DECISION THEORY

The Decision Maker has to make a decision by making use of the information available to
him, regarding states of nature, courses of action, and the corresponding Pay-offs.
The decision making situations or environments can broadly be divided into following types.
● Under Certainty

● Under Uncertainty

● Under Risk

● Under Conflict

DECISION MAKING UNDER UNCERTAINTY

1) Maximax Criterion
2) Maximin Criterion
3) Laplace Criterion
4) Minimax Regret Criterion

Example 1: Given the following pay-off table, decide the best decision using the following
criteria
(i) Maximax Criterion/ LEONID HURWITZ criteria
(ii) Maximin Criterion/ ABRAHAM WALD criteria
(iii) Laplace Criterion/ Thomas Bayes and Simon De LAPLACE criteria
(iv) Minimax Regret Criterion/ L.J./ Savage criteria
Course of Action State of Nature

S1 S2 S3

A1 15 25 10

A2 0 30 20

A3 50 20 -10

A4 40 20 15

Solution:
 Pay-off Table
Course of State of Nature Maximum Minimum Average
Action Pay-off Pay-off Pay-off

S1 S2 S3

A1 15 25 10 25 10 16.67

A2 0 30 20 30 0 16.67

A3 50 20 -10 50 -10 23.33

A4 40 20 15 40 15 25.00
(i) Maximax Criterion / Leonid Hurwitz Criteria
i.e. Maximum(25,30,50,40) = 50
∴ The decision under Maximax Criterion is to choose action A 3 with pay-off 50.

(ii) Maximin Criteria / Abraham Wald Criteria

i.e. Maximum(10,0,-10,15) = 15
∴ The decision under Maximin Criteria is to choose action A 4 with pay-off 15.

(iii) Thomas Bayes and Simon De Laplace Criterion / Equal Likelihood Criterion
i.e. Maximum(16.67, 16.67, 23.33 ,25) = 25
∴ The decision under Laplace Criterion is to choose action A 4 with pay-off 25.

(iv) Minimax Regret Criterion / Savage Criterion

Regret OR Opportunity Loss Table

Course of State of Nature Maximum
Action pay-off
S1 S2 S3

A1 35 5 10 35

A2 50 0 0 50

A3 0 10 30 30

A4 10 10 5 10
i.e. Minimum(35, 50, 30, 10) = 10
∴ The decision under Minimax Criterion is to choose action A 4 with minimum value 10.
 DECISION MAKING UNDER RISK
In this situation, the decision maker has the knowledge of probabilities of occurrence of each
state of nature, from past experience. These probabilities are obtained from past records or from
personal judgment of the decision maker. Thus a 'Prior distribution' of probabilities along with
various states of nature is known to the Decision Maker. It is called 'Probability Distribution'.
For the decision making under risk, the most popular methods used are
(A) EMV (Expected Monetary Value) Criterion
(B) EOL (Expected Opportunity loss) Criterion
(C) EPPI (Expected Profit with Perfect Information)
(D) EVPI (Expected value of Perfect Information)

Example1: A farmer wants to decide which of the three crops he should plant on his farm.
The profit from each crop is dependent on the rainfall during the season. He has
categorized the rainfall as substantial, moderate or light. He estimates his profit for each
crop as shown in the table:
Estimated Profit in Rs.
Rainfall

Substantial Moderate Light

Crop A 7000 3500 1000

Crop B 2500 3500 4000

Crop C 4000 4000 3000

Depending on the projection of the coming season, he estimates the probability of
substantial rainfall as 0.2, of moderate rainfall as 0.3 and that of light rainfall as 0.5.
(i) Determine the optimal decision as to which crop to plant using EMV criterion.
(ii) Do we need perfect information to arrive at a decision? Is it worth it?
(iii) EVPI =EPPI-EMV

Solution:
 Pay-off table
Course of Action State of Nature

Substantial Moderate Light

Crop A 7000 3500 1000

Crop B 2500 3500 4000

Crop C 4000 4000 3000

Probability 0.2 0.3 0.5

Expected pay-off table

Course of State of Nature EMV

Action
Substantial Moderate Light

Crop A 1400 1050 500 2950

=(7000*0.2) =(3500*0.3) =(1000*0.5) =(1400+1050+500)

Crop B 500 1050 2000 3550

Crop C 800 1200 1500 3500

∴ EMV is Maximum for crop B, the optimal decision for the farmer is to plant crop B.

(ii) The maximum of these expectations is the Crop B. Not knowing which crop will be
demanded (only knowing the probability of the rainfall), we expect to make the most profit with
the Crop B.
Thus, EMV = 3550.
On the other hand, consider if we did know ahead of time which way the market would turn.
Given the knowledge of the direction of the market we would (potentially) make a different
decision in taking extra care of another crop.

(iii)Expectation for maximizing profit given the state of the rainfall:

EPPI = 7000×0.2 + 4000×0.3 + 4000×0.5 = 4600
That is, given each market direction, we choose the investment Crop that maximizes the profit.
Hence,
EVPI =EPPI-EMV
 = 4600 - 3550
= 1050.
Conclusion:
Knowing the direction the market will go (i.e. having perfect information) is worth ₹1050.

Example 2: A bakeryman has observed the following demand pattern for cakes produced
in his bakery:

No. of cakes in 20 21 22 23
demand

Probability 0.05 0.25 0.30 0.40

The cost of producing one cake is ₹ 12/- and the selling price is ₹ 20/- per cake.
(i) Use EMV criterion to determine how many cakes the bakery man should produce so as
to maximize his profit.
(ii) Determine optimum EOL.

Solution:
Since the demand (state of nature) is for 20, 21, 22, 23 cakes, the bakery man should produce
(course of action) either 20, 21, 22, 23 cakes.
Let A1,A2,A3,A4 denote the action of producing 20, 21, 22, 23 cakes respectively.
Let S1,S2,S3,S4 denote the demand for 20, 21, 22, 23 cakes respectively.

∴ Pay-off table will consist of four courses of action and four states of nature.
We need to calculate conditional pay-off (profit) for every combination of course of action and
state of nature. This can be done as follows:
If Demand ≤ Production, Profit = 20 × demand - 12 × production &
If Demand > Production, Profit = (20-12) production.

Pay-off Table
 Production Demand

S1 = 20 S2 = 21 S3 = 22 S4 = 23

A1 = 20 160 160 160 160

=(20*20) = (8*20) = (8*20) =(8*20)
-(12*20)

A2 = 21 148 168 168 168

=(20*20) -
(12*21)

A3 = 22 136 156 176 176

=(20*20) -
(12*22)

A4 = 23 124 144 164 184

=(20*20) -
(12*23)

Probability 0.05 0.25 0.30 0.40

Expected pay-off table

Production Demand EMV

S1 = 20 S2 = 21 S3 = 22 S4 = 23

A1 = 20 8 40 48 64 160
= (160*0.05) =(8+40+48+64)

A2 = 21 7.4 42 50.4 67.2 167

A3 = 22 6.8 39 52.8 70.4 169

A4 = 23 6.2 36 49.2 73.6 165

∴ Maximum EMV is ₹ 169/- which corresponds to action A3. Therefore the best action is to
produce 22 cakes.

(ii) To calculate EOL first create a Regret Table.

 Regret Table
Production Demand

S1 = 20 S2 = 21 S3 = 22 S4 = 23

A1 = 20 0 8 16 24

A2 = 21 12 0 8 16

A3 = 22 24 12 0 8

A4 = 23 36 24 12 0

Probability 0.05 0.25 0.30 0.40

Expected Opportunity Loss

Production Demand EOL

S1 = 20 S2 = 21 S3 = 22 S4 = 23

A1 = 20 0 2 4.8 9.6 16.4

=2+4.8+9.6

A2 = 21 0.6 0 2.4 6.4 9.6

A3 = 22 1.2 3 0 3.2 7..4

A4 = 23 1.8 6 3.6 0 11.4

Since EOL is Minimum for action A3 , the optimum action is to produce 22 cakes.

DECISION TREE
Example: A company has to decide about sales of a new product among three regions I, II
and III. The estimated sales figures in region I are 20,000 units, 18,000 units and 15,000
units if the demand is high, fair and low respectively. For region II, the sales figures are
21,000 units, 19,500 units and 14,000 units for these demand conditions respectively.
Whereas, the estimates of sales III are 24,000 units, 14,000 units and 12,000 units
respectively for these demand situations in region III. The probabilities associated with the
demand being high, fair and low are 0.25, 0.45 and 0.3 respectively. Suggest the optimal
region using the Decision Tree Method.
The expected sales are 17600,15150, 15900. So the best region is region I with a maximum
EMV of 17600 units.