Academic Year: 2024-25 Semester: VII

Class/Branch: BE/DS Subject: CSDOL7023 Game Theory for Data

Science LAB

ASSIGNMENT NO. 2
Lab Outcome (LO):

LO1 To apply the principles of the Prisoner's Dilemma to real-world scenarios.

To apply Game Theory Principles of dominant strategies to Real-World
LO2
Problems
To solve extensive form games and repeated games by applying backward
LO3
induction
LO4 To apply game theory concept in mixed and pure strategies
To solve complex game theory problems by game theory algorithms and
LO5
computational tools
LO6 To apply game theory concepts in various data science contexts

1. Two competing coffee shops, A and B, are located in the same neighborhood. Each
shop can choose to advertise (A) or not advertise (N) each day. The shops make their
decision simultaneously. If both shops advertise, they each earn a profit of $50. If
neither advertises, each earns $100. However, if one shop advertises and the other
does not, the one that advertises earns $200, while the other earns $0.

A. Build this situation as a strategic game in matrix form.

B. Identify the mixed-strategy Nash Equilibrium for this game.
C. Model the mixed strategy to a pure strategy and the difference in terms of
decision-making and outcomes. [LO4/L3]

Solution:

A. Game Matrix:

B: Advertise (A) B: Not Advertise (N)

A: Advertise (A) (50, 50) (200, 0)
A: Not Advertise (N) (0, 200) (100, 100)

B. Mixed-Strategy Nash Equilibrium:

Let p be the probability that A advertises, and q be the probability that B advertises.

• Expected payoffs for A:

o If A advertises: 50q+200(1−q)
 o If A does not advertise: 0q+100(1−q)

Set A's payoffs equal:

50q+200(1−q)=100(1−q)
Solving this:

50q+200−200q=100−100q
150q=100⇒q=2/3

• Expected payoffs for B:

o If B advertises: 50p+0(1−p)
o If B does not advertise: 200p+100(1−p)

Set B's payoffs equal:

50p=200p+100(1−p)
Solving this:
50p=200p+100−100p
150p=100⇒p=2/3

Therefore, the mixed-strategy Nash Equilibrium is p=q=2/3, meaning that each

coffee shop advertises 2/3 of the time and does not advertise 1/3 of the time.

C. Difference Between Mixed and Pure Strategies:

• Pure Strategy: A pure strategy involves making a single, definitive choice (e.g., always
advertise or never advertise). If A and B both used pure strategies, their outcomes would
be deterministic.
• Mixed Strategy: A mixed strategy involves randomizing over different choices (e.g.,
advertise with probability 2/3). Here, the decision is not fixed, allowing for a
probabilistic approach, which can lead to more balanced payoffs over time. The mixed
strategy allows A and B to be unpredictable, avoiding consistent exploitation.

2. Consider two rival tech companies, X and Y, that are entering a new market for
cloud services. Each company has two strategic options:
• Aggressive pricing (A): The company slashes prices aggressively to capture market
share.
• Moderate pricing (M): The company sets moderate prices, focusing on long-term
sustainability.
• The payoff for each company depends on the combination of their strategies:
• If both companies choose aggressive pricing, they engage in a price war, and each
earns $200.
• If both companies choose moderate pricing, they peacefully coexist and each earns
$500.
• If one company chooses aggressive pricing and the other chooses moderate pricing,
 the company with aggressive pricing earns $800 while the other earns $100 due to
a loss in market share.

A. Model this scenario as a strategic game in normal form by constructing the payoff
matrix for companies X and Y.
B. Using game theory algorithms and computational tools (such as the Best Response
Algorithm, Minimax, or Linear Programming), Identify the Nash Equilibrium for
this game. [LO5/L3]
Solution:
A. Payoff Matrix:
The payoff matrix for companies X and Y can be represented as follows:
Y: Aggressive (A) Y: Moderate (M)
X: Aggressive (A) (200, 200) (800, 100)
X: Moderate (M) (100, 800) (500, 500)

B. To identify the Nash Equilibrium, we need to determine the best response strategies for
each company given the strategies of the other company. A Nash Equilibrium occurs
when neither player has an incentive to unilaterally change their strategy.

1. For Company X:
o If Y chooses A:
▪ X’s payoff for A: 200
▪ X’s payoff for M: 100
▪ Best response: A (because 200 > 100)
o If Y chooses M:
▪ X’s payoff for A: 800
▪ X’s payoff for M: 500
▪ Best response: A (because 800 > 500)
2. For Company Y:
o If X chooses A:
▪ Y’s payoff for A: 200
▪ Y’s payoff for M: 100
▪ Best response: A (because 200 > 100)
o If X chooses M:
▪ Y’s payoff for A: 100
▪ Y’s payoff for M: 500
▪ Best response: M (because 500 > 100)

Now, we can summarize the best responses:

• If both companies choose A, neither company has an incentive to change, as both are
earning 200.
• If Company X chooses A and Company Y chooses M, Company X will earn 800
while Company Y earns 100, which is not stable as Y would switch to A.
• If both choose M, neither company has an incentive to change, earning 500 each.

Nash Equilibria:

From the analysis above, we see there are two Nash Equilibria:

1. (A, A): Payoff (200, 200)

2. (M, M): Payoff (500, 500)
 C.

3. In the context of data science and machine learning model competition, consider two
data science teams, Team A and Team B, competing to build the best predictive
model for a company's business problem. Each team has two strategic options:

• Invest more resources (I): Team spends more time and computational resources to
improve the model.
• Invest fewer resources (F): Team spends less time and resources, aiming to deliver
quickly but with a potentially lower-performing model.

The success of each team's model depends on the resource investment decisions of both
teams:

• If both teams invest more resources, they each get a medium score, reflecting
competition but with both models performing decently.
• If one team invests more resources and the other invests fewer, the team investing
more resources gets a high score, while the other gets a low score due to their lower-
performing model.
• If both teams invest fewer resources, they both get a low score, as neither model
performs well.

The payoffs are as follows:

• High score: $800

• Medium score: $500
• Low score: $100

A. Use game theory concepts to analyze what would happen if the game were repeated
multiple times (iterated game) and identify how the teams might adjust their
strategies over time.
B. Identify how game theory concepts apply to this competitive scenario in data
science. [LO6/L3]

Solution:

A. Iterated Game and Strategy Adjustments:

If this game is repeated multiple times (an iterated game), both teams may adjust their
strategies over time. This is where concepts like Tit-for-Tat come into play:

• Tit-for-Tat: In an iterated version, each team might start by investing more resources. If
one team cheats (invests fewer resources), the other team may respond by also investing
fewer resources in the next round, leading to worse overall outcomes. Over time, the
teams may realize that consistently investing more benefits both, leading to cooperation.

In iterated games, cooperation can emerge because both teams recognize the long-term
benefits of mutual investment over the short-term advantage of defecting (investing fewer
resources).

B. Application of Game Theory Concepts:

 This game illustrates several important concepts in game theory and their application to
competitive data science contexts:

• Nash Equilibrium: The teams settle into a balance where both choose to invest more
resources, knowing that deviating from this strategy leads to worse outcomes.
• Iterated Games: In practice, data science competitions often involve multiple rounds,
where teams can adjust their strategies based on their opponent’s behaviour. Over time,
cooperation (both teams investing more) becomes more appealing to avoid consistently
poor outcomes from both investing fewer resources.
• Competitive Strategy: In real-world data science competitions, teams must balance
resource allocation (computational power, time, and expertise) with the potential
payoffs of delivering a better model. Game theory helps explain how competitors can
reach equilibrium strategies that optimize their performance in competitive
environments.

4. Three companies, A, B, and C, are competing in a location-based advertising market

where they must choose locations to place their ads. Each company can choose one
of two locations, L1 or L2, to place its ad. The number of customers they reach
depends on whether the companies choose the same location or different locations:

• If all three companies choose the same location (either L1 or L2), they equally split
the customer base, earning 200 customers each.
• If two companies choose one location and the third company chooses the other
location, the company that is alone at a location receives 600 customers, while the
two at the same location share 300 customers equally, earning 150 customers each.
• If all three companies choose different locations (e.g., two at L1, one at L2), the
company with no competition gets 600 customers, and the two others get 0.

A. Identify if any pure strategy Nash equilibria exist.

B. Choose the concept of mixed strategies to identify the probabilities with which each
company should randomize between locations to achieve a mixed-strategy Nash
equilibrium. [LO4/L3]

Solution:

A. Pure Strategy Nash Equilibria:

We check the best responses for each company:

• Company A:

If both B and C choose L1, A's best response is L1 (200 > 150).
If both B and C choose L2, A's best response is L2 (200 > 150).
If B chooses L1 and C chooses L2, A's best response is L1 (150 > 600).
If B chooses L2 and C chooses L1, A's best response is L2 (150 > 600).

By symmetry, B and C will have similar best responses.

There is no pure strategy Nash equilibrium in this scenario because no strategy

consistently provides a best response for all firms. In some cases, each firm would prefer
to switch to another location to improve their payoff.
B. Mixed Strategy Nash Equilibrium:

Let:

• p be the probability that Company A chooses L1.

• q be the probability that Company B chooses L1.
• r be the probability that Company C chooses L1.

The goal is to make each company's choice of L1 and L2 indifferent, meaning their
expected payoffs from choosing L1 or L2 should be the same.

For Company A, the expected payoff from choosing L1 is:

EA(L1)=200(qr)+150(q(1−r)+(1−q)r)+600(1−q)(1−r)

The expected payoff from choosing L2 is:

EA(L2)=200((1−q)(1−r))+150(q(1−r)+(1−q)r)+600(qr)

5. A new startup incubator is accepting applications from two startups, X and Y, to

award a limited investment fund. Each startup can choose one of two strategies:
they can either present an innovative proposal (I) or a conservative proposal (C).

• If both startups present innovative proposals, the incubator divides the investment
equally, but the startups incur higher development costs, each earning a net profit
of $300.
• If one startup presents an innovative proposal and the other presents a
conservative proposal, the innovative startup receives the full investment fund,
earning $500, while the conservative one gets nothing and loses $100.
• If both present conservative proposals, they receive equal investment, but at a
lower profit, each earning $200.

A. Identify whether there is a pure strategy Nash equilibrium.

B. Apply the mixed strategy equilibrium to identify the probabilities with which each
startup should randomize between presenting an innovative or conservative
proposal.
C. Identify how mixed strategies provide flexibility in decision-making compared to
pure strategies. [LO5/L3]

Solution:

A. Pure Strategy Nash Equilibrium:

• If X believes Y will present an innovative proposal, X’s best response is to present a

conservative proposal (to avoid a high-risk development cost).
• If Y believes X will present a conservative proposal, Y’s best response is to present an
innovative proposal.

There is no pure strategy Nash equilibrium, as the startups have conflicting incentives
depending on the other’s strategy.

B. Mixed Strategy Nash Equilibrium:

 Let p be the probability that X chooses to present an innovative proposal and q be the
probability that Y chooses an innovative proposal.

We calculate the expected payoffs for each startup and solve for the probabilities p and
q that make both startups indifferent to presenting an innovative or conservative
proposal.

C. Mixed Strategy Flexibility:

Mixed strategies allow each startup to randomize its decision, introducing uncertainty
into the opponent's strategy choice. This randomness prevents opponents from easily
predicting the outcome and increases flexibility in competitive environments where one
must balance risk and reward.