Chapter 4.

Decision Theory
• The decision analysis is used to determine
optimal strategies - decision maker is faced with
several decision alternatives and an uncertain, or
risky, pattern of future events.
• All decision making situations are characterized
by the fact that two or more alternative course of
action are available to the decision maker to
choose from.

1
 4.1. Basic concepts in Decision Making
The decision making process involves the following steps:
1. Identification of the various possible outcomes
(state of natures, events). The events are
beyond the control of the decision maker.
2. Identification of all the courses of action
(strategies) available to the decision maker. The
decision maker has control over choice of these.
3. Determination of the pay off function which
describes the consequences resulting from
different combination of events and acts.
4. Choose among the various alternatives on the
base of some criteria.

2
 4.2. Decision Making under certainty
• It is the simplest of all circumstances occurs when the
decision making takes place in an environment of
complete certainty.

• The main characteristics of this decision making is that

the decision maker know, with certainty, which event will
occur. He has perfect information about the outcomes.

Degree of certainty
• There can be different degrees of certainty.
– Complete certainty
– Complete Uncertainty
– Risk

3
• Example, if we certainly know the demand is say
medium in the coming five years the best
alternative (the best outcome) is to construct
mid size (just right) plant

4
 4.3. Decision Making Under Uncertainty
• The decision situation where there is no way in
which the decision maker can assess the
probabilities of the various states of nature.
• In such situations, the decision maker has no idea at
all as to which of the possible states of nature
would occur nor has a reason to believe why a given
state is more, or less likely to occur as another.
• With probabilities of the various outcomes unknown
the actual decisions are based on specific criteria.
• There are several decision criteria to recommend a
solution for this kind of problems.

5
4.3.1. Maximax (Minimin) decision rule
 The maximax principle is optimists’
principle of choice.
 This optimist criterion attempts to
describe the decision-making behavior of
people who are perfectly optimistic in
their expectations.
 An optimistic decision maker is attracted
by large rewards and is willing to risk high
losses in order to obtain them.

6
 Steps in Maximax
Maximax" is shorthand for "Maximum of the (row) maxima."
• For each action alternative (matrix row) determine the
maximum payoff possible.
• From these maxima, select the maximum payoff. The
action alternative leading to this payoff is the chosen
decision.
• It is possible to model the optimist profile with the
MAXIMAX decision rule (when the payoffs are positive-
flow rewards) such as profits or revenue.
• When payoffs are given as negative-flow rewards, (such
as costs) the optimist decision rule is MINIMIN. Note that
negative-flow rewards are expressed with positive
numbers.)
7
Example

8
The decision is shown by circling the
maximax value, thus indicating the row of
the chosen action alternative:

9
 4.3.2. Maximin or Minimax principle (Wald Criterion)
• This principle is adopted by pessimistic decision makers who are
conservative in their approach. They assume that the worst
outcome will occur. Then it chooses the alternative that gives the
best of these worst outcomes.

• It is possible to model the pessimist profile with the MAXIMIN

decision rule (when the payoffs are positive-flow rewards, such as
profits or income.

• When payoffs are given as negative-flow rewards, the pessimist

decision rule is MINIMAX.)
Rule
1. For each action alternative (matrix row) determine the minimum
payoff possible. This represents the worst possible outcome if that
strategy were chosen.
2. From these minima, select the maximum payoff. The action
alternative leading to this payoff is the chosen decision
10
Using our Example defined previously:

11
 4.3.3.Savage decision criterion
• Sometimes we are judged not by how well we actually do, but how
well we could possibly have done.
• In such cases there is a regret- the difference between the actual
outcome and best possible outcome.
• It is the decision criterion that we discussed by basing the
opportunity cost of the alternatives (savage MiniMax Regret)
• The Minimax Regret criterion focuses on avoiding the worst
possible consequences that could result when making a decision
• Regret is defined as the opportunity loss to the decision maker if
action alternative Ai is chosen and state of nature Sj happens to
occur.
• It is an approach that does take all payoffs into account.
• Opportunity loss is the payoff difference between the best possible
outcome under Sj and the actual outcome resulting from
choosing Ai given that Sj occurs.

12
Savage Minimax Regret rule:
1. For each event find the best possible outcome (the best entry
in each column)
2. Find the regret for every entry in the column (the difference b/n
the best and the entry).
It is the amount of payoff the decision maker would miss by not
having chosen the alternative that would have yielded the best
payoff if that state of nature occurs.
3. Put the regrets found in step 2 in to regret matrix. There is at
least one zero in each column and regret are always positive.
4. For each alternative (row) find the highest regret
5. Choose the alternative with the lowest value of these highest
regrets.
Let's analyze our previous problem using opportunity losses
instead of the monetary payoffs. First it must derive the OL
matrix from the payoff matrix R.
13
 where the star denotes "best payoff under state of nature Sj ".
The OL matrix can now be obtained by subtracting each entry Rij from
its column's best payoff. The minimax rule is then applied to the OL
(regret) matrix:

14
 S
A
H M W
L 15-15=0 4-3=1 1- (-6)=7
JR 15-9= 6 4-4=0 1- (-2)= 3
S 15-3 =12 4-2=2 1 - 1= 0

15
 4.3.4.The Hurwicz Criterion
• The Hurwicz criterion attempts to find a
middle ground between the extremes
posed by the optimist and pessimist
criteria.
• Instead of assuming total optimism or
pessimism, Hurwicz incorporates a
measure of both by assigning a certain
percentage weight to optimism and the
balance to pessimism.

16
 Hurwicz Criterion…Cont’d
• A weighted average can be computed for every action
alternative with an alpha-weight α, called the coefficient of
optimism (realism).
• Note that 0 ≤ α ≤ 1
• An α = 1 denotes absolute optimism (Maximax) while an α = 0
indicates absolute pessimism (Maximin).
• The α is selected subjectively by the decision maker.
• Selecting a value for α simultaneously produces a coefficient
of pessimism 1 - α , which reflects the decision maker's
aversion to risk.
• A Hurwicz weighted average H can now be computed for
every action alternative Ai in A
H (Ai ) = α (row maximum) + ( 1 - α ) (row minimum) for
positive-flow payoffs (profits, revenues)
H (Ai ) = α (row minimum) + ( 1 - α ) (row maximum) for
negative-flow payoffs (costs, losses)
17
Hurwicz decision rule:
1. Select a coefficient of optimism value α .
2. For every action alternative compute its Hurwicz weighted
average H.
3. Choose the action alternative with the best H as the chosen
decision.
("Best" means Max {H} for positive-flow payoffs, and Min {H}
for negative-flow payoffs.)
• Let's assume that the managers have agreed to assess their
level of optimism at 60%. Thus α = 0.6. Using the decision
matrix defined previously:

18
Hurwicz Awarded Nobel in Economics

19
 4.3.5 Laplace Criterion
• It is based on the simple philosophy that if we are
uncertain about the various events then we may treat
them as equally probable.
• The Laplace criterion is the first to make explicit use of
probability assessments regarding the likelihood of
occurrence of the states of nature. As a result, it is the first
elementary model to use all of the information available in
the payoff matrix.
• The rationale behind the Laplace criterion is, to deal with
uncertainty rationally, probability theory must be invoked.
• This means that for each state of nature Sj in S, the
decision maker must assess the probability pj that Sj will
occur.
• It assumes that all state of natures to be equally probable.
P (Sj ) = 1/n 20
Laplace decision rule
1. Assign pj = P (Sj ) = 1/n to each Sj in S, for
j = 1, 2, ..., n.
2. For each Ai (payoff matrix row), compute its
expected value: E (Ai ) = Σj pj (Rij ), for i = 1,
2, ..., m.
Since pj is a constant in Laplace,
E (Ai ) = Σj pj (Rij ) = pj Σj Rij .
3. Select the action alternative with the best E (Ai )
as the optimal decision. "Best" means max for
positive-flow payoffs (profits, revenues) and min
for negative-flow payoffs (costs).
21
• Let's assume that all three market states (H, M, W ) to
be equally probable. Then and only then is the use of the
Laplace criterion warranted.

22
 4.4. Decision Making Under Risk
• The difference b/n decision making under
uncertainty and risk is the presence of
probabilities for the occurrence of the
various states of nature under decision
making under risk.
• The probabilities may be:
– Subjective estimates from managers
– Subjective estimates from experts in the field
– Historical frequencies

23
Decision Making Under Risk…Cont’d

† If they (probabilities) are reasonably

correct, they provide the decision
maker with additional information that
can dramatically improve the decision
making process.
† The sum of the probabilities for all
states of nature must be 1.00.

24
 4.4.1. Criteria of Expected Monetary Value (EMV)
• EMV approach provides the decision maker with a
value which represents an average payoff for each
alternative.
• The best alternative is, then, the one that has the
highest expected monetary value.
• The average or expected payoff of each alternative is
weighted average; the state of nature probabilities are
used to weight the respective payoffs.
• Thus the expected monetary value is:
• EMV (Ai ) = E (Ai ) = Σj pj ( Rij )

– And using the probability distribution estimated by

the managers of the company, we obtain:
25
EMV…Cont’d

26
4.4.2. Criteria of Expected Opportunity Loss

• Using EOL is an alternate method for

incorporating probabilities in to a decision
making process.
• The approach is nearly the same to the
EMV approach, except that a table of
opportunity losses is used rather than a
table of payoff.

27
• The opportunity losses for each
alternative are weighted by the
probabilities of their respective states of
nature, and the alternative with the
smallest expected loss is selected as the
best.

EOL (Ai ) = E (Ai ) = Σj pj ( OLij )

28
Expected Opportunity Loss…Cont’d
Opportunity Loss matrix table
1- Find state of nature (column) max
2- Deduct column (state of nature) elements from its respective max
3- Multiply step 2 by respective probability
4- Add step 3 in each row (alternative values in each row)
5-Select max in step 4

29
 4.4.4. Expected Value of Perfect Information
(EVPI)
• Sometimes the decision maker might opt
(decide) to delay a decision until it is evident
which state of nature is going to materialize
(be certain).
• Such delays will involve a cost of some sort.

• Hence, the question is whether the cost of

waiting outweighs the potential benefits that
could be realized.

30
 EVPI…Cont’d

• If it would be worth the cost to refine or

eliminate probabilities of states of natures.

• The EVPI is thus, the measure of the

difference between the certain expected
payoff that could be realized under a
condition of certainty and expected payoff
under condition involving risk.
31
 EVPI…Cont’d
• The probabilities of the original state of nature
(column) can be used to weight the payoff, one
of which will occur under certainty. This is called
the expected payoff under certainty (EPC).
• Column (state of nature) max with its probability needed
• EVPI = Σj pj (Rij*)
where Rij* is the best (max for profit or min for
cost) payoff under state Sj

32
 EVPI…Cont’d
Column (state of nature) max with its probability
needed

33
 EVPI…Cont’d
• EPC= 0.1(15) + .6(4) + .3(1) = 4.2

• The difference between this figure (4.2) and the

expected payoff under risk (EMV) is the
expected value of perfect information.
EVPI=EPC-EMV
EVPI= 4.2 - 2.7 =1.5

• Perfect information would increase expected

payoff by $1.5 million, so that is what the perfect
information is worth
34
 Attention !!!!
• The expected opportunity loss due to
imperfect information is equal to the
expected payoff that could be achieved by
having perfect information.

• Therefore EVPI is always equal to EOL

EVPI=EOL

35
 2.4.5. Decision tree approach

• Decision makers sometimes use tree like

diagram to obtain a visual portrays of decision
alternatives.
• A decision tree is a graphic representation of the
sequences of action - event combinations
available to the decision maker. `
• It is an alternative approach to payoff tables.
• It depicts in a systematic manner all possible
sequences of decisions and consequences.
Each such sequence is shown by a distinct path
through the tree.
36
 Decision tree…Cont’d
• It is most commonly used in portraying
sequential decisions (mostly difficult to show in
payoff table).

• A tree is composed of squares, circles, & lines.

– The squares indicate decision points.
– Circles represent chance events
– Lines emanate from the square represent alternatives
– Lines emanate from a circle represent state of nature.

• The expected payoffs are computed in the same

manner as in the previous cases.
37
 Decision tree…Cont’d
• A decision tree enables the decision maker to see
the various elements of his problem in proper
perspective and in a systematic manner.

• It may be mentioned that the criterion on the basis

of which the decisions are made in the decision tree
approach is generally the expectation principle.

• Thus we may choose the alternative that maximizes

the expected profit, or the alternative that minimizes
the expected cost and so on.

38
Example:
• An oil company has recently acquired rights in
certain area to conduct survey and test drillings
to lead to lifting oil if it is found in commercially
exploitable quantities.
• The area is considered to have good potential
for finding oil in commercial quantities. At the
outset, the company has the choice to conduct
further geological tests or to carry out drilling
programme immediately.
• On the known conditions, the company
estimates that there 70:30 of chances of further
test showing a ‘Success”.
39
• Whether the tests show the possibility of ultimate
success or not or even if no tests are undertaken
at all, the company could still pursue its drilling
programme or alternatively consider selling its
rights to drill in the area. Thereafter, however, if it
carries out the drilling programme, the likelihood
of final success or failure is considered dependent
on the following stages. Thus,
• If successful test have been carried out, the
expectation of success in drilling is given as 80:20
• If tests indicate failure, then the expectation of
success in drilling is given as 20:80.

40
• If no tests have been carried out at all the
expectation of success in drilling is given
as 55:45
• Costs and revenues have been estimated
for all possible outcomes and the net
present values of each is as follows:
• Outcome Value (in millions)
• Success:
– With prior tests 100
– Without prior tests 120

41
• Failure:
– With prior tests -55
– Without prior tests -40
• Sale of exploitation rights:
– Prior tests show success 65
– Prior tests shows failure 15
– Without prior tests 45
a) Draw up a decision tree diagram to represent the
above information
b) Evaluate the tree in order to advise the
management of the company on its best course of
action.
42
• The decision tree corresponding to the given problem is
depicted

43
 TABLE Evaluation of Decision Node 3

Alternative Outcome Prob. Conditional Value Expected Value

1. Drill Success 02 100 20

Failure 0.8 (50) (40)

Total (20)

2 Sell 1.0 15 15

44
 TABLE Evaluation of Decision Node 2

Alternative Outcome Prob. Conditional Value Expected Value

1. Sell 1.0 65 65

2 Drill Success 0.8 100 80

Failure 0.2 (50) (10)

Total 70

45
 TABLE Evaluation of Decision Node 1

Alternative Outcome Prob. Conditional value Expected value

Success 0.55 120 66
1. Drill Failure 0.45 (40) (18)
Total 48

Positive 0.7 70 49.0

2. Test
Negative 0.3 15 4.5

Total 53.5

3. Sell 1.0 45 45

46
47