Professional Documents
Culture Documents
INOPER3 Notes
Definition
Decision Theory (DT) is a set of
concepts, principles, tools, and
techniques that aid the decision maker
in dealing with complex decisions under
uncertainty.
Components of a DT Problem
The Decision Maker
Alternative Courses of Actions
This is the controllable aspect of the problem
States of Nature or Events
These are the scenarios or states of the
environment not under the control of the decision
maker. The events defined should be mutually
exclusive and collectively exhaustive.
Components of a DT Problem
Consequences
The consequences that must be assessed
by the decision maker are measures of the
net benefit, payoff, cost or revenue
received by the decision maker. There is a
consequence (or vector of consequences)
associated with each action-event pair. The
consequences are summarized in a
decision matrix.
Classification of DT Problems
Single Stage Decision Problems
A decision is made only once.
Multiple Stage/Sequential Decision Problems
Decisions are made one after another.
Discrete DT Problems
The alternative courses of actions and states of nature are finite.
Continuous DT Problems
The alternative courses of actions and states of nature are infinite.
DT Problems can also be classified as those with or without
experimentation. Experimentation is performed to obtain
additional information that will aid the decision maker.
Discrete Decision Theory
Problems
Decision Trees
A discrete DT problem can be represented
pictorially using a tree diagram or decision tree. It
chronologically depicts the sequence of actions
and events as they unfold.
A square node ( ) precedes the set of possible
actions that can be taken by the decision maker. A
round node ( ) precedes the set of events or
states of nature that could be encountered after a
decision is made. The nodes are connected by
branches ( ).
Decision Tree
E1
2
A1
E2
E1
A2
3
E2
Decision Under Risk and
Uncertainty
If no probabilities are assigned to the
possible consequences, then the
decision situation is called "decision
under uncertainty".
If probabilities are assigned then the
situation is called "decision under risk".
Decision Under Risk and
Uncertainty
Consider a DT Problem with m alternative
courses of actions and maximum of n events
or states of nature for each alternative course
of action.
Define:
Ai = alternative course of action i; I = 1,2,…,m
Øj = state of nature j; j = 1,2,…,n
Decision Under Risk and
Uncertainty
The decision matrix of payoff is given by:
Ø1 Ø2 … Øn
: : : : :
Am v(Am, v(Am, … v(Am,
Ø1) Ø2) Ø n)
Decision Under Risk and
Uncertainty
A. Laplace Criterion
This criterion is based on what is known as the
principle of insufficient reason. Here the
P(
probabilities associated with the occurrence of the
event Øj is unknown. We do not have sufficient
reason to conclude that the probabilities are
different. Hence, we assume that all events are
equally likely to occur.
1
P ( j ) j
n
Decision Under Risk and
Uncertainty
A. Laplace Criterion (con’t.)
Then, the optimal decision rule is to select
action Ai* corresponding to:
1 n
max v( Ai , j )
Ai
n j 1
Example
Blockwood Inc. is a newly organized manufacturer of furniture products.
The firm must decide what type of truck to purchase for use in the
company's operations. The truck is needed to pick up raw material
supplies, to make deliveries and to transport product samples to
commercial exhibits during the coming year. Three alternatives were
identified by the firm:
A small commercial import truck
A standard size pick-up
A large flatbed truck
It is expected that sales in the first year will fall in one of four
categories:
Low (0-200,000)
Moderately low (200,000-400,000)
Moderately high (400,000-600,000)
High (above 600,000)
Example (con’t)
The payoff table for the firm would be:
Actions States of Nature
Truck Type Low Moderately Low Moderately High High
Import 20 10 15 25
Standard 15 25 12 20
Faltbed -20 -5 30 40
•Find the appropriate decisions using
•Laplace Criterion
•Minimax Criterion (Assume loss payoff table)
•Maximin Criterion
•Savage Minimax Regret Criterion
•Hurwicz Criterion ( = 0.6)
Decision Under Risk and
Uncertainty
B. Minimax (Maximin) Criterion
This is the most conservative criterion
since it is based on making the best of the
out of the worst possible conditions. For
each possible decision alternative, we
select the worst condition and then select
the alternative corresponding to the best of
the worst conditions.
Decision Under Risk and
Uncertainty
The Minimax strategy is given by:
if v is loss
Once the loss matrix is constructed using the
above formula, we can now apply the Minimax
criterion defined in b.
Decision Under Risk and
Uncertainty
D. Hurwicz Criterion
This criterion represents a range of attitudes from the
most optimistic to the most pessimistic.
Under the most optimistic conditions, one would
choose the action yielding:
max maxv( Ai , j )
Ai j
Under the most pessimistic conditions, the chosen
action corresponds to:
max minv( Ai , j )
Ai j
Decision Under Risk and
Uncertainty
D. Hurwicz Criterion (con’t.)
The Hurwicz criterion strikes a balance between
the extreme pessimism and extreme optimism by
weighing the above conditions by respective
weights α and (1- α), where 0<α<1. That is, the
action selected is that which yields:
max
max v ( Ai , j ) (1 ) min v ( Ai , )
j
Ai j j
[Note: the above formulas represent the case
where payoffs are expressed as profits]
Decision Under Risk and
Uncertainty
D. Hurwicz Criterion (con’t.)
If α=1, the decision rule is referred to as the
n
EP max Pj * v( Ai , j )
Ai
j 1
EVPI is easily seen as a measure of the maximum
amount a decision maker should be willing to pay
for additional information.
Decision Under Risk and
Uncertainty
G. Decision Making with Experimentation
(con’t.)
Define:
EVSI = expected value of sample information
ENGS = expected net gain from sampling
CAI = cost of getting additional information
Decision Under Risk and
Uncertainty
G. Decision Making with Experimentation
(con’t.)
Then:
l n
EVSI P{Z k } max P j / Z k * v( Ai , j ) EP
Ai
k 1 j 1
Where:
hØ(Ø) = prior distribution function of the states of nature
Continuous Decision Theory
In the above model, it is assumed that no
additional information is available and the
expression is evaluated with respect to the
prior distribution of the states of nature. If
additional information is available, we update
the prior distribution of the states of nature
by determining its posterior distribution,
which is nothing but the conditional
distribution of the states of nature given the
experimental outcome.
Continuous Decision Theory
Hence, the optimization converts to:
max f ( A) v( A, )h / Z z ( )d
Where:
hØ/Z=z(Ø)=conditional distribution of the state of
nature given the experimental outcome
Continuous Decision Theory
hz / ( z )h ( )
h / Z z
hz ( z )
hz/Ø(z) = conditional distribution of the
experimental outcome given the state
of nature
hz(z) = marginal distribution function of
the experimental outcomes