Chapter 02

Decision Analysis
IIMT3636
Faculty of Business and Economics
The University of Hong Kong
Instructor: Dr. Yipu DENG
 2

Recap of Last Lecture

• Poisson distribution à number of arrivals
▫ Relationship with binomial distribution
• Exponential distribution à time
▫ Relationship with Poisson distribution
• Uniform distribution
• Normal distribution
▫ Standard normal distribution and Z table
• Steps in decision making
▫ Define the problem
▫ List all possible alternatives, states of nature, and corresponding
payoff for each combination
▫ Select a decision theory model and make the decision
• Decision making under certainty
• Decision making under risk
▫ Maximize EMV & Minimize EOL
▫ EVPI = EVwPI – max EMV or EVPI = min EOL
 11

In-class Exercise: Café du Donut

• The Café buys donuts each day for $40 per carton of 20 dozen
donuts. Any cartons not sold are thrown away at the end of
the day. If a carton is sold, the total revenue is $60.

DAILY DEMAND PROBABILITY CUMULATIVE

(CARTONS) PROBABILITY
• The original plan 4 0.05 0.05
is to order 6 5 0.15 0.2
cartons per day. 6 0.15 0.35

Should the Café 7 0.20 0.55

increase the 8 0.25 0.8

9 0.10 0.9
order size to 7?
10 0.10 1.0
Total 1.00
 12

In-class Exercise: Café du Donut

• Monetary Payoff (Profit) Table
• Cost = $40, revenue = $60
D=4 D=5 D=6 D=7 D=8 D=9 D = 10 EMV
Q=6 0 60 120 120 120 120 120 105
Q=7 -40 20 80 140 140 140 140 104
Prob. 0.05 0.15 0.15 0.20 0.25 0.10 0.10

• Should we reduce the order size from 6 to 5? What is the EMV of Q=5?

• If we can only choose between 6 and 7, what is the EVPI?

 13

Decision Making under Risk

• If the estimation of probability is changed, how will John change his
decision?
• A sensitivity analysis is needed! It investigates how our decision
might change given a change in the problem data.
STATE OF NATURE

FAVORABLE UNFAVORABLE EMV

ALTERNATIVE MARKET (profit in $) MARKET (profit in $) (in $1,000)

Construct a large plant 200,000 –180,000 380p – 180

Construct a small plant 100,000 –20,000 120p – 20
Do nothing 0 0 0
Probability p 1–p
 15

Decision Making under Risk

EMV

$300,000

$200,000 EMV (large plant)

Point 2 380,000p-180,000
0.45
$100,000 EMV (small plant)
Point 1 120,000p-20,000

0 EMV (do nothing)

0
.167 .615 1
–$100,000 Values of p

–$200,000
 16

Results of Sensitivity Analysis

BEST ALTERNATIVE RANGE OF P VALUES

Construct a large plant >0.615

Construct a small plant 0.167 - 0.615

Do nothing <0.167
 17

Sensitivity Analysis in Excel

• Step 1: setup the payoff table (data and formulas)
• Step 2: setup the data table
▫ Set a list of values in the first column for the parameter to be
explored, except for the first row.
▫ Reference the output values in the first row, starting from the
second column.
• Step 3: generate the one-way data table
▫ Select the entire data table.
▫ Click: DATA -> What-If Analysis -> Data Table.
▫ Set the Column input cell to the parameter cell in the original
model.
 18

In-Class Exercise
• Assume directly measuring monetary payoff is infeasible
• Opportunity Loss Table

State I State II
Option A 5 1
Option B 0 3
Option C 6 0
Probability 0.3 0.7

• Which is the option that maximizes EMV?

 19

In-Class Exercise
• Given Opportunity Loss Table
State I State II
Option A 5 1
Option B 0 3
Option C 6 0
Probability 0.3 0.7

• How to restore the payoff table?

State I State II
Option A 1 3
EVwPI = 66(0.3) + 4(0.7)
Option B 1
= 4.6
Option C 0 4

• What is the EVwPI?

 3

In-class Exercises
• The Monty Hall Problem Revisited
▫ Suppose you're on a game show, and you're given the choice of three doors:
Behind one door is a car; behind the others, goats. You pick a door, say No. 1, and
the host, who knows what's behind the doors, opens another door, say No. 3,
which has a goat. He then says to you, "Do you want to pick door No. 2?" Is it to
your advantage to switch your choice?
• Timeline
• Payoff table (payoff = 1 if you get a car; 0 otherwise)
Given door 1 is Car behind Car behind Car behind Expected
chosen door 1 door 2 door 3 Value
Switch 0 1 1 2/3

Do not switch 1 0 0 1/3

Probability 1/3 1/3 1/3

 4

Decision Making under Uncertainty

What if we do not know the probabilities for all states of
nature?

• Maximax (optimistic)
• Maximin (pessimistic)
Payoff table
• Criterion of realism (Hurwicz)
• Equally likely (Laplace)
• Minimax regret Opportunity loss table
 5

Decision Making under Uncertainty

• Maximax (optimistic)
• Maximize the max payoff for the alternative
STATE OF NATURE

FAVORABLE UNFAVORABL MAXIMUM IN

ALTERNATIVE MARKET ($) E MARKET ($) A ROW ($)
Construct a large plant 200,000 –180,000 200,000

Construct a small plant 100,000 –20,000 100,000

Do nothing 0 0 0
 6

Decision Making under Uncertainty

• Maximin (pessimistic)
• Maximize the min payoff for the alternative
STATE OF NATURE

FAVORABLE UNFAVORABL MINIMUM IN

ALTERNATIVE MARKET ($) E MARKET ($) A ROW ($)
Construct a large plant 200,000 –180,000 –180,000

Construct a small plant 100,000 –20,000 –20,000

Do nothing 0 0 0
 7

Decision Making under Uncertainty

• Criterion of realism (Hurwicz)
▫ Select a coefficient of realism α, with 0 ≤ α ≤ 1.
▫ Weighted average = α(best in row) + (1 – α)(worst in row).
STATE OF NATURE WEIGHTED
AVERAGE
FAVORABLE UNFAVORABL WITH α = 0.8
ALTERNATIVE MARKET ($) E MARKET ($) ($)
Construct a large plant 200,000 –180,000 124,000

Construct a small plant 100,000 –20,000 76,000

Do nothing 0 0 0

• Consider two extreme payoffs for each alternative

 8

Decision Making under Uncertainty

• Equally likely (Laplace)
• Maximize the average payoff for each alternative
STATE OF NATURE
ROW
FAVORABLE UNFAVORABL AVERAGE
ALTERNATIVE MARKET ($) E MARKET ($) ($)
Construct a large plant 200,000 –180,000 10,000

Construct a small plant 100,000 –20,000 40,000

Do nothing 0 0 0
 9

Decision Making under Uncertainty

• Minimax regret à opportunity loss table
• Minimize the max opportunity loss for each alternative
STATE OF NATURE

FAVORABLE UNFAVORABL MAXIMUM IN

ALTERNATIVE MARKET ($) E MARKET ($) A ROW ($)
Construct a large plant 0 180,000 180,000

Construct a small plant 100,000 20,000 100,000

Do nothing 200,000 0 200,000

 10

In-Class Exercise
• Suppose you’re a salesperson and you can sell 3 products to
consumers. Consider the following payoff table.
Scenario 1 Scenario 2 Scenario 3
Optimistic Pessimistic Hurwicz Laplace Max Regret
Option
Option A A 2.5 0 -1.5 2.5
4a – 1.5 1/3 -1.5 2.5

Option B
Option B 1
1 0
0
a 2/3
1 2.5
Option C 1 -1 2a – 1 1/3 2
Option C -1 1 1
• If you want to sell option A, which criterion should you use?
• If you want to sell option B, which criterion should you use?
• If you want to sell option C, which criterion should you use?
 11

Decision Tree
• Any problem that can be presented in a decision table can also be
graphically illustrated in a decision tree.
• Advantages of decision tree:
▫ Incorporate sequential decision making
▫ Incorporate different states of nature and corresponding probabilities for
different alternatives
• While drawing the decision tree, we begin at the left and move to the
right à the tree presents decisions and outcomes in a sequential
order.
• Any decision tree has:
▫ Decision node (square), from which one of several alternatives may be
chosen
▫ State-of-nature node (circle), out of which one state of nature will occur
▫ Lines from a decision node represent alternatives, lines from a state-of-
nature node represent states of nature
 12

John’s Decision Tree

A State-of-Nature Node Payoffs
Favorable Market (0.45)
$200,000
A Decision Node EMV1 = -9k 1
Unfavorable Market (0.55)
cta t –$180,000
stru lan
o n eP
C arg
L Favorable Market (0.45)
EMV2 = 34k $100,000
Construct a
2
Small Plant Unfavorable Market (0.55)
–$20,000
Do
No
th
in
g

$0
 13

Decision Tree with Sample Information

• Before deciding on building a new plant, John has the
option of hiring ABC, Inc. to conduct a market survey, at
a cost of $10,000. The survey can partially show the
future state of nature (positive vs. negative).
• Conditional on a positive survey result, the market will
be favorable with probability 0.78 and unfavorable with
probability 0.22. Conditional on a negative survey result,
the market will be favorable with probability 0.27 and
unfavorable with probability 0.73.
• Suppose P(Positive result) = 0.353;
P(Negative result) = 0.647
 14

Decision Tree with Sample Information

First Decision Second Decision
Conditional probability Payoffs
Point Point Favorable Market (0.78)
$190,000
nt 2 Unfavorable Market (0.22)
ge Pla –$190,000
Lar Favorable Market (0.78)
$90,000
) Small
3 53 Plant 3 Unfavorable Market (0.22)
–$30,000
.
y (0 ts
e l
urv esu tive No Plant
–$10,000
S R osi
Su
1 rv P Favorable Market (0.27)
e $190,000
Re y (0
ey

nt 4 Unfavorable Market (0.73)

rv

Ne su .6 Pla –$190,000
ga lts 47) ge
Su

tiv Lar Favorable Market (0.27)

t

Small $90,000
ke

e 5 Unfavorable Market (0.73)

ar

Plant –$30,000
t M
uc

No Plant
nd

–$10,000
Co

Do Favorable Market (0.45)

N ot $200,000
C on ant 6 Unfavorable Market (0.55)
duc
tS g e Pl –$180,000
urv Lar Favorable Market (0.45)
$100,000
ey Small
Plant 7 Unfavorable Market (0.55)
–$20,000
34k No Plant
$0
 15

Decision Tree with Sample Information

• Given positive survey results,

EMV(node 2) = EMV(large plant | positive survey)

= (0.78)($190,000) + (0.22)(–$190,000) = $106,400
EMV(node 3) = EMV(small plant | positive survey)
= (0.78)($90,000) + (0.22)(–$30,000) = $63,600
EMV for no plant = –$10,000

• Given negative survey results,

EMV(node 4) = EMV(large plant | negative survey)

= (0.27)($190,000) + (0.73)(–$190,000) = –$87,400
EMV(node 5) = EMV(small plant | negative survey)
= (0.27)($90,000) + (0.73)(–$30,000) = $2,400
EMV for no plant = –$10,000
 16

Decision Tree with Sample Information

Second Decision Payoffs
Point
$106,400 Favorable Market (0.78)
$190,000
t 2 Unfavorable Market (0.22)
lan
$106,400

g e P –$190,000
Lar $63,600 Favorable Market (0.78)
$90,000
) Small
3 53 Plant 3 Unfavorable Market (0.22)
. –$30,000
y (0 ts
EMV1 = 39.1k urve esulitive No Plant
–$10,000
S R os
Su P
1 rv –$87,400 Favorable Market (0.27)
ey $190,000
R (0 t 4 Unfavorable Market (0.73)
Ne esu .6 P lan –$190,000
ga lts 47) rg e
$2,400

La $2,400 Favorable Market (0.27)

tiv Small $90,000
e 5 Unfavorable Market (0.73)
Plant –$30,000
No Plant –$10,000
 17

Decision Tree with Sample Information

First Decision Second Decision Payoffs
Point Point Favorable Market (0.78)
$190,000
nt 2 Unfavorable Market (0.22)
ge Pla –$190,000
Lar Favorable Market (0.78)
$90,000
) Small
3 53 Plant 3 Unfavorable Market (0.22)
–$30,000
.
y (0 ts
e l
u rv esu tive No Plant
–$10,000
S R osi
Su
1 rv P Favorable Market (0.27)
e $190,000
Re y (0
ey

39.1k nt 4 Unfavorable Market (0.73)

rv

Ne su .6 Pla –$190,000
ga lts 47) ge
Su

tiv Lar Favorable Market (0.27)

t

Small $90,000
ke

e 5 Unfavorable Market (0.73)

ar

Plant –$30,000
t M
uc

No Plant
nd

–$10,000
Co

Do Favorable Market (0.45)

N ot $200,000
C on ant 6 Unfavorable Market (0.55)
duc
tS g e Pl –$180,000
34k urv Lar Favorable Market (0.45)
$100,000
ey Small
Plant 7 Unfavorable Market (0.55)
–$20,000
No Plant
$0
 18

Decision Tree with Sample Information

• Expected Value of Sample Information (EVSI)
à the increase in EV resulting from the sample information
= EV with SI – EV without SI
= EV of conducting survey + cost of survey - EV without SI
= $39,112 + $10,000 – $34,000 = $15,112

• Efficiency of sample information

!"#$
= ×100%
!"%$
&',&&)
= ×100% = 27%
'*,+++
 19

Sensitivity Analysis
How sensitive is our decision to the probability of positive survey results?
p)
$106,400
(
v ey ts
r l e
Su esu itiv
R os EMV (Conduct Survey) = 106,400*p+2,400*(1-p)
S P
1 ur
ve = 104,000*p+2,400
ey

R y
Ne esu (1-p
rv

EMV (Not Conduct Survey) = 34,000

Su

ga lts )
tiv
t
ke

e $2,400
ar
Mt
uc
nd
Co

Indifference point: 104,000*p+2,400 = 34,000

p = 0.30
Do
N ot
C on
duc
tS urv
ey
$34,000

If p > 0.30 à conduct the market survey;

If p < 0.30 à not conduct the market survey
 20

How Probability Values Are Estimated?

First Decision Second Decision Payoffs
Point Point Favorable Market (0.78)
$190,000
nt 2 Unfavorable Market (0.22)
ge Pla –$190,000
Lar Favorable Market (0.78)
$90,000
) Small
3 53 Plant 3 Unfavorable Market (0.22)
–$30,000
.
y (0 ts
e l
urv esu tive No Plant
–$10,000
S R osi
Su
1 rv P Favorable Market (0.27)
e $190,000
Re y (0
ey

nt 4 Unfavorable Market (0.73)

rv

Ne su .6 Pla –$190,000
ga lts 47) ge
Su

tiv Lar Favorable Market (0.27)

t

Small $90,000
ke

e 5 Unfavorable Market (0.73)

ar

Plant –$30,000
t M
uc

No Plant
nd

–$10,000
Co

Do Favorable Market (0.45)

N ot $200,000
C on ant 6 Unfavorable Market (0.55)
duc
tS g e Pl –$180,000
urv Lar Favorable Market (0.45)
$100,000
ey Small
Plant 7 Unfavorable Market (0.55)
–$20,000
No Plant
$0
 21

Use of Historical Data

• Suppose we do not know P(Pos), P(Neg), P(Fav|Pos), or
P(Unf|Neg). Instead, we have a prior belief that P(Fav) =
0.8, and historical data from ABC, Inc.:
Positive Negative Total
Favorable 35 20 55
Unfavorable 20 25 45
Total 55 45 100

• Then P(Pos)=? P(Neg)=? P(Fav|Pos)=? P(Unf|Neg)=?

• P(Pos) = 55/100?
 22

Use of Historical Data

• How to measure the accuracy of the survey?
• Think about the HIV example:
▫ P(positive test result|infected) = 90% & P(negative test result|not
infected) = 95%.
▫ Note that P(infected|positive test result) heavily depends on the
prior belief (prevalence rate).

• Therefore, we need P(Pos|Fav) and P(Neg|Unf).

Positive Negative Total
Favorable 35 20 55
Unfavorable 20 25 45
Total 55 45
 23

Use of Historical Data

P(Pos|Fav) = 35/55
• For the current market: P(Neg|Unf) = 25/45
Joint Prob. Positive Negative Marginal
Favorable 0.8*35/55 0.8*20/55 0.8
Unfavorable 0.2*20/45 0.2*25/45 0.2
Marginal 28/55+4/45 16/55+1/9

• P(Fav|Pos) = 28/55/(28/55+4/45) = 252/296 = 0.8514

• P(Unf|Neg) = 5/45/(16/55+5/45) = 0.2764
• The prior belief will be left unused if we calculate
P(Fav|Pos) directly from the historical data.
 24

Utility Theory
• People do not always make their decisions based on EMVs.
▫ E.g., a person buys a lottery ticket even though the EMV is negative.

• Why? The MV is not always a true indicator of the overall

value of the outcome of the alternative à utility

• Economists believe rational people make decisions that

maximize the expected utility.
▫ When some alternatives are associated with extremely large payoffs
or extremely large losses, the MV may not be a good indicator of
utility.

• Here, utility is a mental worth/happiness a person associates

with a (monetary) outcome. It only has a relative value.
▫ $1 vs. $2
 25

Construct a Utility Curve

• Assign utility values (e.g., 0~1) to each outcome à utility
assessment is completely subjective.
• A utility curve revealing Alex’s preference for money

Risk avoider
 26

Utility Theory: Risk Avoider

• A decision maker who gets less utility from a greater risk
and tends to avoid situations where high losses might occur.
$2,000 Utility curve for a risk avoider

Utility
(0.5) $5,000

$2,500

(0.5)
$0 $0 $2,000 $5,000

• As the monetary value increases, the utility increases at a

slower rate.
 27

Utility Theory: Risk Seeker

• A decision maker who gets more utility from a greater risk
and higher potential payoff.
$3,000 Utility curve for a risk seeker

Utility
(0.5) $5,000

$2,500

(0.5)
$0 $0 $3,000 $5,000

• As the monetary value increases, the utility increases at an

increasing rate.