Decision Making Under Uncertainty

:
Pay Off Table and Decision Tree

Decision Making Under Uncertainty
A set of quantitative decision-making techniques for decision situations where uncertainty exists

Decision Making
States of nature
± ±

±

events that may occur in the future decision maker is uncertain which state of nature will occur decision maker has no control over the states of nature

Payoff Table 

A method of organizing & illustrating the payoffs from different decisions given various states of nature A payoff is the outcome of the decision 

Payoff Table
States Of Nature Decision a b 1 Payoff 1a Payoff 1b 2 Payoff 2a Payoff 2b

Decision-Making Models Under Uncertainty 

Maximax
choose decision with the maximum of the maximum payoffs 

Maximin
choose decision with the maximum of the minimum payoffs 

Minimax regret
choose decision with the minimum of the maximum regrets for each alternative 

Hurwicz
±

±

choose decision in which decision payoffs are weighted by a coefficient of optimism, E coefficient of optimism (E) is a measure of a decision maker¶s optimism, from 0 (completely pessimistic) to 1 (completely optimistic) 

Equal likelihood (La Place)
±

choose decision in which each state of nature is weighted equally

Decision Making Under Uncertainty Example
States Of Nature Good Foreign Poor Foreign Decision Competitive Conditions Competitive Conditions Expand $ 800,000 $ 500,000 Maintain status quo 1,300,000 -150,000 Sell now 320,000 320,000

Maximax Solution
Expand: $ 800,000 Status quo: 1,300,000 Sell: 320,000 Maximum

Decision: Maintain status quo

Maximin Solution
Expand: $ 500,000 Status quo: -150,000 Sell: 320,000 Decision: Expand Maximum

Minimax Regret Solution
Good Foreign Competitive Conditions $ 1,300,000 - 800,000 = 500,000 1,300,000 - 1,300,000 = 0 1,300,000 - 320,000 = 980,000 Poor Foreign Competitive Conditions $ 500,000 - $500,000 = 0 500,000 - (-150,000) = 650,000 500,000 - 320,000 = 180,000

Regret Value Expand: $ 500,000 Status quo: 650,000 Sell: 980,000 Decision: Expand

Minimum

Hurwicz Solution
E = 0.3, 1- E = 0.7

Expand: $ 800,000 (0.3) + 500,000 (0.7) = $590,000 ** Status quo: 1,300,000 (0.3) -150,000 (0.7) = 285,000 Sell: 320,000 (0.3) + 320,000 (0.7) = 320,000 Decision: Expand
** Maximum

Equal Likelihood Solution
Two decisions, weight = 0.50 for each state of nature

Expand: $ 800,000 (0.50) + 500,000 (0.50) = $650,000 ** Status quo: 1,300,000 (0.50) -150,000 (0.50) = 575,000 Sell: 320,000 (0.50) + 320,000 (0.50) = 320,000 Decision: Expand
**Maximum

Decision Making With Probabilities 

Risk involves assigning probabilities to states of nature Expected value is a weighted average of decision outcomes in which each future state of nature is assigned a probability of occurrence 

Expected Value
n EV ( x) ! § p xi xi i !1 where xi ! outcome i p xi ! probability of outcome i

Expected Value Example
70% probability of good foreign competition 30% probability of poor foreign competition $ 800,000 (0.70) + 500,000 (0.30) = $710,000 EV(status quo) $1,300,000 (0.70) - 150,000 (0.30) = 865,000 Maximum EV(sell) $ 320,000 (0.70) + 320,000 (0.30) = 320,000 EV(expand) Decision: Maintain status quo

Case of Pay off Table application
An ICT (Information and communication technology) company wants to analyze the future of its business. There are 4 decision alternatives: expand the company, maintain status quo, decrease the business size up to 50% of the current size and sell the company. From the business analysis there will be two possibilities: good economic condition and bad economic condition. If the economic condition is good the profit of the expansion will be Rp. 900 million and only Rp. 400 million when the economic condition is bad. If the economic condition is good the profit of maintain status quo will be Rp. 1.000 million and only Rp. 50 million when the economic condition is bad. If the economic condition is good the profit of decrease the business will be Rp. 600 million and only Rp. 300 million when the economic condition is bad. When the company is sold the current price is Rp. 350 million. Solve this decision problem by using maximax, maximin, minimax, hurwicz (with alpha = 0.3) and Equal likelihood. Based on the analysis provide your best suggestion.

Sequential Decision Trees 

A graphical method for analyzing decision situations that require a sequence of decisions over time Decision tree consists of
Square nodes - indicating decision points Circles nodes - indicating states of nature Arcs - connecting nodes 

Decision tree basics: begin with no uncertainty
Example: deciding where to eat dinner 

Basic setup:
Trees run left to right chronologically. Decision nodes are represented as squares. Possible choices are represented as lines (also called branches). The value associated with each choice is at the end of the branch.

Japanese

North Side
Greek

Vietnam

South Side
Thai

Assigning values to the nodes involves defining goals.
Example: deciding where to eat dinner Taste
Japanese

versus

Speed

4

1

North Side
Greek

3 1

2 4

Vietnam

South Side
Thai

2

3

To solve a tree, work backwards, i.e. right to left.
Example: deciding where to eat dinner Speed
Japanese

1

North Side

Value =2 Greek

2 4

Value =4 Vietnam

South Side

Value =4 Thai

3

Decision making under uncertainty
Example: a company deciding whether to go to trial or settle a lawsuit

Win [p=0.6]

Go to trial
Lose [p= ]

Chance nodes are represented by circles. Probabilities along each branch of a chance node must sum to 1.

Settle

Solving a tree with uncertainty:
Win [p=0.6]

Go to trial -$3.7M
EV= -$3.7M

$0
EV= -$3.2M Lose [p=0.4]

-$.5M

-$8M

Settle -$4M

The expected value (EV) is the probability-weighted sum of the possible outcomes: pwinx win payoff + plosex lose payoff In this tree, ³Go to trial´ has a cost associated with it that ³Settle´ does not. We¶re assuming the decisionmaker is maximizing expected values.

Decision tree notation
Chance nodes (circles) Expected value of chance node (or certainty equivalent) Probabilities (above the branch) Win [p=0.6]

Go to trial -$3.7M
EV= -$3.7M

$0
EV= -$3.2M Lose [p=0.4]

-$.5M

Terminal values corresponding to each branch (the sum of payoffs along the branch).

-$.5M

-$8M

-$8.5M

Decision nodes (squares)

Settle -$4M
Value of optimal decision Running total of net expected payoffs (below the branch)

-$4m -$4M
Payoffs (below the branch)

Example of a Decision Tree Problem
A glass factory specializing in crystal is experiencing a substantial backlog, and the firm's management is considering three courses of action: A) Arrange for subcontracting B) Construct new facilities C) Do nothing (no change) The correct choice depends largely upon demand, which may be low, medium, or high. By consensus, management estimates the respective demand probabilities as 0.1, 0.5, and 0.4.

Example of a Decision Tree Problem (Continued): The Payoff Table
The management also estimates the profits when choosing from the three alternatives (A, B, and C) under the differing probable levels of demand. These profits, in thousands of dollars are presented in the table below:

A B C

0.1 Low 10 -120 20

0.5 Medium 50 25 40

0.4 High 90 200 60

Step 1. We start by drawing the three decisions

A B C

Step 2. Add our possible states of nature, probabilities, and payoffs
High demand (0.4) Medium demand (0.5) Low demand (0.1)

$90 $50 $10 $200 $25 -$120 $60 $40 $20

A B C

High demand (0.4) Medium demand (0.5) Low demand (0.1) High demand (0.4) Medium demand (0.5) Low demand (0.1)

Step 3. Determine the expected value of each decision
High demand (0.4) Medium demand (0.5)

$62
A

Low demand (0.1)

$90 $50 $10

EVA=0.4(90)+0.5(50)+0.1(10)=$62

Step 4. Make decision
High demand (0.4) Medium demand (0.5)

$62
A B C

Low demand (0.1)

$90 $50 $10 $200 $25 -$120 $60 $40 $20

$80.5

High demand (0.4) Medium demand (0.5) Low demand (0.1) High demand (0.4)

$46

Medium demand (0.5) Low demand (0.1)

Alternative B generates the greatest expected profit, so our choice is B or to construct a new facility

Format of a Decision Tree
Decision Point Chance Event
2

Payoff 1 Payoff 2

Payoff 3
1

B
Payoff 4
2

Payoff 5 Payoff 6

Case of Decision Tree application

See Attached Problem