ü Decision Theory focuses on how individuals make choices under

Decision and Game Theory uncertainty and with available information.

ü Previous OR courses have focused mainly on DM when the consequences

Introduction of alternative decisions are known with a reasonable degree of certainty;
which enabled formulating helpful mathematical models (such as LP, IP,

Sonia REBAI - M. Naceur Azaiez NLP).

ü Decision-making is important in everyday life, business, economics, public

policy, and even personal situations. Decisions often must be made in
Tunis Business School
University of Tunis environments that are much more Loaded with uncertainty.

For example: The Dinner Dilemma:

o Personal Decisions: Career choices, financial investments, health ü Imagine you are deciding what to have for dinner—should you try
decisions. something new or stick to a familiar choice? This involves trade-offs (cost,
o Business Decisions: Marketing strategies, pricing, project time, satisfaction).
management. ü Factors influencing the decision:
o Public Policy: Economic policy, disaster management, resource o Preferences (what you like)
allocation.
o Probabilities (e.g., uncertainty about whether you'll like the new dish)
ü Decision theory is relevant to various fields such as economics, o Payoffs (potential satisfaction from each option)
psychology, finance, medical decision-making.
ü The Decision-Making Process outlines the following steps: ü Normative vs. Descriptive Decision Theory
1. Define the problem: Identify the decision to be made.
o Normative Decision Theory: Focuses on how rational agents should make
2. Identify alternatives: What are the different options available?
decisions (using logic and probabilities).
3. Evaluate outcomes: Assess the potential consequences and
uncertainties. o Descriptive Decision Theory: Examines how people actually make

4. Maximize expected outcome: Choose the option with the best expected decisions, often influenced by biases, emotions, and cognitive limitations.
outcome. ü Growing field of behavioral decision theory that studies these human
ü Decision analysis is a process that provides a structured method with
deviations from rationality.
analytical tools designed to improve one’s decision-making skills. It provides
a conceptual framework allowing to break the problem down into smaller ü Game Theory studies strategic interactions where the outcome depends not
and easier sub-problems. just on a single decision maker but on the choices of others.

Why decisions are hard? Complexity of the decision problem

The problem could be naturally complex

Four broad sources of difficulty:
Ø Too many uncertain scenarios.
²complexity,
Ø Too many factors.
²uncertainty,
Ø Too many alternatives some of which may be sequential.

²multiple objectives, and Ø Too many consequences, some of which may have impacts that are
hard to evaluate such as intangible benefits.
²competing viewpoints.
7 8
Uncertainty Multiple objectives
Uncertainty could be the major handicap in making decision. Future Ø Progress towards one objective may impede progress in others
events are beyond the DM’s control and hence could result in good or bad ü Quality vs. cost
consequences. ü Short-term profit vs. long-term profit
ü Economic benefits against environmental damage
ü Shareholders’ benefits against employees’

ü Expected profit vs. risk

Ø A DM may be interested in trading off multiple objectives.
9 10

Competing viewpoints Are Humans Rational?

Ø A decision problem may be difficult if different perspectives lead to
Ø Rationality says that if you have better alternative and worse alternative,
different conclusions.
you choose the better one.
Ø Also, from a single perspective, changes in the decision criteria may
Ø Humans are often very certain of knowledge that is false (cognitive
change the optimal decision policy.
illusions).
Ø Various decision-makers may view the same decision problem from
different perspectives Ø Let’s take for example the Allais Paradox.

Ø They may rank priorities or preferences differently

11 12
Allais Paradox First set of gambles

Ø In 1953 Maurice ALLais, winer of Nobel prize in Economics (1988), has • Gamble A: Win $1 million (1.0)

published a paper that contradicts the expected utility hypothesis based • Gamble B: Win $5 million (0.10) + Win $1 million (0.89) + Win $0 (0.01)

on a number of experiments that he conducted. Second set of gambles

• Gamble C: Win $1 million (0.11) + Win $0 (0.89)

Ø This is inconsistent with rational behaviour.
• Gamble D: Win $5 million (0.10) + Win $0 (0.90)
Ø Consider the following two sets gambles:
ü Which one do you prefer?
13

ü In numerous experiments, most people prefer A to B and D to C. • The expected values of C and D are
Are these choices consistent with one another ? EV(C) = 0.11*$1m + 0.89*$0m = $110,000
• The expected values of A and B are EV(D) = 0.1*$5m + 0.90*$0m = $500,000
EV(A) = 1.0*$1m = $1m • So if someone prefer D over C, this supposes a maximization of the
EV(B) = 0.1*$5m + 0.89*$1m + 0.01*$0m = $1.39m expected value.

• So if someone prefer A over B, this supposes a maximization of the • If we chose A over B and D over C, we are violating expected utility
expected utility and not of the expected value. Which means: theory. Let’s get a better understanding of the paradox by rearranging

U($1m) > 0.1*U($5m) + 0.89*U($1m) + 0.01*U($0m) what we have written out these gambles.
15 16
 The framing of the problem confuses participants

• Gamble A: Win $1 million (1.0) • Gamble A: Win $1 million (0.11) + Win $1 Decision and Game Theory
million (0.89)
• Gamble B: Win $5 million (0.10) + • Gamble B: Win $5 million (0.10) + Win $0
Win $1 million (0.89) + Win $0 (0.01) (0.01) + Win $1 million (0.89)
Chapter 1: Problem Formulation

• Gamble C: Win $1 million (0.11) + • Gamble C: Win $1 million (0.11) + Win $0 Sonia REBAI - M. Naceur Azaiez
Win $0 (0.89) (0.89)

• Gamble D: Win $5 million (0.10) + • Gamble D: Win $5 million (0.10) + Win $0

Win $0 (0.90) (0.01) + Win $0 (0.89) Tunis Business School

University of Tunis

The first step in the decision analysis process is the problem formulation. We
Setting objectives
begin with a verbal statement of the problem. We then identify the following
An objective is a specific goal that a decision-maker attempts to reach.
elements:
Setting the right objectives is vital for a decision problem.
• the objective to optimize,
• It should be measurable to evaluate the alternatives leading to different
• the decision alternatives, levels of the objective.
• the uncertain future events, referred to as chance events or state of • Knowing the right objectives helps identifying the adequate alternatives
nature to consider.

• the consequences associated with each decision alternative and each • It also dictates the way to measure the results and the type of

chance event outcome, referred to the outcome (gain or loss) function uncertainty inherited in the problem. 3
• Some objectives may be related; others may serve for more important
• It is sometimes difficult to understand the objective itself. For
ones. The DM should distinguish between what is important for an
instance, business firms seek “success” or “excellence”.
objective and what is the ultimate objective.
ü How to measure it accurately?
• It is possible to have a hierarchy of objectives. For example; by having a
ü Would profit maximization, cost reduction, or operating according to
university degree, one may get a good job that can generate a high
international standards be good indicators to track progress towards
salary which can improve the quality of life of the DM. Hence,
such an objective?
improving the quality of life is the ultimate objective in this hierarchy.
4 5

Identifying alternatives
• Among possible actions to consider:
Once the decision situation is clearly identified and the objectives are
ü Wait & see: in order to gain more information before acting.
adequately set, one may attempt to discover and create alternatives.

• Some of the alternatives may arise in a straightforward way: e.g., ü Do nothing: in order to not to get worse than the current situation.
accept/reject, select among a list. Others may be obtained only after
ü Pay for safety or insurance: in order to get rid of risk.
some careful consideration and analysis of the problem.

• The DM should make sure that the identified actions would serve the
final objective(s) either directly or indirectly.
6 7
• In complex situations, some alternatives are less obvious to identify. It is Environment of the decision problem
the case where alternatives are of technical nature: legal procedures,
The decision problem environment can be identified through the level of
financial solutions, reforms, engineering methods, medical procedures.
knowledge of the states of nature. Three levels can be distinguished.
These would be obtained by consulting the right experts.

• Other alternatives may be less technical yet require deep thinking. These
More information
may be identified by calling for meetings, to take more time to think, to
employ competencies & know-how. Among scientific tools used are Uncertainty Risk Certainty

creativity, checklists, brainstorming. Less information

8 9

ü Uncertainty In general, it is not necessary that the problem falls in one of the three
It reflects the extreme case where situations are out of the control of the decision- cases above. However, the modeling approach would dictate the
maker without knowledge on the likelihood of their occurrence environment as a choice such as investigating the uncertainty or totally
ü Risk neglecting it.
The available information is used to assess a probability distribution of the
occurrences of the situations that are out of the control of the DM Uncertain Environment

ü Certainty Risky Environment

Deterministic
Another extreme case where all outcomes of the decision problem are Environment
known beforehand.

10 11
Example 1
Alternatives
A company manufacturing electronic components realizes that the computer market
• A1 : Buying machine 1
grow rapidly. It has reached the maximum of its production for the moment and can no
longer meet any additional demand.
• A2: Buying machine 2

The managers, whose policy is to satisfy the demand, are considering the opportunity of • A3: Subtracting
purchasing a machine from two preselected ones or or move towards subcontracting. States of Nature
The possible results in million TD are estimated as follows: • E1: High Additional Demand
Additional Additional Additional Weak No Additional • E2: Medium Additional Demand
High Demand Medium Demand Demand Demand
Machine 1 0.7 0.3 -0.1 -0.3 • E3: Weak Additional Demand
Machine 2 0.5 0.4 0.1 -0.2 • E4: Null Additional Demand
Subcontracting 0.3 0.2 0.3 0 12 13

Payoff Table and Decision Tree

• The consequence resulting from a specific combination of a decision

alternative and a state of nature is a payoff. E1 E2 E3 E4
A1 0,7 0.3 -0,1 -0,3
• A table showing payoffs for all combinations of decision alternatives and
A2 0,5 0.4 0,1 -0,2
states of nature is a payoff table.
A3 0,3 0,2 0,3 0
• Payoffs can be expressed in terms of profit, cost, time, distance or any
other appropriate measure.
14 15
 Ø To structure the decision problem, a graph called a decision tree can
be used.
E1
0.7
Ø The decision tree is a graphical representation that structures in a
A1 E2 0.3
chronological order the decision alternatives and the states of nature. E3
-0.1
A2 E4
Ø It consists on representing nodes connected to each other by arcs. E1
-0.3
0.5
Nodes are of two types: E2 0.4

A3
E3 0.1
• Decision nodes represented by squares and E4 -0.2
E1 0.3
• Probability nodes represented by circles.
E2 0.2
E3
Ø the different results are indicated at the extremities of the tree. 0.3
E4 0
16 17

Solving the Decision Problem Sensitivity Analysis

• Solving the decision problem by identifying the optimal set of actions to • Once the problem is solved, sensitivity analysis is performed to answer
take is usually obtained through backward iterations “What if” questions
• Calculations are usually simple and mostly performed through software • One should find out if a minor change in the problem inputs would make
packages such as: significant change in the solution
ü Expert Choice • In such a case, one should carefully re-evaluate sensitive parameters (for
ü Logical Decisions
which a small change may result in a change in the solution)
ü Decision Tools
• Sensitivity analysis may help the DM feel the need to redefine objectives
ü Tree Plan
or to reconsider alternatives leading to a more structured model
ü Supertree
18 19
• After several iterations of revising the decision model, the DM may reach
Decision and Game Theory
a more adequate structure or more pertinent solution

• With the tedious iterative process of conducting sensitivity analysis, the

perception of the problem may change, some better understanding of
Chapter 2: Decision Under Uncertainty
uncertainty may be obtained and the views on the various alternatives
may be modified. Sonia REBAI - M. Naceur Azaiez
• The DM will reach higher maturity in considering the problem with the
deep thinking involved Tunis Business School
University of Tunis
20

Optimistic approach
• In this chapter we consider approaches that do not require knowledge of
the probabilities of the states of nature. • The DM assumes that each alternative will generate the best possible
result, then he chooses the one that gives him the best result among
• These approaches are appropriate in situations in which a simple best-
the most favorable results.
case and worst-case analysis is desirable.
• If the objective of the problem is a maximization, the DM uses the
• Because different approaches sometimes lead to different decision
Max-max criterion.
recommendations, the DM needs to select the specific approach that,
according to him is the most appropriate.
• If the objective is a minimization, the DM uses the criterion Min-min.
 Conservative (Pessimistic) approach
• A DM assumes that each decision alternative generates the worst
e1 e2 e3 e4 Max
payoff, then he chooses the action that gives him the best result among
a1 250 350 350 400 400
the least favorable ones (Wald's criterion).
a2 225 300 380 420 420
• If the objective of the problem is a maximization of the payoff, the DM
a3 200 200 400 500 500 Max
uses the criterion Max-min.

• If the goal is a minimization of the payoff, the DM uses the Min-max

criterion.

Hurwicz criterion
• Most decision makers are not totally pessimistic or optimistic.
e1 e2 e3 e4 Min
• To measure the degree of optimism of such a DM, we use a scale aÎ [0,1],
a1 250 350 350 400 250 Max where
a2 225 300 380 420 225 ü 0 indicates a totally pessimistic DM.

a3 200 200 400 500 200 ü 1 indicates a totally optimistic DM.

• The approach uses a weighed sum of best payoff times a with worst
payoff times (1- a).
• Then, the highest weighed sum is selected.
 9

Savage (Min-max Regret)

• This criterion is based on the regret that the DM can manifest after
e1 e2 e3 e4 Max Min 0.4*Max+0.6*Min choosing an alternative decision.
a1 250 350 350 400 400 250 310
• It consists in comparing the corresponding payoff with the best possile
a2 225 300 380 420 420 225 303 one for the same state of nature.
a3 200 200 400 500 500 200 320 Max • We first compute a regret table by calculating the difference between
each payoff and the best payoff under the same state of nature.
• For each decision, identify the maximum regret.
• Select the decision with the minimum of the maximum regret.

10

e1 e2 e3 e4 Remark
a1 250 350 350 400
a2 225 300 380 420 • In all the above criteria the maximum and/or the minimum payoff is
a3 200 200 400 500 used as a basis for decision

• Except for the regret criterion, all other entries are ignored in
Regret table
e1 e2 e3 e4 Max choosing the “best” alternative
a1 0 0 50 100 100 • Note that these extreme payoffs that were used for decision might be
a2 25 50 20 80 80 Min extremely unlikely

a3 50 150 0 0 150
 Laplace criterion
• A rational DM tries to incorporate all the information into the evaluation e1 e2 e3 e4 Mean
process. He uses the average payoff over all entries. a1 250 350 350 400 337.5 Max

• Because no probabilities are identified, all payoffs are considered as a2 225 300 380 420 331.25

equally likely a3 200 200 400 500 325

• The criterion selects the best average (largest when maximizing a payoff
and the smallest when minimizing it)

In many decision-making situations, we can obtain probability assessments

for the states of nature. When such probabilities are available, to identify
Decision & Game Theory the best decision alternative, four possible criteria can be used

• Maximum Likelihood Criterion

Chapter 3: Decision Under Risk
• Expected Monetary Value (EMV)
Sonia REBAI - Naceur Azaiez
• Expected Opportunity Loss

Tunis Business School • Expected-Variance Criterion

University of Tunis
 3 4

Maximum Likelihood Criterion Expected Monetary Value (EMV)

The DM Identifies the most likely event, ignores others, and picks act with The DM calculates for each alternative the weighted average payoff,
the best payoff. then he selects the act providing the best expected value.

(0.2) (0.4) (0.1) (0.3) (0.2) (0.4) (0.1) (0.3)

EMV
S1 S2 S3 S4 S1 S2 S3 S4

a1 250 350 350 400 350 Max a1 250 350 350 400 345 Max

a2 225 300 380 420 300 a2 225 300 380 420 329

a3 200 200 400 500 200 a3 200 200 400 500 310
3 4

5 6

Expected Opportunity Loss (EOL) Expected Value-Variance Criterion

The DM computes the weighted average of the opportunity losses for • The DM optimizes both the EV and the variance.
each alternative. • In the case of a maximization problem, the DM Maximizes (E(Z)-ks(Z)),

(0.2) (0.4) (0.1) (0.3) where Z is a random variable representing payoff and k is a predefined
EOL
S1 S2 S3 S4 constant reflecting the decision-maker's risk aversion factor. It is used as
a1 0 0 50 100 35 Min a weight indicating the degree of importance of the variance of Z in
a2 25 50 20 80 51 relation to its expected value. For example, a decision maker who is
a3 50 150 0 0 70 particularly sensitive to results far below EV will choose a value for k
greater than 1.
Note that EMV and EOL yield the same optimal decision. 5 6
 7

Example 1 : Texaco vs. Pennzoil

Let’s suppose that k=1.
In early 1984, Pennzoil and Getty Oil agreed to the terms of a merger. But

(0.2) (0.4) (0.1) (0.3)

before any formal documents could be signed, Texaco offered Getty a
E(Z) E(Z²) Var = E(Z²)-E(Z)² E(Z)-s(Z)
S1 S2 S3 S4
substantially better price, and Gordon Getty, who controlled most of the
a1 250 350 350 400 345 121750 2725 292.798 Max
Getty stock, reneged on the Pennzoil deal and sold to Texaco. Naturally,
a2 225 300 380 420 329 113485 5244 256.585
Pennzoil felt as if it had been dealt with unfairly and immediately filed a
a3 200 200 400 500 310 115000 18900 172.523
lawsuit against Texaco alleging that Texaco had interfered illegally in the

Pennzoil-Getty negotiations.
7

Pennzoil won the case; in late 1985, it was awarded $11.1 billion, the largest In April 1987, just before Pennzoil began to file the liens, Texaco offered to

judgment ever in the United States at that time. A Texas appeals court pay Pennzoil $2 billion to settle the entire case. Hugh Liedtke, chairman of
Pennzoil, indicated that his advisors were telling him that a settlement
reduced the judgment by $2 billion, but interest and penalties drove the total
between $3 and $5 billion would be fair.
back up to $10.3 billion. James Kinnear, Texaco’s chief executive officer, had
said that Texaco would file for bankruptcy if Pennzoil obtained court What do you think Liedtke should do? Should he accept the offer of $2
billion, or should he refuse and make a firm counteroffer? If he refuses the
permission to secure the judgment by filing liens against Texaco’s assets.
sure $2 billion, he faces a risky situation. Texaco might agree to pay $5 billion,
Furthermore, Kinnear had promised to fight the case all the way to the U.S.
a reasonable amount in Liedtke’s mind. If he counteroffered $5 billion as a
Supreme Court if necessary, arguing in part that Pennzoil had not followed
settlement amount, perhaps Texaco would counter with $3 billion or simply
Security and Exchange Commission regulations in its negotiations with Getty.
pursue further appeals.
 Below is a decision tree that shows a simplified version of Liedtke’ s
Experts expect that the Supreme Court will keep the fine with only 20% problem.
chance, will reduce it to $5 billion with 50% chance or will eliminate it
completely with a 30% chance.
It was also believed that Texaco accepts a counter-offer of $5 billion with
1/6 chance and would place a counter-offer of $3 billion with 1/3 chance.
Find the optimal strategy.

Example 2
A government is bidding for the exploitation of an oil field. A company plans to
make an offer of 110 million TD. The company estimates that it has a 60%
chance of winning the contract. If it wins the contract it can choose between
three methods for exploiting the field: the new method (NM), the existing
method (EM), or subcontracting (SC). The outcomes are provided below.
The cost of preparing and submitting the bid is 2 million TD. If the company
The decision tree shows that his best current choice is to make the $5 billion does not bid, it can invest in another alternative that guarantees a return of 30
counteroffer with an expected payoff of $4.63 billion. The decision tree shows clearly million TD. Build a decision tree and determine the company's optimal
what Liedtke should do if Texaco counteroffers $3 billion: He should refuse.
strategy.
 142 Do not submit an offer
30

Probability Profit (million TD) 350 S (0.3) 600

Success 0.3 600

Sub
NM MS (0.6) 300
NM Moderate Success 0.6 300 350

mit )
F (0.1) -100
Failure 0.1 -100

(-2
an
) EM S (0.5) 300
Success 0.5 300 0.6 202

offe
( MS (0.3)
in ) 200
EM Moderate Success 0.3 200 W 10

r
( - 1 F (0.2)
144 -40
Failure 0.2 -40 SC
250
SC Moderate Success 1 250
Lose (0.4)
0

Risk Profiles
ü EMV alone does not tell us the whole story; it does not inform us about
how much variation there is in the consequences.

ü To help us choose the best alternative, we should consider both the EMV
and the set of possible consequences for each alternative.

Let’s reconsider Penzoil vs. Texaco example.

 Alternative/Strategy EMV Consequence Values Consequence Prob.
Accept $2 Billion $2 Billion $2 Billion 100%
Counteroffer $0 24.9%
$5 Billion; Refuse $4.63 $5 Billion 58.5%
Texaco Counteroffer Billion $10.3 Billion 16.6%

Counteroffer $0 15%
$5 Billion; Accept $3 $4.12 $3 Billion 33%
Billion” strategy Billion $5 Billion 42%
$10.3 Billion 10%

Risk profile for the “Accept $2 Billion” alternative Risk profile for the “Counteroffer $5 Billion; Refuse Texaco Counteroffer” strategy
 ü Risk profiles can be calculated for strategies that might not have
Risk profile for the “Counteroffer $5 Billion; Accept $3 Billion” strategy
appeared as optimal in an expected-value analysis.

ü For example, Comparing the two previous figures indicates that the
strategy “Counteroffer $5 Billion; Accept $3 Billion,” which we ruled out
on the basis of EMV, yields a smaller chance of getting nothing, but also
less chance of a $10.3 billion judgment. Compensating for this is the
greater chance of getting something in the middle: $3 or $5 billion.

• In a risky environment, the probability distribution of the states of

Decision & Game Theory nature plays an important role in the choice of the optimal decision.

• DMs have preliminary or prior probability assessments that are the

Chapter 4: Decision Analysis under Risk with best values available at that time.
Additional Information • It would then be of importance to seek additional information before
making a final decision.

• This new information can be used to revise or update the prior

Sonia REBAI - M. Naceur Azaiez probabilities so that the final decision is based on more accurate
Tunis Business School
University of Tunis probabilities.
• The revised probabilities are called posterior probabilities. Example 1

• Most often, additional information is obtained through experiments, a A company has the choice between producing itself or buying from a supplier

consulting advice, or a forecast. one of the electronic components that it uses in its activity. Net profit depends

• Such information is usually acquired at some cost. The question is on the level of demand for the product requiring this component. This

whether the cost paid is worth. demand can be low, medium or high. The a priori distribution is estimated at
0.35; 0.35 and 0.30.
• To answer this question we need to calculate the Expected Value of the
S1 S2 S3
additional information. Low Demand Medium Demand High Demand
Produce -20 40 100
buy 10 45 70 4

S1 -20
S2
The company has the opportunity to test the demand on the market. Two d uce 40
Pro S3 100
S1 10
results are possible for this test favorable (F) or unfavorable (U). The

F
Buy S2 45
S3
following conditional probabilities were estimated: S1
70
-20
ce S2
du 40

ey
Pro S3

v
100
S1 S2 S3 U

Sur
S1 10
Buy S2
Favorable P(F|S1)=0.10 P(F|S2)=0.40 P(F|S3)=0.60 45
S3 70
Unfavorable P(U|S1)=0.90 P(U|S2)=0.60 P(U|S3)=0.40 S1
No -20
Su S2
r ve Produce 40
y S3 100
Bu S1
10
Find the optimal strategy to adopt. y
S2 45
5
S3 70
Bayes’ theorem

Si: State of Nature (i = 1, …, n)

P(Si): Prior Probability

S1 S2 S3 The
P(S1│“.”) P(S2│“.”) P(S3│“.”)
Ij: Professional Information (Experiment)( j = 1, …, n) 0.35 0.35 0.30 sum

P(Ij | Si): Conditional Probability F 0.1 0.4 0.6 0.035 0.14 0.18 0.355 0.09859 0.39437 0.50704

U 0.9 0.6 0.4 0.315 0.21 0.12 0.645 0.48837 0.32558 0.18605
P(IjÇSi) = P(SiÇIj): Joint Probability

P(Si | Ij): Posterior Probability

P(Si Ç I j ) P( I j | Si ) P(Si )
P(Si | Ij) = =
P( I j ) n
å P( I j | Si ) P(Si ) 8
i =1

64.51 S1 (0.09859) -20

S2 (0.39437)
40
d uce S3 (0.50704) 100 • Thus, the optimal strategy is to perform the survey, if the result of
64,51 Pro
F (0.355) S1 (0.09859) 10
43.90
54.23
S2 (0.39437) 45
the survey is favorable then the company should produce the
Buy S3 (0.50704) 70
U
product, however, if the result is unfavorable, the company should
(0. S1 (0.48837)
6 21.86 -20
45
) S2 (0.32558) buy the product.
vey

c e 40
32,56 Produ S3 (0.18605) 100
S ur

32.56 S1 (0.48837) 10
Buy S2 (0.32558) 45
• EVII = 43.90 – 40.25 = 3.65
43.90
S3 (0.18605) 70

37
S1(0.35)
-20
• So if the cost paid for this information is less than 3.65, it will be
No S2(0.35)
Su 40.25 Prod uce 40 worthy to acquire. Otherwise, the information will be worthless.
rve S3(0.30)
y 100
S1(0.35) 10
Bu 40.25
y S2(0.35) 45
S3(0.30) 70
 • Additional information reduces risk in decision making. It can in some • The value of EPVI is just simply the expected value under certainty minus the

extreme situations completely remove the risk and provide a certainty expected value under uncertainty.

environment. In General, such information is very expensive. EVPI = Expected Payoff - Expected payoff
under Certainty with no information
• How to evaluate the value of perfect information (EVPI)?
• To compute the expected value under certainty simply take the best payoff
• The idea behind EVPI is that if the state of nature that will occur is
under each state of nature and multiply it by its prior probability and sum
known with certainty, then the best alternative can be determined
these.
with certainty as well.
• EVPI places an upper bound on what one would pay for additional information.

• Expected Payoff under certainty = 10*0.35+45*0.35+100*0.30 = 49.25

Example 2
Company A is gearing up to introduce a new product to the market priced at
• EVPI = 49.25 – 40.25 = 9
20 TD per unit. Market conditions are anticipated to fall into three categories:
• This means that we should not be willing to pay more than 9. favorable, moderately favorable, and unfavorable, with respective

• The Efficiency of the imperfect information is the ratio of EVII to EVPI. probabilities of 20%, 30%, and 50%. Correspondingly, the estimated profits
under these scenarios are 100,000 TD, 60,000 TD, and a loss of 20,000 TD,
• As the EVPI provides an upper bound for the EVII, efficiency is always a
respectively.
number between 0 and 1.
1. Say whether the introduction of the new product would be beneficial for
• The efficiency of the survey = EVII/EVPI = (3.65)/(9) = 40.556% Company A.
Company A can gather additional information about the state of the market via unit enabling Company A to collect a profit of 27,000 TD. If negotiations fail,
a cost of 5,000 TD. Historical data indicates a 50% probability of accurately Company B has the right to withdraw its offer entirely, leaving Company A
predicting a given situation, and a 25% likelihood of misjudging one of the two without an avenue to introduce its product to the market. The prevailing belief
remaining scenarios. is that there's a 40% likelihood for the first event, a 50% chance for the
2. What is your advice to Company A? second, and a 10% probability for the third event to occur.
Company B has put forth an offer to buy all of Company A's production at 12 3. Assist Company A in making a decision.
TD per unit. Company A may accept and receive a profit of 25,000 TD. 4. Determine the expected value of perfect information in relation to
Alternatively, Company A has the option to counter with a price of 15 TD per Company B's response to Company A's potential counter-offer?
unit. If Company B consent to this counter-offer Company A stands to earn a
profit of 30,000 TD. Company B might also propose a revised offer of 13 TD per

18

(0.5) (0.5)
Example 3: Marketing a new product Favorable market Unfavorable market
Construct a large plant $200,000 -$180,000
Construct a small plant $100,000 -$20,000
Assume that a feasibility study at Getz company of a new product led to
Do nothing $0 $0
encouraging the introduction of this product to the market. The management
of Getz is not sure whether a large plant or a small one should be built to 2. Getz has the possibility to conduct a survey for $10,000. The survey will result in
a favorable or an unfavorable prediction. Past experience indicates a 70%
manufacture the product. The relevant data is presented in the table below.
probability of predicting favorable market conditions when the market is
1. A marketing research company requests $65000 for a perfect information.
favorable, and a 20% probability when the market is unfavorable. What should
How much does this information worth? be the optimal strategy of the company?
3. Calculate the efficiency of the survey
 19

(0.5) (0.5)
EMV
Favorable market Unfavorable market 2. Posterior distribution calculations
Construct a
$200,000 -$180,000 $10,000
large plant
Construct a States FAV UNF The P(FAV│ “.”) P(UNF│ “.”)
$100,000 -$20,000 $40,000
small plant 0.5 0.5 sum
Survey
Do nothing $0 $0 $0
“FAV” 0.7 0.2 0.35 0.10 0.45 0.78 0.22
“UNF” 0.3 0.8 0.15 0.40 0.55 0.27 0.73
1. EVPI = Expected Value Under Certainty - Max(EMV)
= ($200,000*0.50 + 0*0.50) - $40,000
= $100,000 - $40,000 = $60,000
So Getz should not be willing to pay more than $60,000

25

1st decision point 2nd decision point $106,400 Fav. Mkt (0.78)
$190,000
2
plant
Unfav. Mkt (0.22)
Example 3: Texaco vs. Pennzoil
$106,400

-$190,000
Large
Small $63,600 Fav. Mkt (0.78) $90,000
)
V”
(.4
5
No
plant 3 Unfav. Mkt (0.22) -$30,000
“FA
pla
nt
$49,200 -$87,400 Fav. Mkt (0.27)
-$10,000
1. Calculate the EVPI regarding Texaco’s reaction to a counteroffer of $5
1 “UN $190,000
F” ( 4 Unfav. Mkt (0.73)
plant
.5 5) -$190,000
billion? Can you explain this result intuitively?
y

Large
Surve

$2,400

$2,400 Fav. Mkt (0.27)

Small $90,000
No plant 5 Unfav. Mkt (0.73) -$30,000
$49,200

pla
nt -$10,000
No
s
$10,000 Fav. Mkt (0.5)
$200,000
2. The timing of information acquisition may make a difference.
urv 6 Unfav. Mkt (0.5)
ey nt
e pla -$180,000
$40,000

Larg Sma $40,000 Fav. Mkt (0.5)

No p plant
ll
7 Unfav. Mkt (0.5)
$100,000
-$20,000
a. suppose that Penzoil could obtain information about the final court
lant
$0
decision before making his current decision (taking the $2 billion or
Hence, if the survey results are favorable, the company should build a large
counteroffer $5 billion). What would be the EVPI of this information?
plant. However, if they are unfavorable, it should build a small plant.
The efficiency of the survey = EVII/EVPI = (19,200)/(60,000) = 32%
 b. Suppose that Penzoil knew that it would be able to obtain perfect
information only after it has made its current decision but before it
would have to respond to a potential Texaco counteroffer of $3 billion.
What would be the EVPI in this case?

3. In question 2, EVPI for (b) should be less than EVPI calculated in (a). Can
you explain why?

4. What is EVPI if Liedtke can learn both Texaco’s reaction and the final
court decision before he makes up his mind about the current $2 billion The decision tree shows that his best current choice is to make the $5 billion
offer? Can you explain why the interaction of the two bits of information counteroffer with an expected payoff of $4.63 billion. The decision tree shows clearly
should have this effect? what Liedtke should do if Texaco counteroffers $3 billion: He should refuse.

1.
Texaco accept Refuse Counter-offer 3 EMV
(0.17) (0.5) (0.33)
Accept 2 2 2 2 2
Couter- 5 4.56 4.56 4.63
offer 5

EVPI = EMV(Information) - EMV(Counteroffer $5 billion)

= (5*0.17 + 4.56*0.5 + 4.56*0.33) – 4.63
= 4.63 - 4.63 = 0.
Counteroffering $5 billion is optimal for each of Texaco’s 3 responses.
2. a.
EVPI = EV(under certainty) - EMV
= $4.98 billion - $4.63 billion
= $0.35 billion.
This assumes that Penzoil is the only who receives the information regarding
the court award.

2. b.
EVPI = EV(under certainty) - EMV
= 4.93 - 4.63
= $0.30 billion.
3. The earlier the information, the more valuable it is. In fact, we saw in
part (a) that, if the award turned out to be zero, Pennzoil would accept
the $2 billion. Obtaining the information later does not allow him to use
the information in this way.
4.
EVPI = EV(with Information) - EMV(without) = 5.23 - 4.63 = $0.60 billion.

Note that 0.60 > 0 + 0.35.

The information about Texaco’s reaction alone was not enough to be

worthwhile. However, this information along with the court information
helps Pennzoil refine its strategy. Now it knows exactly what to do in each
case, especially when the court award = 0.

Decision & Game Theory Introduction

²The expected value criterion offers an advantage of simplicity by
Chapter 5: Utility Functions
summarizing all the available information on a one score.

²However, this advantage represents as well his main weakness. In fact, by

summarizing the information, the criterion loses much of it, which can
Sonia REBAI - Naceur Azaiez
make its application dangerous.
Tunis Business School
²The paradox of St. Petersbug, analyzed in the 18th century by Niclas
University of Tunis
Bernoulli, displays this phenomenon.
² Consider the following game: Toss repeatedly a fair coin and obtain a profit
2n if head appears on the nth trial. As soon as head appears, the game is ² The EV criterion is not adequate to approach such situations.
completed. How much are you willing to pay for this game? ² In addition, several real situations require the consideration of intangible
² Based on expected value criterion, to access the game we need to pay an factors that are difficult to express by a simple measure of the EV.
amount equal to the expectation of winning. The expected payoff of the
² Suppose you have a choice between the following two options:
game is k
k =¥
kæ 1 ö § Toss a coin and win 1000 TD if head and zero if tail
VE = å 2 ç ÷ = 1+1+ = ¥
k =1 è2ø
§ or accept the sum of 600 TD.
² Even if the expected payoff is infinite, still many people would not pay more
than few amount of money.

² We are interested in a measure that takes into consideration any

qualitative or quantitative objective . ² Also, it takes into consideration the fact that the utility of any additional
² We need to use a new measure taking into account the risk factor and monetary unit decreases with the increase of capital.
reflecting the preferences of the DM (his attitude towards risk). ² The utility function is an ordinal concept (as opposed to a cardinal
² How can we model a DM’s preferences? concept).
² We introduce in the following an approach called utility theory that
² It permits to assign a value to each choice according to the preferences
allows us to incorporate the riskiness of an alternative when deciding on
of the decision maker.
the best course of action.
² Such a measure must satisfy a number of axioms of rationality
² The theory of utility allows individuals to have different attitudes towards
money and against hazards.
 Von Neumann and Morgenstern (1944) were the first to establish
Von Neumann-Morgestern axioms
the axiomatic theory of the rule of expected utility.
Notations & Terminology
Total Ordering Axiom
²A lottery L is any uncertain event having random outcomes.
Consider two lotteries L and L’, then exactly one of the following
²Consider two lotteries L and L’, then
types of preferences must hold:
§ L is more preferred than L’ is denoted by: L  L’
L  L’, L’ L, or L » L’
§ L is less preferred to L’ is denoted by L ≺ L’
Transitivity Axiom
§ L is equally preferred to L’ is denoted by L » L’
For all lotteries L, L’, L”, If L≻L' and L'≻L'' Then L≻L''

Independence Axiom Von Neumann-Morgenstern Theorem

For all lotteries L, L’, and L’’, for all p Î [0,1], we have that
There exists a utility function U so that for any lotteries L and L’ such that
L ≻ L' ⇔ pL +(1− p)L'' ≻ pL' +(1− p)L'' L ≻ L' if and only if U(L) ≥ U(L') (1)
Continuity Axiom How to develop a utility function?
For all lotteries L, L’, and L’’
If L ≻ L' ≻ L'' ⇒ there is p ∈ [0,1] such that ²Set the utility of the best outcome (denoted by x*) to be 1 and that of
L' = pL + (1− p)L''
the worst (denoted by x0) to be 0.
An agent accepting all four axioms above is called Von Neumann-
Morgestern rational or just VNM-rational. U(x*) =1 and U(x0) =0
²Let x be an intermediate outcome and let p a probability in [0,1]. Then, ²It is important to note that x is not necessarily a monetary value. It can
ask the DM whether he prefers x with certainty or the lottery, L, leading
even be a qualitative attribute.
to x* with probability p and x0 with probability 1-p.
An alternative approach: Keeney & Raiffa 5-points procedure.
²By trial & error find p so that the DM becomes indifferent between both
lotteries. ²It is often difficult for DM to manipulate probabilities. However, they feel
²Set U(x) = p. x is called the certainty equivalent of the lottery L, denoted more comfortable playing with 50-50 lotteries.
by CE(L). ²With this approach, we proceed as follows:
²Take other intermediate values of x and find their corresponding p.
§ U(x*) = 1 and U(x0) = 0.
²Fit the curve (x, u(x)).

§ Then, the DM is asked to determine CE(L), where L(x*, 0.5; x0, 0.5),
DM attitudes toward risk
denoted x0.5. It is obvious that U(x0.5)= 0.5.
²A DM is averse to risk if for any lottery L, the certainty equivalent of L is
§ Next, find the CE of the lottery L(x*, 0.5; x0.5 , 0.5) denoted x0.75, with
U(x0.75) = 0.75. worst than the expected value of the lottery L. In profit case, we have

§ Next, find the CE of the lottery (x0.5, 0.5; x0, 0.5) denoted x0.25, with E(L) > CE(L)

U(x0.25) = 0.25. § The difference RP(L) = E(L)-CE(L) > 0 is called the risk premium of L.

§ Fit the curve passing through (x*,1); (x0.75,0.75); (x0.5,0.5); (x0.25,0.25); and § RP(L) is the amount of money the DM is willing to pay to avoid risk.
(x0,0).
 ²A DM is prone to risk, risk seeker, or risk taker if for any lottery L,
²A DM is risk-neutral if for any lottery L, the certainty equivalent
the certainty equivalent of L is better than the expected value of
of L is equal to the expected value of L
the lottery L. In profit case we have E(L) < CE(L)
E(L) = CE(L)

§ Therefore, the risk premium is zero § The risk premium of L, RP(L)=E(L)-CE(L) < 0

RP(L) = E(L)-CE(L) = 0 § RP(L) is the amount of money the DM is willing to receive in

addition to the expected value of the lottery to give up risk.

Profit case
Convexity/Concavity and Attitude toward risk

²A function f(x) is said to be convex if

" x, y; "α Î [0, 1]; f(αx+(1-α)y) ≤ αf(x)+(1-α)f(y)

²If f is differentiable, then f is convex if its derivative is an increasing

function

²If f is twice differentiable, then it is convex if its second derivative is

non-negative: f”(x)≥0.
 Convex Function ²A function f(x) is said to be concave if –f is convex.

"x, y,"αÎ [0,1], f(αx+(1-α)y) ≥ αf(x)+(1-α)f(y)

²If f is differentiable, then f is concave if its derivative is a decreasing

function

²If f is twice differentiable, then it is concave if its second derivative is

non-positive: f”(x)≤0.

Interpretation
Concave Function

²Let the lottery L have values x with probability p and y with probability 1-p.

²If the DM is neutral to risk, he would accept the expected value criterion.

²Therefore, he will be indifferent between the average z with certainty and

the risky lottery L. (z = E(L) = (px+(1-p)y))

²That is, u(z)=u(px+(1-p)y) = u(E(L)) = E(u(L)).

Cost case
That is, ²If the DM is risk-averse, then the
u(E(L))=u(px+(1-p)y) = E(u(L)) certain situation, z, is preferred to
U(x)
U(x)

U(z)= U(E(L))
the lottery L
U(z)
= pU(x)+(1-p)U(y)
U(L)=
pU(x)+(1-p)U(y) ²z=px+(1- p)y
U(y)
U(y) ²U(z)=U(E(L))≥E(U(L)).
x Z y
x z y
²The utility function is concave

Theorem
²If the DM is risk-seeker, the
U(x)
² A DM is risk-averse if and only if his utility function is concave.
lottery L is preferred to the
U(L)=
certain situation z (px+(1-p)y) ² A DM is risk-neutral if and only if his utility function is linear.
pU(x)+(1-p)U(y)
U(z)
²U(z) = U(E(L))≤E(U(L)) ² A DM is risk-seeker if and only if his utility function is a convex

U(y)
function
²The utility function is convex
x z y
 U(M) U(M) U(M)
Example
An individual owns a house with a value of 150,000 TND. The risk to lose his

house from fire or other catastrophic event is 0.1% each year. His utility

Risk Averse
M
Risk prone
M
Risk Neutral
M function for the house is of the form:

U (x) = ax + b

²The higher the curvature the more averse or seeker the DM. How much he is willing to pay for an insurance company to preserve the value
²The same DM may have different risk attitudes based on the
of his asset? What is the corresponding risk premium?
problem situation or the values considered.

Let us find x, such that U(x) = 0.999; or the certainty equivalent of L, CE(L).
Solution
Let u be the utility function of the loss of the asset of the DM, where u(0) 1
Then solving 0.999 = − x +1
=1 and u(150,000)=0. Then, b=1 and a=-1/150,000. That is, 150,000

1
u(x)= − x +1
150,000 yields x = 299.850 = 300 TND

The risk premium RP(L) = 300 -150,000*0.001 = 150 TD

Consider now the lottery L: [(0; 0.999), (150,000; 0.001)]. Then, E(u(L))=
That is the DM is willing to pay 150 TD to avoid losing his asset.
0.999.
 Decision & Game Theory
What is Game Theory?
ü Game theory is a mathematical discipline that studies the behaviors and
Chapter 6
decisions of agents placed in a situation of interdependence.
Introduction to Game Theory
ü The main originality of game theory is that the players are aware not only

Sonia REBAI - Naceur Azaiez of their own goals but also of those of other protagonists.
ü Decision-makers are supposed to be rational and to pursue exogenous
Tunis Business School
objectives
University of Tunis

ü Game theory concerns social problems for which you are not the only Three elements should be specified
person who makes a decision. You interact with others.
ü Players: individuals, firms, countries, etc. A player has the ability to choose
ü You need to think about what is best for you that depends on what
among a set of possible actions. The specific identity of the players is
others do. Such a situation is called a strategic situation.
irrelevant to the game.
ü Game theory helps in examining and predicting how people behave in a
ü Strategies can be very simple or very complex but each is assumed to be
strategic situation.
well-defined.
ü Game theory provides you a unified way for solving many kinds of
ü Payoffs: The final returns of the players at the end of the game. They are
problems.
measured in terms of utility obtained by the player.
 Types of Games
ü The goal of each player is to maximize his utility, which depends on his
Games are generally divided into two categories:
decisions (strategies) and that of all other players.
✓ Cooperative games: players can make irrevocable agreements with each
ü Each player is free to choose his decision, but not that of the others. other.

✓ Non-cooperative games:
ü They take into account the knowledge they have or the anticipations they
• Either it is assumed that players can not make irrevocable agreements
make of the behavior of other decision-makers.
before engaging in the action (impossibility of communicating or a
prohibition of consultation between competitors).

• or it is assumed that the players are in no way compelled to take any Description of Non-Cooperative Games
action respecting the agreements that they may have previously A non-cooperative game can be described in two ways
concluded between them.
1. A matrix form: it consists in giving strategies and payoffs for each set of
Further, players.
✓ Sequential vs. Simultaneous moves
2. An extensive form: it provides an extended description of the game as a
✓ Single Play vs. Iterated tree revealing outcomes from each set of players’ strategies and possible
✓ Zero vs. non-zero sum actions that each set of players can take in response to other players’
✓ Perfect vs. Imperfect information moves. This description is suitable for dynamic games or those with

✓ Static vs. Dynamic incomplete information or with uncertainty.

Examples
Player R
The children’s hand game
R S P
ü Each player simultaneously makes a figure of rock, paper, or scissors R (D,D) (W,L) (L,W)
Player F
ü Outcome: Rock dominates scissors, scissors dominate paper, and paper S (L,W) (D,D) (W,L)
P (W,L) (L,W) (D,D)
dominates rock. In a two-person game, player who makes dominating
figure wins the game. When both make same figure, it’s a draw and
Within the parentheses the first payoff is player F’s and the second is for
neither player wins
player R.

ü For this example sequential moves put player who moves first at a
disadvantage. The other player will always choose an action that
results in a win.

ü When players simultaneously reveal their actions, neither player has

any prior information on the actions of the other player.
Battle of the Sexes Prisoners’ Dilemma

Ø A wife and husband may either go to shopping or to a football match ✓ Two suspects are arrested for a crime. They are
• both prefer spending time together put in different rooms and being interrogated
• the wife prefers shopping and the husband prefers football by police force.
Wife Football Shopping ✓ Both players have two choices. Either to

Husband cooperate or to defect. Cooperation means

Football 3, 2 0, 0 remain silent.

Shopping 0, 0 2, 3 ✓ The prosecutor wants to extract a confession

so he offers each a deal.

✓ The Deal

• “if you defect to confess, but your companion remains silent, you will 2 Cooperate Defect
1
be set free, and your companion will be harshly punished for 15
Cooperate -1, -1 -15, 0
years”
Defect 0, -15 -10, -10
• “if you both defect to confess, you will each get a ten-year sentence”

• “if both remain silent, each will get a one-year sentence”

Dominance
Let’s denote by ü A strategy xi of the player i strictly dominates a strategy xi’ if:

ü fi : the payoff of player i in terms of the choices of all players. It can be a

f i (xi , x−i ) > f i (xi ', x−i ) for all x−i
profit, an assessment of psychological satisfaction,… (In any case, a player
prefers the issues that give him the best value of fi).
ü A strategy xi of the player i weakly dominates a strategy xi’ if
ü xi : strategy of player i
f i (xi , x−i ) ≥ f i (xi ', x−i ) for all x−i
ü x-i : strategies of the other players.

Equilibrium of Dominant Strategies

✓ If a player has a strictly dominant strategy, then he chooses it without ✓ By assuming that the dominated strategies will never be played, we can

hesitation since his best choice is independent of that of the other eliminate them.

players. ✓ If this elimination process converges towards a single outcome, everyone

✓ When all players have a strictly dominant strategy, an equilibrium of
can unambiguously anticipate the behavior of the others. The obtained
dominant strategies exists that is determined without having to consider
outcome is called equilibrium by the elimination of strictly dominated
the behavior of the other players.
strategies.
✓ The existence of an equilibrium in weakly dominant strategies is almost
miraculous. In general, a player must consider other players’ strategies.
He may then reduce his set of strategies based on rational behavior.
 Example
✓ Such equilibrium only rarely exists. The elimination process often stops
while the players still have several alternatives, none of them being Player 2
strictly dominated. A B C
1 (4,2) (1,4) (2,2)
✓ The process of elimination can be multiple. Player 1
2 (3,0) (1,5) (2,8)
3 (1,4) (1,4) (3,4)
✓ The process can converge to several possible outcomes according to the
order in which the eliminations succeed. But there is no reason to
² The successive elimination of the strategies 2, C, 3 and A converges
assume that players follow the same order in the elimination process.
towards (1,B) whereas the successive elimination of A, 1, B and 2
converges towards (3, C).

Decision & Game Theory Zero-Sum Games

✓ Zero-sum games are games for which player strategies can not increase
Chapter 7 or decrease the available resources. The sum of the gains is constant
Two-Person Zero-Sum Games and therefore if the gain increases for one of the players it decreases
inevitably for the other player.
Sonia REBAI - Naceur Azaiez ✓ The Battle-of-the-Sexes game or the Prisoner's dilemma are non-zero
sum games.
Tunis Business School
University of Tunis
 Player 2
A B C ✓ A two-players constant-sum game is a two-player game in which, for
1 (7,-7) (4,-4) (3,-3) any choice of both player’s strategies, the row player’s reward and
Player 1 2 (9,-9) (5,-5) (2,-2) the column player’s reward add up to a constant value C .
3 (6,-6) (3,-3) (1,-1)
✓ A two-players zero-sum game is just a two-players constant-sum
This payoff table corresponds to a zero-sum game and can be simplified as
game with C = 0.
follows: Player 2
A B C
1 7 4 3
Player 1 2 9 5 2
3 6 3 1

Pure Strategy - Saddle Point Player 2

A B C Min
✓ A game is a pure strategy, if the optimal strategy for a player is the same
1 7 4 3 3 Max
whatever the game of the 2nd. PLayer 1 2 9 5 2 2
3 6 3 1 1
✓ In a game of pure strategy, the 1st player identifies the minimum (what the
Max 9 5 3
2nd player allows him) for each of his strategies. Then he selects the
maximum among the identified values. In other words the 1st player Min

maximizes the minimum allowed gains (offensive). Player 1 should play strategy 1 and player 2 should play C.
Since min-max = max-min, we say that the game possesses a saddle point
✓ However, the 2nd player minimizes the maximum loss (defensive).
leading to a stable solution.
✓ A saddle point is the combination of strategies in which each player can ✓ Every saddle point is an equilibrium point but not every equilibrium point
find the highest possible payoff assuming the best possible play by the is a saddle point.

opponent. ✓ A saddle point occurs when each player is achieving the highest possible
✓ None of the players will benefit from changing his move in an attempt payoff and thus neither would benefit from changing strategies if the other
to make advantage of the opponent strategy to improve his own didn’t also change - which is why it is also called an equilibrium point.
position. ✓ The optimal strategies and value for a two-players constant-sum game may
✓ When there is a saddle point, both players should stick to their min-max be found by the same methods used to find the optimal strategies and
and max-min solutions respectively. value for a two-players zero-sum game.
✓ A game does not always possess a saddle point.

Example Two players and Zero-Sum Games: Randomized Strategies

During the 8 to 9 P.M. time slot, two networks are vying for an audience of
B C Min
100 million viewers. The networks must simultaneously announce the type
1 1 5 1
of show they will air in that time slot. The possible choices for each network
2 7 4 4 Max
and the number of network 1 viewers (in millions) for each choice are shown Max 7 5
below. Does this game have a saddle point? What is the value of the game to
Network 2 Min
network 1?
S1 S2 S3 ✓ If player 1 chooses 1, player 2 will choose B. Player 1 will then change
S1 35 15 60
his preference for strategy 2. Hence, player 2 will change for C.
Network 1 S2 45 58 50
S3 38 14 70
✓ A game is said to have randomized or mixed strategies, if a player needs
to change strategy when the game is repeated. Let denote by

pi = probability that player 1 will choose strategy i (i=1,2,…,m).

✓ Whenever a game does not have a saddle point, game theory advises
each player to assign a probability distribution over his/her set of qj = probability that player 2 will choose strategy j (j=1,2,…,n).
strategies.

✓ Each player has to look for the frequency to play each of the strategies
so that the result will be independent of the policy adopted by the
other player.

Matrix 2x2 ü Similarly, for the 2nd player, the probability q of playing strategy B is
B C
1 1 5 chosen so that the expected value of his loss is independent of the game
2 7 4 of the 1st player.

ü Let P be the probability of playing strategy 1. The expected value of the 1*q + 5*(1-q) = 7*q + 4*(1-q) or q = 1/7
gain is independent of the game of the 2nd iff
ü Thus, player 2 must play 1/7 of the time the strategy B and 6/7 of the
1 * P + 7 * (1-P) = 5*P + 4*(1-P) or P = 3/7
time the strategy C. His expected loss is
ü Thus, player 1 must play 3/7 of the time the 1st strategy and 4/7 of the
time the 2nd strategy. His expected profit = 31/7 1 * 1/7 + 5 * 6/7 = 31/7
 Player 2
Matrix 2xm A B C D
1 16 4 18 6
Player 1
2 16 22 12 18

In this case we can use a graphical solution procedure

ü Let P be the probability of playing strategy 1. The expected payoff of
player 1 depends on the choice of the second player.
𝜋! = 𝜋 (p, A) = 16
𝜋" = 𝜋 (p, B) = P * 4 + 22 * (1-P) = 22-18p
𝜋# = 𝜋 (p, C) = 18 * p + 12 * (1-P) = 6p + 12
𝜋$ = 𝜋 (p, D) = P * 6 + 18 * (1-P) = 18-12p Remember that player 2 wants to minimize the expected payoff for player 1.

ü Player 2 can minimize the expected payoff for player 1 by choosing a

ü The 2nd player will then play strategies C or D. If we denote by q the
strategy corresponding to the bottom line in the figure.
probability of playing strategy C, then
ü Player 1 wants to maximize this minimum expected payoff. Consequently,
18*q + 6*(1-q) = 12*q + 18*(1-q)
player 1 should select the value of p where the bottom line peaks. Then
the Max-min is determined by q = 2/3

𝜋# = 𝜋$ Hence p* = 1/3. ü Thus player 2 must play 2/3 of the time strategy C and 1/3 of the time
strategy D. His expected loss = 14
ü Thus player 1 must play 1/3 of the time the 1st strategy and 2/3 of the
time the 2nd strategy. Game value = 14
Matrix mxn Let pij = the payoff under strategies i, j respectively for player 1 and 2
(i=1,2,…,m) and (j=1,2,…,n)
ü Let xj= the probability that Player 1 will use strategy j, (j =1,2,…,m).

ü yj = the probability that Player 2 will use strategy j, (j=1,2,…,n). Player 2

1 2 .. .. n
ü Player 1 will choose x1, x2,…, xm so that his expected payoff is maximized. 1 P11 P12 .. .. P1n
2 P21 P22 .. .. P2n
ü On the other hand, Player 2 will choose y1, y2,…, yn so that his expected Player 1 : : : :
payoff is minimized. : : : :
m Pm1 Pm2 .. .. Pmn

Player 1 will solve the following LP Player 2 will solve the following LP (dual of the player 1’s LP
Note that the jth
Max v constraint implies that the Min w The ith constraint in the column
s.t. p11x1+ p21x2+ …+ pm1xm ≥ v s.t. p11y1+ p12y2+ … +p1nyn ≤ w player’s LP implies that if the
expected reward against
p12x1+ p22x2+ … + pm2xm ≥ v p21y1+ p22y2+ …+ p2nyn ≤ w
column j must at least row player chooses row i, then
. .
equal v; otherwise, the the column player’s expected
. .
column player could hold . losses cannot exceed w ;
.
p1nx1+ p2nx2+ … + pmnxm ≥ v the row player’s expected pm1y1+ pm2y2+ … +pmnyn ≤ w otherwise, the row player could
x1+x2+ …+xm= 1 reward below v by y1+y2+ …+yn=1 obtain an expected reward that
x1, x2, …, xm ≥0 y1, y2, …yn ≥ 0 exceeded w by choosing row i .
choosing column j.
Example
A B 1st player
1 14 6 Max V
14X1+2X2 ≥ V
2 2 12
6X1+12X2 ≥ V
X1+X2=1
Let denote by X1, X2 ≥ 0
2nd player
² X1 and X2 the probabilities of playing respectively Strategies 1 and 2. Min W
14Y1+6Y2 ≤ w
² Y1 and Y2 the probabilities of playing respectively Strategies A and B.
2Y1+12Y2 ≤ w
Y1+Y2 = 1
Y1, Y2 ≥ 0

ü Game theory became a field since the book of John von Neumann (a
computer scientist and mathematician and theoretical physicist) and
Decision & Game Theory
Oskar Morgenstern (a professor of economics at Princeton University)
published in 1944.
Chapter 8
Nash Equilibrium

Sonia REBAI - Naceur Azaiez

O. Morgenstern

Tunis Business School J. von Neumann

University of Tunis
ü The founders of game theory argued that all social problems can be ü Six years later the answer to this open

formulated as a simple mathematical model of game by specifying the question was discovered by a

players, the strategies and the payoffs. mathematical genius, John Nash.

ü They posed the question about finding a unified principle or governing ü He received his Ph.D. from Princeton
principle which can be applied to all social. University with a 28-page thesis on his
22nd birthday.
ü Specifically they ask the following question “Is there any single solution

concept which can be applied to all social interactions?” that remains an ü He invented the notion of Nash
open question. equilibrium.
John Nash(1928-2015)

ü He had a problem of schizophrenia, and he ü Nash found the answer in a cup of coffee.
couldn't continue his research.
ü Nash get the ingenious idea of considering
ü Sooner It was discovered that his unified
the surface of a coffee as the set of all
principle was very useful in addressing lots of
economics questions. possible human behavior in a social

problem.
ü In 1994 he received Nobel Prize in economics.
ü If you stir the surface of a coffee, you get a The vortex
ü His life was so interesting that it was made into a
movie called A Beautiful Mind. vortex, a point that doesn't move.
 ü A point in this picture represents
Nash Equilibrium
best replies original behavior of people in a
ü A Nash equilibrium is a set of strategies, one for each player, that are each
social situation.
best responses against one another.
ü The destination represents
another social situation where ü In a two-player game, a Nash equilibrium is a pair of strategies (a*,b*)
All players players are moving towards best such that a* is an optimal strategy for player A against b* and b* is an
Original behavior of are doing
players replies. optimal strategy for player B against a*.
their best
against ü At the vortex point, all players are
others ü A Nash equilibrium is where the strategy of one player is the best strategy
doing their best against the others
when other players play their best strategy.
ü This is called Nash equilibrium

ü It does guarantee that, if a pure-strategy Nash equilibrium does not exist, a

ü A Nash equilibrium is a situation where no player has an interest in
mixed-strategy Nash equilibrium does.
deviating (alone) from the obtained situation.
ü Every game has a Nash equilibrium, and some games may have several Nash
ü Not all games have a Nash equilibrium in pure strategies. They may equilibria.
have Nash equilibrium in mixed strategies.
ü A drawback to Nash equilibrium is unclear how a player can choose a best
ü Nash proved the existence of a Nash equilibrium in all finite games. response strategy before knowing how rivals will play

ü The existence theorem does not guarantee the existence of a pure- ü A quick way to find the Nash equilibrium is to underline the best-response

strategy Nash equilibrium. payoffs. The Nash equilibria correspond to the boxes in which every player’s
payoff is highlighted.
Example

tell if the following games have a pure Nash equilibrium

Husband
ü The Battle-of-the-Sexes game O F
O (4,2) (1,1)
ü The children’s hand game Wife
F (0,0) (2,4)
ü The Prisoner’s Dilemma

Two Nash equilibria (O,O) and (F,F).

Player R
R S P C D
R (D,D) (W,L) (L,W)
Player F C -1, -1 -15, 0
S (L,W) (D,D) (W,L)
P (W,L) (L,W) (D,D) D 0, -15 -10, -10

The children’s hand game has no pure Nash equilibrium D is always best, irrespective of the opponent’s strategy

Rational players play Nash equilibrium (D,D)

ü A strategy that is a best response to any strategy the other players might Does the following games have Nash equilibrium?

choose is called a dominant strategy.

Player 2 Player 2
A B A B C
ü Defect is a dominant strategy for both players in the prisoner dilemma 1 (3,7) (0,0) 1 (3,0) (0,2) (0,3)

Player
2 (4,1) (5,3) Player 1 2 (2,0) (1,1) (2,0)

1
game. 3 (6,5) (4,7) 3 (0,3) (0,2) (3,0)

Player 2
ü When a dominant strategy exists, it is the unique Nash equilibrium. A B C D
1 (5,2) (5,3) (7,5) (6,3)

Player 1
ü There is no dominant strategy in the battle of the sexes game. 2 (2,0) (6,5) (4,5) (4,3)
3 (0,8) (4,3) (3,7) (5,6)
4 (4,2) (2,2) (3,0) (8,3)

Solve the following game To find the pure Nash equilibria, we need to identify the best response for
each player with regard to each possible strategy chosen by his opponent.
A B C D
1 38 16 38 18 Each identified pair of best response strategies corresponds to a pure Nash
2 34 52 30 42
3 18 50 14 26 equilibrium. In the following examples, the best responses for each player
with regard to the choice of his opponent are tinted in red and the pure
Nash equilibria are highlighted in yellow.
Player 2 Player 2
A B A B C
1 (3,7) (0,0) 1 (3,0) (0,2) (0,3)
Player

2 (4,1) (5,3) Player 1 2 (2,0) (1,1) (2,0)

1

3 (6,5) (4,7) 3 (0,3) (0,2) (3,0)

 A B C D Strategy C dominates strategy A
1 38 16 38 18
Player 2 2 34 52 30 42 B C D
A B C D 3 18 50 14 26 1 16 38 18
1 (5,2) (5,3) (7,5) (6,3) 2 52 30 42
Player 1 2 (2,0) (6,5) (4,5) (4,3) Strategy 2 dominates strategy 3 3 50 14 26
3 (0,8) (4,3) (3,7) (5,6)
B C D Let’s denote by p the probability that
4 (4,2) (2,2) (3,0) (8,3)
1 16 38 18 player 1 chooses Strategy 1.
2 52 30 42 The payoffs of player 1 are as follows:

Π! = 16𝑝 + 52 1 − 𝑝 = 52 − 36𝑝
Π" = 38𝑝 + 30 1 − 𝑝 = 30 + 8𝑝
Π# = 18𝑝 + 42 1 − 𝑝 = 42 − 24𝑝

60
Optimal value of p is obtained by
50 solving Π" = Π# ; 30 + 8𝑝 = 42 − 24𝑝
40 then 𝑝 = 3/8
30 Player 1 should play strategy 1 with a
20
probability 3/8 and probability 2 with
probability 5/8
10
However, player 2 should play
0
0 0,2 0,4 0,6 0,8 1 strategies C and D.
Π_B Π_C Π_𝐷 Let’s denote by q the probability to play
strategy C. The payoffs of player 2 are
Π$ = 38𝑞 + 18 1 − 𝑞 = 18 + 20𝑞
Π% = 30𝑞 + 42 1 − 𝑞 = 30 − 12𝑞 then as follows:
Π$ = Π%; 𝑞 = 3/8

Player 2 should play strategy C with a probability 3/8 and probability D with probability 5/8