 2 4 2
Y, L, A 3 2 5
Y, L, B 1 1 1

Table 3.2: Payoff Table for Players 1, 2, and 3

Solution: We are asked to show that (s1 , s2 , s3 ) = (Y, L, A) is a Nash Equilibrium. In this
case s−1 = (L, A) and one needs to check, if B1 (s−1 ) is equal to Y . Let us check it:

B1 (s−1 ) = B1 (L, A) = Y (3.4)

This is done, by checking if Y is the best response to (L, A). As

u1 (X, L, A) = 2
u1 (Y, L, A) = 3

we see that B1 (L, A) = Y . Now it is your turn to check, if s2 = L is the best response to
s−2 = (Y, A), and if s3 = A is the best response to s−3 = (Y, L).

22
Exercise 13. Show that (Betray, Betray) is a Nash Equilibrium in Prisoner’s Dilemma.

Solution:

3.3 Cournot Duopoly


In the Cournot duopoly model, two firms compete by choosing quantities to produce. The firms
produce homogeneous goods, and the market price depends on the total quantity supplied. Let us
denote:
• q1 = quantity produced by firm 1

• q2 = quantity produced by firm 2

• Q = q1 + q2 = total quantity produced

• P (Q) = market price as a function of total quantity

• C1 (q1 ) = cost function of firm 1

• C2 (q2 ) = cost function of firm 2


Assume the inverse demand function is linear:

P (Q) = a − bQ (3.5)

where a > 0 and b > 0. Moreover, assume that both firms have identical cost functions with
constant marginal cost c:
Ci (qi ) = cqi , i = 1, 2 (3.6)
In this model, the aim of both firms are to maximize their profits. We assume that, all the supply
will be consumed by the consumers.
In the remaining of this section, we will investigate the methods of finding Nash Equilibriums
of the Cournot model. We start with a particle case.
Example 1. Let us assume that a = 200, b = 1, and c = 10. We further assume that q2 = 50, and
try to find the best response of the first firm to q2 . i.e. we will answer the question

B1 (50) =?

Since the first firm aims to maximize its profit by deciding their supply q1 , given that q2 = 50, we
should write down the profit of the first firm with respect to their decision q1 . It is clear that

23
• Profit of the first firm π1 = Income - Cost

• Income = Number of sold items by the first firm * Price per item

• Cost = Number of sold items by the first firm * Cost per item

• Number of sold items by the first firm = The supply of first firm, because all will be sold by
assumption

• Number of sold items by the first firm = q1

Relying on our observations, we can calculate π1 , using the equations

• The price per unit is P (Q) = P (q1 + q2 ) = 200 − 50 − q1 = 150 − q1 ,

• the number of sold items by the first firm is q1 ,

• the cost per unit is 10.

Thus

π1 (q1 ) = P (Q)q1 − cq1


= (150 − q1 )q1 − 10q1
= 150q1 − q12 − 10q1
= 140q1 − q12 (3.7)

As 140q1 −q12 is the profit that the first firm is going to make, if the second firm supplies 50 products
and the first firm supplies q1 products. Thus, the highest amount of profit that the first firm can
make is the maximum value of π1 (q1 ) = 140q1 − q12 . As we have seen in the class, the maximum
value can be found by finding the critical points of this curve, which requires us to find the roots of
the derivative of it. So, we need to solve
∂π1
0 = (q1 )
∂q1
= 140 − 2q1 (3.8)

Obviously, q1 = 70 is the only zero of this function. As this function is a shape down quadratic, it
has no minimum and a unique maximum, 70 gives us a maximum.1 In conclusion,

B1 (50) = 70.

Exercise 14. Check, if the following equality holds:

B2 (70) = 50. (3.9)


1
We have to make this observation, because the zero of the derivative of a function only tells us that that point
is worth attention. It does not mean that that point gives maximum by default.

24
Solution:

1. Given the supply q1 = 70 of the first firm, calculate the P = (Q) = P (q1 + q2 ) in terms
of the supply q2 of.

2. Using the previous part, calculate the profit function π2 (q2 ).

3. Calculate the derivative π2′ (q2 ) of π2 (q2 ), and find its zero:

π2′ (X) = 0. (3.10)

4. Check, if the zero that you find equal to 50, and decide if B2 (70) = 50.

5. Does (70, 50) a Nash Equilibrium of the Cournot duopoly defined above? Explain your
answer.

Theorem 3.3.1. The Cournot duopoly with P (Q) = a − q1 − q2 , where Q = q1 + q2 , with constant
cost per unit c has a unique Nash equilibrium, q1∗ , q2∗ , where
a−c
q1∗ = q2∗ = (3.11)
3
Proof. Let (q1 , q2 ) be a combination of two strategies. The strategy profile (q1 , q2 ) is a N.E. if and
only
B1 (q2 ) = q1 ,
and at the same time
B2 (q1 ) = q2 .
Let us calculate the condition that q1 should require. By definition, q1 should maximize the profit
function π1 of the first firm. Given the second firms supply q1 , the profit function π(q1 ) of the first

25
firm is the following:
Total income
z }| { Total cost
z }| {
π(q1 ) = P (q1 + q2 ) ·q1 − c · q1
| {z }
Price per unit
= (a − q1 − q2 ) · q1 − c · q1
= a · q1 − q1 · q1 − q2 · q1 − c · q1
= (a − q2 − c) · q1 − q12

The maximal of this function should also be a zero of the derivative of this function, which is:

π ′ (q1 ) = a − q2 − c − 2 · q1 (3.12)

Thus, the maximal q1∗ of the profit function should satisfy the following equation

0 = π ′ (q1 )
= a − q2 − c − 2 · q1 (3.13)

So we get
a − q2 − c
q1 = B1 (q2 ) if and only if q1 = (3.14)
2
By the same reasoning, we see that
a − q1 − c
q2 = B2 (q1 ) if and only if q2 = (3.15)
2
So, (q1 , q2 ) is a N.E. if anf only if
a − q2 − c a − q1 − c
q1 = if and only if q2 = (3.16)
2 2
a−q2 −c a−q1 −c
Let q1 = 2 into q2 = 2 we get

a − q1 − c
q2 =
2
a − ( a−q22 −c ) − c
q2 =
2
(3.17)

By multiplying both sides with 2, the last equation can be written as


a − q2 − c
2q2 = a − −c
2
2a a − q2 − c 2c
2q2 = − −
2 2 2
a + q2 − c
2q2 =
2
(3.18)

26
Again by multiplying both sides with 2, we get

4q2 = a + q2 − c (3.19)

Which in turn gives


a−c
q2 = (3.20)
3
By substituting the q2 instead of q1 , we can derive the same equality for the first equation.
a−q2 −c
Exercise 15. Do it. Substitute q2 in q1 = 2 and solve it for q1 .

Solution:

Thanks to the exercise, if (q1 , q2 ) is a N.E., then


a−c
q1 = q2 = (3.21)
3
On the other hand, we can show that a−c 3 , 3 is a N.E. by showing that both strategies are the
a−c

best responses to each other. Which finishes the proof of the theorem.

Exercise 16. Do everything that is done in the Cournot Duopoly section, for b = 2. So, first start
with the following example: Let us assume that a = 200, b = 2 (this was 1 in the example above),
and c = 10. As before, further assume that q2 = 50, and try to find the best response of the first
firm to q2 . i.e. try to answer the question

B1 (50) =?

And continue in this fashion. That is, if for example you found that the best response is 110, check
if
B2 (100) = 50
Finally, show that in this case (q1 , q2 ) is a N.E. if and only if
a−c a−c
q1 = q2 = = (3.22)
b·3 2·3
For this last part, just rewrite the proof that is presented above, and put b when it is necessary.
Believe me, this exercise will be rewarding.

27

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy