Professional Documents
Culture Documents
Making in Organizations
UNIT 1.3
DECISION MAKING UNDER
UNCERTAINTY
1
Where are we now?
Goodwin, P. & Wright, G. (2009). Decision Analysis for Management Judgment, 4th Edition. UK: John Wiley &
Sons. Chapter 6 (pp. 115-129) and Chapter 9. 3
Think about it
Each box contains 10 cards.
A dollar amount is printed on each card.
You are asked to choose a box to draw a card and win the
amount of $$ stated on the card.
Box 1: Box 2:
some “$1000” some “$600”
some “$500” some “$550”
some “$0” some “$350”
Box 1: Box 2:
some “$1000” some “$600”
some “$500” some “$550”
some “$0” some “$350”
Box 1: Box 2:
1 card: $1000 3 cards: $600
8 cards: $500 3 cards: $550
1 card: $0 4 cards: $350
Which box would you prefer? Why?
6
SOLVING DECISION TREES
USING THE EXPECTED MONETARY VALUE (EMV) CRITERION
• What is expected monetary value?
• Suppose you play the following game, you flip a coin:
If the coin lands heads, you win $100.
If the coin lands tails, you win nothing.
• Let X be our consequences for one flip
8
EMV: Business deal example
• Suppose you are considering a business deal that requires a $60M
investment, and the return will depend on economy conditions.
13
Think about it
Each box contains 10 cards.
A dollar amount is printed on each card.
You are asked to choose a box, and draw a card ONCE.
You will gain the amount of $$ stated on the card.
Box 1: Box 2:
1 card: $1000 3 cards: $600
8 cards: $500 3 cards: $550
1 card: $0 4 cards: $350
Which box would you prefer? Did you follow the EMV criterion?
14
Not the EMV criterion?
Consider the following two games:
Game 1 Game 2
Flip a coin: Flip a coin:
If it lands heads, win $30. If it lands heads, win $2000.
If it lands tails, lose $1. If it lands tails, lose $1900.
15
The expected utility (EU) criterion
Sometimes, people don’t choose the option that maximize EMV. Why?
• The EMV criterion is most suitable when the decision can be repeated a
large number of times so that a long run average result would be of
relevance.
• In some situations, however, decision makers only have one
opportunity to choose the best course of action. If things go wrong,
then there will be no chance of recovering losses. In such circumstances,
some people might prefer a less risky course of action.
• Thus, we need to consider expected utility:
EU(X) = U(x1)*p(x1) + U(x2)*p(x2) + … + U(xi)*p(xi) where xi is the various
outcomes of the uncertain event X
• EU takes into account our risk attitudes which can be reflected by our
utility functions.
16
UTILITY FUNCTIONS
17
Defining our utility functions using the method of bisection
0.6
0.4
0.2
U($0) = 0
0
$0 $200 $400 $600 $800 $1,000
18
Money ($)
• Then, consider a gamble:
“50% chance to get $1000, 50% chance to get $0.”
What is my certainty equivalent of the gamble?
• In other words, what is $x1 such that:
U($x1) = EU(gamble) = ½ U($1000) + ½ U($0) = ½ *1 + ½ *0 = 0.5
• For me, $x1 is $250.
Utility
1
U($1000) = 1
0.8
0.6
U($250) = 0.5
0.4
0.2
U($0) = 0
0
$0 $200 $400 $600 $800 $1,000
19
Money ($)
• Then, consider a gamble:
“50% chance to get $1000, 50% chance to get $250.”
What is my certainty equivalent of the gamble?
• In other words, what is $x2 such that:
U($x2) = EU(gamble) = ½ U($1000) + ½ U($250) = ½ *1 + ½ *0.5 = 0.75
• For me, $x2 is $600.
Utility
1
U($1000) = 1
0.8
U($600) = 0.75
0.6
U($250) = 0.5
0.4
0.2
U($0) = 0
0
$0 $200 $400 $600 $800 $1,000
20
Money ($)
• Then, consider a gamble:
“50% chance to get $250, 50% chance to get $0.”
What is my certainty equivalent of the gamble?
• In other words, what is $x3 such that:
U($x3) = EU(gamble) = ½ U($250) + ½ U($0) = ½ *0.5 + ½ *0 = 0.25
• For me, $x3 is $70.
Utility
1
U($1000) = 1
0.8
U($600) = 0.75
0.6
U($250) = 0.5
0.4
U($70) = 0.25
0.2
U($0) = 0
0
$0 $200 $400 $600 $800 $1,000
21
Money ($)
• Then, consider a gamble:
“50% chance to get $250, 50% chance to get $70.”
What is your certainty equivalent of the gamble?
• In other words, what is $x4 such that:
U($x4) = EU(gamble) = ½ U($250) + ½ U($70) = ½ *0.5 + ½ *0.25 = 0.375
• For me, $x4 is $125.
Utility
1
• Then, joint U($1000) = 1
the points 0.8
together U($600) = 0.75
0.6
and this is
U($250) = 0.5
MY utility 0.4 U($125) = 0.375
function.
U($70) = 0.25
0.2
U($0) = 0
0
$0 $200 $400 $600 $800 $1,000
22
Money ($)
Define YOUR utility function over the range $0 - $1000
Utility
$0/$x1 $x3 = 0.25
0.5
$0/x3 $x4 = 0.125 0.4
0.3
$x3/$x1 $x5 = 0.375 0.2
0.1
$x1/$x2 $x6 = 0.625 0
0 100 200 300 400 500 600 700 800 900 1000
23
Utility Functions and Risk Attitudes
Utility
Convex
Money
EU(gamble) = 50%*U(w-1000)+50%*U(w+1000)
EU(gamble) < U(w)
EU(gamble) < EU(doing nothing)
U(w+1000)
Unwillingness to accept a
“fair” gamble indicates
U(w) that you are risk averse.
EU(gamble)
w-1000 w w+1000
0 200 400 600 800 1000
Wealth ($)
25
Linear utility function implies risk neutrality
Utility
EU(gamble) = 50%*U(w-1000)+50%*U(w+1000)
EU(gamble) = U(w)
EU(gamble) = EU(doing nothing)
w-1000 w w+1000
0 200 400 600 800 1000
Wealth ($)
26
Convex utility function implies risk seeking
Utility
EU(gamble) = 50%*U(w-1000)+50%*U(w+1000)
EU(gamble) > U(w)
EU(gamble) > EU(doing nothing)
Willingness to accept a
“fair” gamble indicates
that you are risk seeking.
Hints:
Draw a utility function that reflects “risk averse”
On the x-axis, identify 3 points: w, w+100, and EMV(gamble)
On the y-axis, identify EU(gamble)
On the x-axis, identify CE(gamble)
What can you said about the EMV and CE?
28
Solving the PRS question:
29
SOLVING DECISION TREES
USING THE EXPECTED UTILITY (EU) CRITERION
30
Solving decision tree for EU: The R&D decision example
Stop development
U($0B)
Patent
Low demand (20%)
awarded U($3B)
(70%)
License technology
U($23B)
31
Step 1: For each of the consequence, define the utility of the consequence.
Utility function of the decision maker is given as below:
($43B, 1.0)
Utility
($23B, 0.80)
($21B, 0.78)
($3B, 0.25)
($0B, 0.15)
(-$2B, 0)
-2 8 18 28 38
Money ($B)
32
Steps 2 & 3: When you encounter a chance node, calculate the EU.
When you encounter a decision node, choose the branch with the highest EU.
Stop development
U($0B)=0.15
Patent EU =
EU = 0.729 Low demand (20%)
awarded U($3B)=0.25
0.56 (70%)
License technology
U($23B)=0.8
33
Maximizing EMV or EU?
• Scholars argue that the EMV criterion is also appropriate in one-off
decisions.
• Although a single decision may be unique, over time a decision-maker
may make a large number of decisions involving similar monetary
sums, so that returns should still be maximized by consistent
application of the EMV criterion.
• Moreover, large organizations may be able to sustain losses on
projects that represent only a small part of their operations.
34
DECISION INVOLVING MULTIPLE OBJECTIVES AND UNCERTAINTY
(PUTTING SMART AND DECISION TREES TOGETHER)
35
Retreat example:
• Our department is considering organizing a retreat.
• There are two alternatives: “Boat trip” and “Dinner at hotel”
• Evaluation criteria: “Cost” and “Level of fun”
• Uncertainty: Level of fun in boat trip depends on weather condition.
Cost: Fun:
Rainy (10%)
$50,000 Low
Sunny (70%)
$50,000 High
36
Step 2: For each evaluation attribute,
assign utility scores to reflect the Cost: Fun:
desirability of outcomes. Rainy (10%)
Low
$50,000
(1.0) (0.0)
Step 3: Assign swing weight to each attribute, then normalize the weights.
Hypothetical worst Hypothetical best Swing Normalized
alternative alternative Weight Weight
60/160 =
Cost $60,000 $50,000 60
0.375
100/160 =
Fun Low High 100
0.625 37
Step 4: Calculate the weighted utility at each branch,
then solve the decision tree using EU criterion.
EU = 0.9
Boat trip & ball Cloudy (20%) $50,000 Medium 0.813
game on beach (1.0) (0.7)
38
Step 5: For figures that you are not very sure of (e.g. utility of an intermediate
outcome or a swing weight), conduct a sensitivity analysis.
1.00
0.90
0.80
EU of alternatives
0.70
0.60
0.50
Boat trip
0.40
0.30 Hotel dinner
0.20
0.10
0.00
0 10 20 30 40 50 60 70 80 90 100
Weight on Cost
40
ASSESSING PROBABILITY
41
Sometimes, you might be able to obtain more information on your
specific case.
Examples
• Mrs. Chen is 35 and is pregnant for the first time. Doctors: 1% chance
that a woman of her age will have a baby with Down syndrome.
Mrs. Chen can have a Triple Screen test – this test can detect if her
baby has Down syndrome.
• You are planning for a boat trip on 26th July. Historical data shows that
10% of days in July are rainy.
You can consult a weatherman – this expert can tell you if 26th July this
year will be a rainy day.
How should you revise the probability judgment in the light of new
information?
42
Revising probability judgments in the light of new information
• If Mrs. Chen had a 100%-accurate Triple Screen test done, and the test
indicated that the baby doesn’t have Down syndrome:
• Revised probability judgment: “Baby has the disease (0%)”
• Similarly, if the 100%-accurate test indicated that the baby does have
Down syndrome:
• Revised probability judgment: “Baby has the disease (100%)”
43
In fact, the Triple Screen test is NOT 100% accurate.
• If the baby has the disease, there is a 0.80 probability that the test will
correctly show a positive result.
• If the baby doesn’t have the disease, there is a 0.90 probability that the test
will correctly show a negative result.
Let’s combine the probability that characterizes the test performance and the
base rate.
In reality, the baby has
Disease (1%) No Disease (99%)
Disease Correct False Alarm
The test
indicates:
No Disease Missed Detection Correct
Total = 100% 44
In reality, the baby has
Disease (1%) No Disease (99%)
• If now this imperfect test indicates that the baby has Down syndrome, what
should be your estimate of the probability that the baby has the disease?
• That is, given that the test indicated “disease,” what is the probability that the
baby has the disease?
• If now this imperfect test indicates that the baby has Down syndrome, what
should be your estimate of the probability that the baby has the disease?
• That is, given that the test indicated “disease,” what is the probability that the
baby has the disease?
P(baby has the disease| test indicated “disease”)
= P(baby has the disease AND test indicated “disease”) /
P(test indicated “disease”)
= 0.8% / (0.8% + 9.9%)
= 7.48%
• Revised probability judgment:
Baby has the disease (7.48%)
• If now this imperfect test indicates that the baby does not have Down syndrome,
what should be your estimate of the probability that the baby has the disease?
1) 0.2%
2) 0.22%
3) 7.5%
4) 10.7%
47
VALUE OF INFORMATION
48
Value of Information – Planting Example
• One year ago a farmer suffered serious losses when a virus affected the crop.
• Since then, the farmer has taken measures to eliminate the virus from the soil.
On the basis of scientific evidence, it is estimated that there is a 70% chance that
the elimination was successful.
• Now the farmer must make a decision about what to do next season:
Virus Conditions
Plant options Virus was eliminated Virus was NOT
(70%) eliminated (30%)
Grain $90M - $20M
Soybeans $30M
Soybeans
$30M
Soybeans
$30M
Virus was eliminated (70%)
$90M
Grain
Virus was NOT eliminated (30%)
- $20M
EMV = $57M
Grain
Test indicates no virus (70%) $90M
Test Soybeans
Soil $30M
Grain
EMV= $72M Test indicates virus (30%) - $20M
Soybeans
$30M
Base rate
51
• How much would it be worth paying to have the test done?
• If no further information is obtained, the EMV of the decision is $57M.
• If perfect information is obtained, the EMV of the decision is $72M.
• Perform the test if ($72M - cost of test) > $57M
i.e. if cost of test < $15M
52
• Of course not all information is perfect (not 100% accurate).
• Suppose the information provided by the soil test is imperfect, and data showed
that:
• If the virus is still present, the test has a 90% chance of correctly detecting it.
• If the virus has been eliminated, the test has a 20% chance of incorrectly
indicating that the virus is still present.
• Let’s combine the probabilities that characterize the performance of the soil test,
and the base rate of successful virus elimination:
In reality, the virus is:
Present (30%) Eliminated (70%)
• Now, we want to know the EMV of the decision, with the option of
gathering this imperfect information.
53
Soybeans
$30M
54
In reality, the virus is:
Present (30%) Eliminated (70%)
• What is the probability that the test indicates that the virus is present?
P(test indicates that virus is present) = ____________________
• Given that the test indicates that virus is present, what is the probability that the virus is present?
P(virus is present | test indicates that virus is present) = ____________________
• Given that the test indicates that virus is present, what is the probability that the virus has been
eliminated?
P(virus has been eliminated | test indicates that virus is present) = ____________________
• What is the probability that the test indicates that the virus is eliminated?
P(test indicates that virus is eliminated) = ____________________
• Given that the test indicates that virus has been eliminated, what is the probability that the virus is
present?
P(virus is present | test indicates that virus has been eliminated) = ____________________
• Given that the test indicates that virus has been eliminated. What is the probability that the virus has
been eliminated?
P(virus has been eliminated | test indicates that virus has been eliminated) = _________________
Soybeans
$30M
57
Time for REFLECTION
58
INTENDED LEARNING OUTCOMES FOR UNIT 1.3
59