1

Decision Making With Uncertainty

(EMSE 6020)

George Washington University

 School of Engineering and Applied Science

Chapter 4 – Evaluating Decision Trees

Dr. Basel Ali

Spring 2, 2020
2

Evaluating Decision Tree

• Decision trees can be evaluated using two methods:

• Expected monetary value
• Risk profile
3
Evaluating Decision Tree
Using EMV- Folding Back the Tree
• Procedure

1. Start at the right most part of the tree

2. Replace each uncertainty node with its respective

expected

value : E[X] = xxP(X=x)

3. At each decision node select the optimal decision and

cancel the paths to all other decisions
4

Evaluating Decision Tree

• Decision Tree Examples - 1

Success
$5,000
0.4
Invest

$0
0.6

Do Not Invest
$2,200
5

License
$23
Patent Awarded

(1-p) $22.9
Development Marketing

$-2 P $-2
Patent not Awarded

Stop development
$0
6

G (0.4)
1 C
0.7 S
Go

0.15 P
0 F

Don’t go
0.37
7

Strike Oil
$190,000
(0.1)

(0.9)
-$10,000
No Oil
 Strike Oil
$190,000
(0.1)
EMV = $10,000
Dril

(0.9)
-$10,000
No Oil

Don’t drill
0

Drill
With Consultation $190,000

EMV = $19,000 Strike Oil

(0.1) Don’t drill
0

Drill
-$10,000
(0.9)
No Oil Don’t drill
0
 Strike Oil
$190,000
(0.1)
Dril
EMV = $10,000
(0.9)
-$10,000
No Oil

Don’t drill
$0
Strike Oil
$190,000
(0.41)
Consult geologist Dril
EMV = $16,560 Good (0.59)
-$10,000
(0.23) No Oil
Don’t drill
$0
Strike Oil
$190,000
Dril (0.01)

Poor (0.99)
-$10,000
(0.77) No Oil
Don’t drill $0
10

Evaluating Decision Tree

• Decision Tree Examples - 2

A Success
Invest in $5,000
0.4
Opportunity A
A Flop
-$2,000
0.6
B Success
Invest in $1,000
0.9
Opportunity B
B Flop
$0
0.1
11

Evaluating Decision Tree

• Decision Tree Examples - 3

Success
$8,000
0.8
Invest High
Flop
-$2,000
0.2
Success
$5,000
0.4
Invest Low
Flop
$0
0.6
Do Not Invest
$2,200
12
Evaluating Decision Tree
Texaco-Penzoil
• 1984 Penzoil and Getty Oil agree to terms of merger
• Before deal formalized Texaco offers Getty Oil a better deal
• Getty Oil go back on on Penzoil Deal, sells to Texaco
• Penzoil sues and wins $11.1Bil
• Appeals court reduces the award to $10.3Bil
• Texaco vows to fight to the bitter end, even potentially declaring
bankruptcy
• Before Penzoil can file liens, Texaco offers $2Bil settlement
• Hugh Liedtke, President of Penzoil, is faced with a decision
• His advisers tell him $3-$5Bil is a fair settlement
13
Evaluating Decision Tree
Texaco-Penzoil – The Model
Accept $2Bil $2 Bill

Texaco Accepts $5Bil $5 Bill

.17
Award High $10.3 Bill
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil $5 Bill
.50 .50
Award Low $0 Bill
Texaco Counters .30
$3Bil Award High $10.3 Bill
.33 .20
Refuse Texaco
Counteroffer Award Medium $5 Bill
.50
Award Low
.30 $0 Bill
Accept
$3Bil $3 Bill
14
Evaluating Decision Tree
Texaco-Penzoil – The Model
Only fundamental
Accept $2Bil 2 objective is money

Texaco Accepts $5Bil 5

.17
Award High
.20
10.3 Other
Counteroffer Texaco Refuses counteroffers
Counteroffer Award Medium
$5Bil
.50 .50
5 possible
Award Low 0
Texaco Counters .30
$3Bil Award High 10.3 Court decision
.33 .20
can be in
Decision could be Refuse Texaco
Counteroffer Award Medium 5 [0,$10.3Bil]
anywhere between .50
$3-$5Bil Award Low
.30 0 What about
Accept bankruptcy
$3Bil
Could counter again 3
option?
15
Evaluating Decision Tree
Random Variable Decomposition
Decision: Accept $2Bil 2
Accept $2Bil
Texaco Accepts $5Bil 5
.17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco Counters .30
$3Bil Award High 10.3
.33 .20
Refuse Texaco
Counteroffer Award Medium 5
.50
Award Low
.30 0
Accept
$3Bil 3
16
Evaluating Decision Tree
Random Variable Decomposition
Decision: Accept $2Bil 2
Counteroffer $5Bil
If Texaco Counters Texaco Accepts $5Bil 5
then Refuse .17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco .30
Counters Award High 10.3
.33 $3Bil .20
Refuse Texaco
Counteroffer Award Medium 5
.50
Award Low
.30 0
Accept
$3Bil 3
17
Evaluating Decision Tree
Random Variable Decomposition
Decision: Accept $2Bil 2
Counteroffer $5Bil
If Texaco Counters Texaco Accepts $5Bil 5
then Accept .17
Award High
.20 10.3
Counteroffer Texaco Refuses
Counteroffer Award Medium 5
$5Bil
.50 .50
Award Low 0
Texaco .30
Counters Award High 10.3
.33 $3Bil .20
Refuse Texaco
Counteroffer Award Medium 5
.50
Award Low
.30 0
Accept
$3Bil 3
 18
Texaco-Penzoil
Folding Back the Tree
Accept $2Bil 2

Texaco Accepts $5Bil 5

.17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50
x P(X=x) xP(X=x)
Award Low 0 0.00 0.30 0.00
Texaco Counters .30 5.00 0.50 2.50
$3Bil Award High 10.30 0.20 2.06
10.3
.33 .20 E[X]= 4.56
Refuse Texaco
Counteroffer Award Medium 5
.50
Award Low
.30 0
Accept
$3Bil 3
 19
Texaco-Penzoil
Folding Back the Tree
Accept $2Bil 2

Texaco Accepts $5Bil 5

.17

Counteroffer Texaco Refuses

Counteroffer
$5Bil 4.56
.50

Texaco Counters
$3Bil
.33
Refuse Texaco
Counteroffer 4.56

Accept
$3Bil 3
 20
Texaco-Penzoil
Folding Back the Tree
Accept $2Bil 2

Texaco Accepts $5Bil 5

.17

Counteroffer Texaco Refuses

Counteroffer
$5Bil 4.56
.50

x P(X=x) xP(X=x)
Texaco Counters 4.56 0.83 3.78
$3Bil
5.00 0.17 0.85
.33
Refuse Texaco E[X]= 4.63
Counteroffer 4.56
21
Texaco-Penzoil
Folding Back the Tree
Accept $2Bil 2

Counteroffer
$5Bil 4.63

Note, however, that it is impractical to redraw the tree

each time so we simply put the expected value above
the chance node and the optimal value above each
decision node
22
Texaco-Penzoil
Folding Back the Tree
Accept $2Bil 2

Texaco Accepts $5Bil 5

4.63 .17
Award High 10.3
.20
4.63 Texaco Refuses 4.56
Counteroffer
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco Counters .30
$3Bil Award High 10.3
.33 .20
Refuse Texaco 4.56
Counteroffer Award Medium 5
4.56 .50
Award Low
.30 0
Accept
$3Bil 3
23

Some Decision Tree Terminology

• Decision Strategy
The collection of decision paths connected to a branch of the left most
node by selecting one alternative from each decision node along these
paths. You cannot select an uncertainty outcome as part of your
decision strategy!

• Decision Path
A path starting at the left most node up to the values at the end of a branch
by selecting one alternative from a decision node and by following one
outcome from each uncertainty node. This yields one potential outcome
of a decision strategy and occurrence of random outcomes i.e. the
value of a random variable

• Optimal Decision Strategy

That Decision Strategy which results in the highest EMV if we maximize
profit and the lowest EMV if we minimize cost.
 24

Decision Strategies

Accept $2Bil 2

Texaco Accepts $5Bil 5

.17
Award High 10.3
.20
Counteroffer Texaco Refuses Accept $2Bil
Counteroffer Award Medium
$5Bil 5
.50 .50
Counteroffer $5 Bil,
Award Low 0 if Texaco counters
.30
Texaco Counters then refuse
$3Bil Award High 10.3
.33 .20
Refuse Texaco Counteroffer $5 Bil,
Counteroffer Award Medium 5 if Texaco counters
.50
then accept
Award Low
.30 0
Accept
$3Bil 3
 25

Decision Paths

Accept $2Bil 2

Texaco Accepts $5Bil 5

.17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco Counters .30
$3Bil Award High 10.3
.33 .20
Refuse Texaco
There are 9
Counteroffer Award Medium 5 such paths
.50
(one for each
Award Low
0 end branch of
.30
Accept the tree)
$3Bil 3
26

Optimal Strategy

Accept $2Bil 2

Texaco Accepts $5Bil 5

4.63 .17
Award High 10.3
.20
4.63 Texaco Refuses 4.56
Counteroffer
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco Counters .30
$3Bil Award High 10.3
.33 .20
Refuse Texaco 4.56
Counteroffer Award Medium 5
4.56 .50
Award Low
.30 0
Accept
$3Bil 3
27
Evaluating Decision Tree
Using Distributions – Risk Profiles
• Procedure

 For each decision strategy collapse the tree into its most
basic uncertainty representation
 Compare Risk Profiles (pmf) or Cumulative Risk Profiles
(CDF) for outcomes
 Cumulative Risk Profiles can be used to explore
dominance in decision strategies. Dominated strategies
can be removed from consideration
28
Dominance
Assuming Max is Optimal
1 The CDF of 2 has
Strategy 1 reached 1 before the
Strategy 2 CDF of 1 leaves 0.
Then 1
deterministically
dominates 2
0
Deterministic Dominance
1 The CDF of 2 lies
entirely to the left
Strategy 1
and above the CDF of
Strategy 2
1. Then 1
stochastically
dominates 2
0 When Min is
optimal dominance
Stochastic Dominance is reversed
29
Dominance

1
Strategy 1
Strategy 2

•Note
When cumulative risk profiles cross use
EMV to solve the problem
Deterministic Dominance
Stochastic Dominance
Solution by EMV
30
Risk Profiles
Texaco-Penzoil
Strategy: Accept $2Bil

Accept $2Bil 2

Risk Profile Cumulative Risk Profile

1
1
0.8
0.6
0.4 0.5
0.2
0 0
0 1 2 3 4 5 6 7 8 9 10 11 0 2 4 6 8 10 12
31
Risk Profiles
Texaco-Penzoil
Strategy: Counteroffer $5Bil, If Texaco Counters Refuse
Texaco Accepts $5Bil 5
.17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco Counters .30
$3Bil Award High 10.3
.33 .20
Refuse Texaco
Counteroffer Award Medium 5
.50
Award Low
.30 0

Combine sequential probability nodes

32
Risk Profiles
Texaco-Penzoil
Strategy: Counteroffer $5Bil, If Texaco Counters Refuse

5
.17
10.3
.10
Counteroffer
$5Bil 5
.25

0
.15
10.3
.066

5
.165

.099 0

Combine probabilities for the same value

33
Risk Profiles
Texaco-Penzoil
Strategy: Counteroffer $5Bil, If Texaco Counters Refuse

10.3
.10 + .066 = .166
Counteroffer
$5Bil 5
.17 +.25 + .165 = .585

0
.15 + .099 =.249

Risk Profile Cumulative Risk Profile

1
1
0.8
0.6
0.4 0.5
0.2
0 0
0 1 2 3 4 5 6 7 8 9 10 11 0 2 4 6 8 10 12
34
Risk Profiles
Texaco-Penzoil
Strategy: Counteroffer $5Bil, If Texaco Counters Accept
Texaco Accepts $5Bil 5
.17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50
Award Low 0
Texaco Counters .30
$3Bil
.33

Accept
$3Bil 3

Combine sequential probability nodes

35
Risk Profiles
Texaco-Penzoil
Strategy: Counteroffer $5Bil, If Texaco Counters Accept
5
.17
10.3
.10
Counteroffer
$5Bil 5
.25
.33 0
.15

Combine probabilities for the same value

36
Risk Profiles
Texaco-Penzoil
Strategy: Counteroffer $5Bil, If Texaco Counters Accept

10.3
.10
Counteroffer
$5Bil 5
.17 + .25 = .42

3
.33

0
.15

Risk Profile Cumulative Risk Profile

1
1
0.8
0.6
0.4 0.5
0.2
0 0
0 1 2 3 4 5 6 7 8 9 10 11 0 2 4 6 8 10 12
37
Risk Profiles
Texaco-Penzoil
Comparison of RIsk Profiles

1.2
1 Accept $2Bil
0.8
0.6 Counter $5Bil, If Counter
0.4 Refuse
0.2
Counter $5Bil, If Counter
0
Accept
0 1 2 3 4 5 6 7 8 9 10 11

Comparison of Cumulative Risk Profiles

1.5
Accept $2Bil
No
1
Counter $5Bil, If Dominance
0.5 Counter Refuse
Counter $5Bil, If
0
Counter Accept
0 5 10 15
38

Example: Deterministic Dominance

Accept Texaco Offer of $2Bil 2

Texaco Accepts
Counteroffer
5
.17
Award High 10.3
.20
Counteroffer Texaco Refuses
Counteroffer Award Medium
$5Bil 5
.50 .50 All
Award Low 2.5
payoffs
Texaco Counters .30 >2
$3Bil Award High 10.3
.33 .20
Refuse Texaco
Counteroffer Award Medium 5
.50
Award Low
.30 2.5
Accept
$3Bil 3
39

Example: Deterministic Dominance

Comparison of RIsk Profiles

1.2
1 Accept $2Bil
0.8
0.6 Counter $5Bil, If Counter
0.4 Refuse
0.2
Counter $5Bil, If Counter
0
Accept
0 2 3 3 4 5 6 7 8 9 10 11
Accept $2Bil is
Comparison of Cumulative Risk Profiles deterministically
dominated
1.2
1 Accept $2Bil
0.8
Counter $5Bil, If
0.6
Counter Refuse
0.4
Counter $5Bil, If
0.2 Counter Accept
0
0.0 5.0 10.0 15.0
 40

Example: Dominance
Suppose if it goes to court Leidtke must choose representation
Award High 10.5
.20
Firm 1 Award Medium 5.2
.50
Award Low 0
.30
Award High 10.3
.20
Firm 2 Award Medium 5
.50
Comparison of Cumulative Risk Profiles Award Low
.30 0
1.5

1 Firm 1
0.5 Firm 2 Firm 1 stochastically
0 dominates Firm 2
0 2 4 6 8 10 12