DECISION THEORY

Presenter:
Apolonio Cabangal Jr.
 TOPICS TO DISCUSS

Introduction
Payoff Matrix
Regret Table
Decision Making Under Uncertainty
Different Methods of Decision Making Under
Uncertainty
Decision Making Under Risk
 TOPICS TO DISCUSS(CONTINUED)
Different methods of decision making under risk
Maximum likelihood principles
Maximum expectation principles
Concept of decision tree
Decision making with perfect information and
economic value of perfect information(EVPI)
Decision making with revised probabilities(Using
Baye’s Theorem)
Expected opportunity loss/regret principle
 INTRODUCTION

External environment imposes certain constrain

on us. Based on our response to such constrains,
we get different payoffs.
We cannot change the payoff matrix.
At best we can use the available information
judiciously to arrive at the optimal decision and
maximize our payoffs in the long run
 EXAMPLE
A wholesaler of fruits buy strawberries at P 20 a case
and sells them at P 50 a case. The product is perishable
by nature and cannot be stored overnight. It has be sold
on the day of purchase itself.
From experience, the wholesaler knows that the daily
demand will range between 10 to 13 cases.
Every case of strawberries bought and would not sold
will lead to a marginal loss of P 20, while every case that
could not be sold because of stock out would lead to an
opportunity loss of P 30.
 PAYOFF MATRIX IN PHP
POSSIBLE POSSIBLE STOCK ACTION
DEMAND IN IN CASES
CASES 10 11 12 13

10 300 280 260 240

11 300 330 310 290
12 300 330 360 340
13 300 330 360 390
 REGRET TABLE IN PHP
Possible Possible Stock Action in Cases
Demands in 10 11 12 13
Cases

10 0 20 40 60
11 30 0 20 40
12 60 30 0 20
13 90 60 30 0
 DECISION UNDER UNCERTAINTY
Here, we are no information about the likelihood(probability)
of any particulars state of demand.
In such a condition, we are making decision under uncertainty.
The following principles used to take decisions under
uncertainty:
 Laplace principles

 Maximin or Minimax criterion

 Maximax or Minimin criterion

 Hurwicz criterion

 Salvage principle
 LAPLACE PRINCIPLE
 Assumes all external Stock Payoff(PHP)
constrains(here
demand of
Action(Cases)
strawberries in cases 10 300
to be equiprobable)
 Maximum Payoff is 11 317.5
achieved by pursuing
the strategy of 12 322.5
stocking 12 units.
13 315
 MAXIMIN. OR MINIMAX. PRINCIPLE
Here, minimum payoffs Stock Action(Cases) Min. Payoff(PHP)
from each strategy with
max. of this profits is 10 300
selected. This is known as
maximin principle. For
cost, the strategy with min 11 280
of maximum cost is
chosen. That is known as
12 260
minimax. Principle. This
is pessimistic decision
making. 13 240
Maximum payoffs is
achieved by pursuing the
strategy of stocking 10
 MAXIMAX. OR MINIMIN PRINCIPLE
Here are the maximum payoffs Stock MAX.
from each strategy with max. of Action(cases) Payoff(PHP)
these maximum profit is
selected. This is known as 10 300
maximax. principle for cost,
the strategy with min. of the
11 330
minimum cost is chosen. That is
known as minimin. principle.
This is highly optimistic 12 360
decision making.
Maximum payoff is achieved by 13 390
pursuing the strategy of
stocking 13 units.
 HURWICZ PRINCIPLE
Index of optimism = α
Criterion Value= α(Max. Profit) + (1- α)(Min. Profit)
For costs,
Criterion Value = α(Min. Cost) + (1- α)(Max. Cost)
Criterion Value = α(Max. Payoff) + (1- α)(Min. payoff)
α=0 stand for maximin and minimax principle
α=1 stand for maximax or minimin principle
For our example, let us assume for an index of optimism
of 60%(α=0.6)
 HURWICZ PRINCIPLE CONTINUED
The strategy with Stock Hurwicz
Action(Cases) Criterion Value
maximum Hurwicz (Php)
criterion value is
chosen. 10 300
Here, the
11 310
maximum hurwicz
criterion value is 12 320
achieved pursuing
the strategy of 13 330
stocking 13 units.
 SALVAGE PRINCIPLE
Here, we select the Stock Max. Regret
strategy that Action(Cases) Value(PHP)
minimizes the
10 90
maximum regret. This
is also pessimistic 11 60
decision making.
Here, the minimum of 12 40
max. regret(salvage)
13 60
value is achieved by
pursuing the strategy
of stocking 12 units.
 DECISION MAKING UNDER RISK
Now, let us assume that based on the sales of past
100 days, the wholesaler has the following
information about the market demand.
Daily Sales(Cases) No. of Days Sold Prob. Of Demand
10 15 0.15
11 20 0.20
12 40 0.40
13 25 0.25
Total 100 1.00
DECISION MAKING UNDER RISK(CONTINUATION)

Here, we have additional information on the

probability of each demand state.
When we have probabilities associated with each
demand state available to us, the decision making
techniques is called decision making under risk.
The following principles are used to arrived at the
optimal decision.
Maximum likelihood principle
Maximum expectation principle
Minimum expected opportunity loss/regret principle
 MAXIMUM LIKELIHOOD PRINCIPLE
Here, we choose to stock according to that demand state
which has maximum probability of occurrence. In this
case, the wholesaler should stock 12 cases, if he adopts
this principles.
Daily Sales(Cases) Prob. Of Demand
10 0.15
11 0.20
12 0.40
13 0.25
Total 1.00
 MAXIMUM EXPECTATION PRINCIPLE

Here, we choose that strategy which has maximum

expected payoff. This is the most acceptable
principle since, the expected payoff will always
come true in the Long Run.
In our example, the wholesaler should choose that
strategy which will maximize his expected profit.
Now, we have to examine the expected profit for
each strategy(stock action).
EXPECTED PROFIT FROM PURSUING A
STRATEGY OF STOCKING 10 CASES
Demand in Conditional Prob. of Expected
Cases Profit Demand Profit
(PHP)

10 300 0.15 45.0

11 300 0.2 60.0

12 300 0.4 120.0

13 300 0.25 75.0

TOTAL 1.00 300.0

EXPECTED PROFIT FROM PURSUING A STRATEGY
OF STOCKING 11 CASES

Demand in Conditional Prob. of Expected Profit

Cases Profit Demand (PHP)

10 280 0.15 42.0

11 330 0.20 66.0

12 330 0.40 132.0

13 330 0.25 82.5

Total 1.00 322.5

EXPECTED PROFIT FROM PURSUING A STRATEGY
OF STOCKING 12 CASES

Demand in Conditional Prob. of Expected Profit

Cases Profit Demand (PHP)

10 260 0.15 39.0

11 310 0.20 62.0

12 360 0.40 144.0

13 360 0.25 90.0

Total 1.00 335.0

EXPECTED PROFIT FROM PURSUING A STRATEGY OF
STOCKING 13 CASES

Demand in Conditional Prob. of Expected Profit

Cases Profit Demand (PHP)

10 240 0.15 36.0

11 290 0.20 58.0

12 340 0.40 136.0

13 390 0.25 97.5

Total 1.00 327.0

 DECISION TREE

So, the maximum profit come from pursuing a

strategy of stocking 12 cases.
Now, how do we represent this logic
diagrammatically? Well, we use a diagram called
decision tree.
In a decision tree, we represent a decision(strategy)
with a rectangle while an outcome(demand in this
case) is represented by a circle.
DECISION TREE FOR THE WHOLESALER PROBABILITIES
10,45.0, 0.15

11,60,0.2
P 300.0

12,120.0,0.4

13,75,0.25
P 322.5
P 335.0

P 335.0

P 327.5
DECISION MAKING WITH PERFECT INFORMATION
(EVPI)
If we go back to the payoff matrix, we have marked our
best decisions for each demand scenario. If we could
have perfect information about the market demand, our
expected payoff table would have looked like this.
Payoff matrix in PHP.
Expected Profit
Market Conditional Prob. of
in
Demand(Cases) Profit Demand
(PHP)
10 300 0.15 45.0
11 330 0.2 66.0
12 360 0.4 144.0
13 390 0.25 97.5
Total 1.00 352.5
DECISION TREE FOR THE WHOLESALER WITH PERFECT
INFORMATION
10, 300.0

P 300.0 11, 280.0

12, 260.0

P 330.0 13, 240.0

$ 352.5

P 360.0

P 390.0
 ECONOMIC VALUE OF PERFECT INFORMATION
( EVPI)

Increase in expected profit with perfect

information.
= ( 352.5- 335) P
= 17.5 P
This is known as economic value of perfect
information(EVPI)
DECISION MAKING WITH REVISED PROBABILITIES

Let, there be a market research firm, that provides additional

information(forecasting) about the possible state of demand, and
charges a fee for the same.
What is the additional value of this forecast and how much can the
wholesaler pay for it?
The agency forecasts the demand by rating it as above normal(an) or
below normal(BN)
It has been observed in the past that in 80% of the instances, when
the demand was 10 cases, the agency had forecasted below
normal(BN). In 60% instances, when the demand was 11 cases, the
agency had forecasted below normal(BN). In 30% of the instances,
when the demand was 12 cases, the agency had forecasted below
normal(BN). In 20% of the instances, when the demand was 13
cases, the agency had forecasted below normal(BN).
 REVISED PROBABILITIES
(USING BAYE’S THEOREM)

P(Forecast P(Forecast& Revised Prob.

Forecast Event P(Event)
/Event) Event) (Event/Forecast)
10 0.15 0.2 0.03 0.05
Above
11 0.2 0.4 0.08 0.14
Normal
12 0.4 0.7 0.28 0.47
(AN)
13 0.25 0.8 0.2 0.34

TOTAL 1.00 P(AN) = 0.59 1.00

10 0.15 0.8 0.12 0.29

Below
11 0.2 0.6 0.12 0.29
Normal
12 0.4 0.3 0.12 0.29
(BN)
13 0.25 0.2 0.05 0.13

TOTAL 1.00 P(BN) = 0.41 1.00

 DECISION TREE FOR THE REVISED PROBABILITIES
DO NOT BUY FORECAST
A
P 335.0 10
B
AN, 0.59

P 348.14
C

P 335.165 P 335.165 13
D

A’ 11
BUY FORECAST

B’
P 316.5
EVPI= P 0.165
C’
BN, 0.41
D’ 12
 NODE ‘A’

STOCKING 10 CASES
10,300.0,0.05

P 300.0

11,300,0.14

12,300.0, 0.47

13, 300.0, 0.34

NODE ‘B’
STOCKING 11 CASES
10,280.0,0.05

P 327.46

11,330,0.14

12,330.0, 0.47

13, 330.0, 0.34

NODE ‘C’
STOCKING 11 CASES
10,260.0,0.05

P 348.14

11, 310.0,0.14

12,360.0, 0.47

13, 360.0, 0.34

NODE ‘D’
STOCKING 11 CASES

10,240.0,0.05

P 345.08

11, 290.0,0.14

12,340.0, 0.47

13, 390.0, 0.34

NODE ‘ A’ ’
STOCKING 10 CASES
10,300.0,0.29

P 300.0

11,300,0.29

12,300.0, 0..29

13, 300.0, 0.13

NODE ‘ B’ ’
STOCKING 11 CASES
10,280.0,0.29

P 315.50

11,330,0.29

12,330.0, 0..29

13, 330.0, 0.13

NODE ‘ C’ ’
STOCKING 12 CASES
10,260.0,0.29

P 316.50

11,310,0.29

12,360.0, 0..29

13, 360.0, 0.13

NODE ‘ D’ ’

STOCKING 13 CASES
10,240.0,0.29

P 303.00

11,290,0.29

12,340.0, 0..29

13, 390.0, 0.13

MINIMUM EXPECTED REGRET/OPPORTUNITY LOSS PRINCIPLE

This principle would give same answer as the

previous principle.
This is because, ER(j)+Ep(j)=EPPI (Expected
Payoff under Perfect Information)
Hence, the strategy that would minimize ER(j)
would automatically maximize EP(j).
The Min. ER(j) is also the EVPI in this case.
EXPECTED OPPORTUNITY LOSS/REGRET FOR DIFFERENT
STOCK ACTIONS

Stock Action(Cases) Expected Regret (PHP)

10 52.5

11 30.0

12 17.5

13 25.0
 EVSI AND EFFICIENCY OF EVSI

In our example, economic of sample

information(EVSI)=(335.165- 335) P
= 0.165 P
Efficiency of EVSI = 0.165/335 = 0.05 %
QUESTIONS PLEASE

THANK YOU