MS&E 252 Handout #28

Decision Analysis I December 4
th
. 2004

Page 1 oI 27 HW#7 Solutions
Homework Assignment #7- Solutions

0
2
4
6
8
10
12
14
16
18
-INF - 0 0.5 - 1 1.5 - 2 2.5 - 3 3.5 - 4 4.5 - 5 5.5 - 6 6.5 - 7 7.5 - 8 8.5 - 9 9.5 - 10
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
Number of Grades within Range Cumulative


























Score on HW Question (./1)
-2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00
Question 1
10% 50%90%
Question 2
10% 50%90%
Question 3
10% 50% 90%
Question 4
10% 50% 90%
Question 5
10% 50%90%
Question 6
10% 50% 90%
Question 7
10% 50%90%
Question 8
10% 50% 90%
Question 9
10% 50% 90%
Question 10
10% 50%90%
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 2 oI 27 HW#7 Solutions
Distinctions
These distinctions were prepared by the teaching team and reIlect our best belieI oI the
meanings oI these terms.
A decision diagram shows the structure oI a decision situation: the decisions.
uncertainties and values involved as well as the relations between them. We
represent these relations with arrows. An arrow Irom node A to node B shows that
we wish to condition B on A. This leads to several diIIerent 'types¨ oI arrows.
depending on the nodes which they connect.

Relevance arrows. indicating the conditioning oI one uncertainty on another.
show the possibility oI relevance between the two uncertainties.

Informational arrows. indicating the conditioning oI a decision on an
uncertainty. show that we know the outcome oI the uncertainty beIore we make
the decision.

Functional arrows indicate the conditioning oI a deterministic node on other
nodes. The deterministic node is a Iunction oI the nodes pointing to it.

Influence arrows. indicating the conditioning oI an uncertainty on a decision.
show that the probabilities we assign to the uncertainty may change depending on
the alternative we choose.

Probabilistic questions

1) Solution: a
The certain equivalent oI the bet decision without the machine is $2.00.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 3 oI 27 HW#7 Solutions
0.5
Smarty Jones
14
0.3
Bet Smarty Jones Lion Heart
2 -10
0.2
Imperialism
-10
0.5
Smarty Jones
-10
0.3
Bet Lion Heart Lion Heart
1
2 2 30
0.2
Imperialism
-10
0.5
Smarty Jones
-10
0.3
Bet Imperialism Lion Heart
2 -10
0.2
Imperialism
50


Flip the tree Ior the machine. in order to get the posterior probabilities oI the
machine`s results.


Prior Cond'l Prob Joint Prob Pre-Posterior Posterior Joint Prob
Smar 0.5 "Smar" 0.8 0.4 "Smar" 0.45 Smar 0.889 0.4
"Lion" 0.1 0.05 Lion 0.067 0.03
"Impe" 0.1 0.05 Impe 0.044 0.02
Lion 0.3 "Smar" 0.1 0.03 "Lion" 0.31 Smar 0.161 0.05
"Lion" 0.8 0.24 Lion 0.774 0.24
"Impe" 0.1 0.03 Impe 0.065 0.02
Impe 0.2 "Smar" 0.1 0.02 "Impe" 0.24 Smar 0.208 0.05
"Lion" 0.1 0.02 Lion 0.125 0.03
"Impe" 0.8 0.16 Impe 0.667 0.16


The certain equivalent oI the bet decision with the machine is $8.80.
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 4 oI 27 HW#7 Solutions
0.889
Smarty Jones
4
0.067
Bet Smarty Jones Lion Heart
1.333333 -20
0.044
Imperialism
-20
0.889
Smarty Jones
-20
0.45 0.067
"Smarty Jones" Bet Lion Heart Lion Heart
1
1.333333 -17.3333 20
0.044
Imperialism
-20
0.889
Smarty Jones
-20
0.067
Bet Imperialism Lion Heart
-17.3333 -20
0.044
Imperialism
40
0.161
Smarty Jones
4
0.774
Bet Smarty Jones Lion Heart
-16.129 -20
0.065
Imperialism
-20
0.161
Smarty Jones
-20
0.31 0.774
"Lion Heart" Bet Lion Heart Lion Heart
2
8.8 10.96774 10.96774 20
0.065
Imperialism
-20
0.161
Smarty Jones
-20
0.774
Bet Imperialism Lion Heart
-16.129 -20
0.065
Imperialism
40
0.208
Smarty Jones
4
0.125
Bet Smarty Jones Lion Heart
-15 -20
0.667
Imperialism
-20
0.208
Smarty Jones
-20
0.24 0.125
"Imperialism" Bet Lion Heart Lion Heart
3
20 -15 20
0.667
Imperialism
-20
0.208
Smarty Jones
-20
0.125
Bet Imperialism Lion Heart
20 -20
0.667
Imperialism
40

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 5 oI 27 HW#7 Solutions
2) Solution: d
Assume a u-curve oI u(x) ÷ 1 3
-x/10

The certain equivalent oI the bet decision without the machine is -$4.32.

0.5
Smarty Jones
$ u-Value
0.785202 14 0.785202
0.3
Bet Smarty Jones Lion Heart
-0.6074 -2 -10 -2
0.2
Imperialism
-2 -10 -2
0.5
Smarty Jones
-2 -10 -2
0.3
Bet Lion Heart Lion Heart
1
-0.6074 -1.11111 0.962963 30 0.962963
-4.32015
0.2
Imperialism
-2 -10 -2
0.5
Smarty Jones
-2 -10 -2
0.3
Bet Imperialism Lion Heart
-1.40082 -2 -10 -2
0.2
Imperialism
0.995885 50 0.995885



The certain equivalent oI the bet decision with the machine is -$6.69. ThereIore. iI
you were to place a bet. buying the use oI the machine is not a good decision.
However. since the CE oI the bet decision without the machine is negative. you are
better oII not placing a bet at all.


MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 6 oI 27 HW#7 Solutions
0.889
Smarty Jones
$ u-Value
0.355606 4 0.355606
0.067
Bet Smarty Jones Lion Heart
-0.57279 -8 -20 -8
0.044
Imperialism
-8 -20 -8
0.889
Smarty Jones
-8 -20 -8
0.45 0.067
"Smarty Jones" Bet Lion Heart Lion Heart
1
-0.57279 -7.40741 0.888889 20 0.888889
0.044
Imperialism
-8 -20 -8
0.889
Smarty Jones
-8 -20 -8
0.067
Bet Imperialism Lion Heart
-7.60055 -8 -20 -8
0.044
Imperialism
0.987654 40 0.987654
0.161
Smarty Jones
0.355606 4 0.355606
0.774
Bet Smarty Jones Lion Heart
-6.65232 -8 -20 -8
0.065
Imperialism
-8 -20 -8
0.161
Smarty Jones
-8 -20 -8
0.31 0.774
"Lion Heart" Bet Lion Heart Lion Heart
2
-1.0864 -1.11828 -1.11828 0.888889 20 0.888889
-6.69426
0.065
Imperialism
-8 -20 -8
0.161
Smarty Jones
-8 -20 -8
0.774
Bet Imperialism Lion Heart
-7.42015 -8 -20 -8
0.065
Imperialism
0.987654 40 0.987654
0.208
Smarty Jones
0.355606 4 0.355606
0.125
Bet Smarty Jones Lion Heart
-6.25925 -8 -20 -8
0.667
Imperialism
-8 -20 -8
0.208
Smarty Jones
-8 -20 -8
0.24 0.125
"Imperialism" Bet Lion Heart Lion Heart
3
-2.00823 -6.88889 0.888889 20 0.888889
0.667
Imperialism
-8 -20 -8
0.208
Smarty Jones
-8 -20 -8
0.125
Bet Imperialism Lion Heart
-2.00823 -8 -20 -8
0.667
Imperialism
0.987654 40 0.987654

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 7 oI 27 HW#7 Solutions
p CE
0 6.6
0.1 6.54
0.2 6.48
0.3 6.42
0.4 6.36
0.5 6.3
0.6 6.24
0.7 6.18
0.8 6.12
0.9 6.06
1 6
Sensitivity to p for Risk Neutral Case
5.9
6
6.1
6.2
6.3
6.4
6.5
6.6
6.7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
p
C
E
p CE
0 -7.18709
0.1 -7.82247
0.2 -8.36703
0.3 -8.79941
0.4 -9.08972
0.5 -9.19426
0.6 -9.04544
0.7 -8.53039
0.8 -7.43855
0.9 -5.30113
1 -0.64981
Sensitivity Analysis to p for Risk Averse Case
-10
-8
-6
-4
-2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
p
C
E
3) Solution: b
The certain equivalent value oI using the machine with any particular value oI p can
be derived using similar methods as in question 1. using 5¹15p as the cost oI using
the machine. Note that the 'Ilipped¨ probabilities will also change with p. The
Iollowing table lists the certain equivalents associated with betting when the machine
is used with various values oI p:












From the graph oI the data at right. we see that the certain equivalent value oI the bet
decision is a linear Iunction oI the value oI p. This value is maximized when p÷0.

4) Solution: c
Assume a u-curve oI u(x) ÷ 1 3
-x/10

Again. we can plot certain equivalent values as a Iunction oI p. computing these
values Ior each value oI p as we did in the solution to question 2. using the price oI
5¹15p and revising the value oI the 'Ilipped¨ probabilities according to p.











MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 8 oI 27 HW#7 Solutions

The certain equivalents are always negative hence iI one bought the use oI the
machine at any p. it would always be better not to bet at all. However. in the case
where p÷1. we can see that the certain equivalent is an improvement to the situation
where the machine is not used. which has a certain equivalent oI -$4.32. as computed
in question 2. Hence. iI Cory were to place a bet. it is to his advantage to use the
machine at a value oI p÷1. However. he is still better oII not placing a bet at all.

5) Solution: b or c
Kim`s u-value Ior the indoors alternative. which she would choose without the rain
detector. is 0.63 (see page 3-7 oI the course reader). The u-value oI the deal with the
use oI the weather detector. Ior any given value oI the accuracy a. can be computed
Irom the Iollowing tree. (We compute the u-values according to the Iormula on page
3-13 oI the course reader.)
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 9 oI 27 HW#7 Solutions
0.79
S
$85.00
Outdoor 100 $85.00
0.92
0 $49.80 0.21
0.67 R
-$15.00
0 -$15.00
-0.31
0.79
S
0.43 $75.00
"S" Porch 90 $75.00
2 0.86
0 $53.70 0 $53.70 0.209302326
0.70 0.70 R
$5.00
20 $5.00
0.09
0.79
S
$25.00
Indoor 40 $25.00
0.39
0 $26.98 0.209302326
0.42 R
$35.00
50 $35.00
0.51
$41.74 0.11
0.59 S
$85.00
100 $85.00
0.92
0 -$9.08 0.89
-0.18 R
-$15.00
0 -$15.00
-0.31
0.11
S
0.57 $75.00
"R" Decision 13 90 $75.00
3 0.86
0 $33.88 0 $9.87 0.89
0.50 0.17 R
$5.00
20 $5.00
0.09
0.11
S
$25.00
Decision 14 40 $25.00
0.39
0 $33.88 0.89
0.50 R
$35.00
50 $35.00
0.51

Figure 1: Tree for a detector with accuracy 85º and cost of detector of $15. u values are in red.
Plugging the values Ior the detector accuracy a and the corresponding detector cost allows
the computation oI the u-value oI the party deal with detector inIormation.
Detector u-value
A .59
B .63
C .66
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 10 oI 27 HW#7 Solutions
Detector C only is clearly economic.
Please not that B is Iree but is useless. since whatever its indication the optimal decision
is the same (indoor). Using it does not increase the value oI the deal. so it is not
economic. However. since it is Iree. some people might have considered that it is
economic because it is Iree (we are not loosing anything by using it). We accepted this
reasoning as well. that is why the solution oI this question is b) or c).
6) Solution: c
Statement I is true.
Statement II is Ialse. Consider a deal about which both Bill and John believe there is a
50° chance oI getting either $0 or $100 dollars. Bill values it $50. John is not risk-
neutral since his u-curve is not linear. and he cannot Iollow the delta property (his u-curve
is not oI an exponential Iorm). However. his certain equivalent Ior the deal is $50 as well.

Statement III is also true: iI you are sure about how an uncertainty will come out. the
Value oI Clairvoyance on this distinction is iust zero. ThereIore. iI you are sure as to
which answer is the correct answer in a probabilistic question. your PIBP Ior the
Clairvoyant`s services is zero.

7) Solution: d

Statement I is true The chart shows that Ior ¦i
0
' &} ÷ 0.7. B is the alternative which
yields the highest u-value;
Statement II is true The equation that gives the u-value oI alternative D as a Iunction oI
¦i
0
' &} is: u
D
÷ 2 ¹ 7 * ¦i
0
' &}. ThereIore. Ior ¦i
0
' &} ÷ 2/3. u
D
÷ 20/3. which is also the
u-value yielded by alternatives A and D. Also. Ior any value oI ¦i
0
' &} · 2/3. we can see
that alternative A should be preIerred to alternative D. and similarly Ior ¦i
0
' &} ~ 2/3.
alternative B should be preIerred to alternative D;
$50

u($100)
.5u($100)¹.5u($0)
÷ u(CE)
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 11 oI 27 HW#7 Solutions
Statement III is true The u-value oI the deal with Iree clairvoyance as a Iunction oI ¦i
0
'
&} is given by: u
FC
÷ 5 ¹ 5 * ¦i
0
' &}. Since Cheryl is risk-neutral. her value oI
clairvoyance is directly proportional to the diIIerence in u-value between the deal with
Iree clairvoyance and the deal with no clairvoyance:
• On the interval |0. 0.25|. u
NC
÷ u
C
÷ 5 so u
FC
u
NC
÷ 5 * ¦i
0
' &}. The
maximum oI this Iunction is 1.25. which is obtained Ior ¦i
0
' &} ÷ 0.25;
• On the interval |0.25. 2/3|. u
NC
÷ u
A
÷ 4 ¹ 4 * ¦i
0
' &} so u
FC
u
NC
÷ 1 ¹
¦i
0
' &}. The maximum oI this Iunction is 1 ¹ 2/3. which is obtained Ior
¦i
0
' &} ÷ 2/3;
• On the interval |2/3. 1|. u
FC
u
NC
is a decreasing Iunction oI ¦i
0
' &}.
ThereIore. the value oI clairvoyance reaches its maximum Ior ¦i
0
' &} ÷ 2/3.

8) Solution: b

Statement I is true For p ÷ 0.5 and q ÷ 0.9. Jackie`s belieIs can be represented by the
Iollowing probability tree:

“i
0
” 0.2
i
0
0.5
0.4 “i
1
” 0.2
0.5
“i
0
” 0.06
i
1
0.1
0.6 “i
1
” 0.54
0.9

Flipping this tree:

i
0
0.2
“i
0
” 0.77
0.26 i
1
0.06
0.23
i
0
0.2
“i
1
” 0.27
0.74 i
1
0.54
0.73

With her prior belieIs. ¦i
0
' &} ÷ 0.4 and so her best decision is alternative A.
However. once she observes the test results. her belieIs will change into either ¦i
0
' 'i
0
¨.
&} ÷ 0.77 or into ¦i
0
' 'i
1
¨. &} ÷ 0.27. ThereIore. iI she observes 'i
0
¨. her best decision
becomes alternative B. which shows that the test is relevant and material to her decision;
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 12 oI 27 HW#7 Solutions
Statement II is Ialse Following up on the previous reasoning. and using the sensitivity
analysis chart. Jackie`s optimal decision becomes alternative B iI she observes 'i
0
¨. but
remains alternative A iI she observes 'i
1
¨;
Statement III is Ialse Jackie`s belieIs can now be represented by the Iollowing
probability tree:


“i
0
” 0.4p
i
0
p
0.4 “i
1
” 0.4(1-p)
1-p
“i
0
” 0.6(1-p)
i
1
1-p
0.6 “i
1
” 0.6p
p

Flipping this tree:

i
0
0.4p
“i
0
” 0.4p / (0.6 - 0.2p)
0.6 - 0.2p i
1
0.6(1-p)
0.6(1-p)/ (0.6 - 0.2p)
i
0
0.4(1-p)
“i
1
” 0.4(1-p) / (0.4 + 0.2p)
0.4 + 0.2p i
1
0.6p
0.6p / (0.4 + 0.2p)

We want to Iind the smallest value oI p > 0.5 such that the change in Jackie`s belieIs
about I will be enough Ior her to make decision B or C instead oI decision A. In other
words. we want to Iind the smallest value oI p > 0.5 such that ¦i
0
' 'i
0
¨. &} > 0.75 or such
that ¦i
0
' 'i
1
¨. &} < 0.1.
• Solving ¦i
0
' 'i
0
¨. &} ÷ 0.75: 0.4p / (0.6 - 0.2p) = 0.75 => p 0.82;
• Solving ¦i
0
' 'i
1
¨. &} ÷ 0.1: 0.4(1-p) / (0.4 + 0.2p) = 0.1 => p 0.86;
ThereIore. the smallest value oI p > 0.5 that will make the test material Ior Jackie is
approximately 82°. At this level oI accuracy. observing 'i
0
¨ will make her choose
alternative B over alternative A.

9) Solution: d
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 13 oI 27 HW#7 Solutions
Statement I is true because we can easily Ilip the arrow Irom E to D to condition B and E
on & only. From this. we can conclude that B and E are irrelevant given & due to the
absence oI an arrow between them.

Statement II is true because arrows Irom uncertainty nodes to decision nodes are
inIormation arrows and an arrow Irom E to C but no arrow Irom E to A implies that the
outcome oI uncertainty E is known beIore decision C but aIter decision A.

Statement III is true because by the time we make decision C. we should have
clairvoyance on uncertainty E. Since the outcome oI uncertainty D aIIects our Iinal value
only through uncertainty E. any inIormation on uncertainty D is worth $0.

10) Solution: c
Statement I is true as mentioned in one oI the lectures. try to imagine a test whose
results would not be made available to you. It is then clear that such a test would be worth
nothing to you.
Statement II is Ialse imagine a test that is irrelevant to your distinction oI interest. In
that case. there is no way that knowing the outcome oI this test will change your best
decision.
Statement III is true in the Iirst example oI weather detector given in the manuscript. the
test that Kim considers purchasing is relevant to the distinction oI interest. and even
material. but not economic: the inIormation provided by the detector is not worth its cost.

Quantitative Problems

1) Superlight
a.
High
0.3 -200
Superlight -65 0.5 -20
** Medium
-65 0.2 25
Low
-125
Basic

As the decision tree shows. the company should thus choose the Superlights. and their
certain equivalent equals -$65.


MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 14 oI 27 HW#7 Solutions
b.
Superlights
-200
High -125
0.3
** -125
Basic
Superlights
** -20
-42.5 0.5 -20
Medium
-125
Basic
Superlights
** 25
0.2 25
Low
-125
Basic


Since the value with Iree clairvoyance equals 50 . 42 $ − and the value without
clairvoyance equals 65 $ − . the value oI clairvoyance must equal
50 . 22 $ ) 65 $ ( ) 5 . 42 $ ( = − − − K.
c. Here is the sensitivity to probability oI the CEs oI the two alternatives.

-250.0
-200.0
-150.0
-100.0
-50.0
0.0
0 0.2 0.4 0.6 0.8 1



The crossing point is the point . 61 . 0 ≈ p

Superlights
Basic
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 15 oI 27 HW#7 Solutions

d. Here is the sensitivity to probability oI the VOC.

0
5
10
15
20
25
30
35
40
45
50
0 0.2 0.4 0.6 0.8 1

The maximum occurs at . 61 . 0 ≈ p Is it a coincidence that the answers to this problem
and the previous problems are the same?
e.
First. we take the prior and the likelihood given and calculate the posterior and the
preposterior.

Pass High
0.3 0.3 0.09 0.13 0.09
High 0.7 0.21 Pass 0.60 0.4
No pass 0.67 Medium
Pass 0.27 0.18
0.5 0.8 0.4 Low
Medium 0.2 0.1 High
No pass 0.64 0.21
Pass 0.33 0.30 0.1
0.2 0.9 0.18 No pass Medium
Low 0.1 0.02 0.06 0.02
No pass Low


Then we plug the new probabilities into the tree.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 16 oI 27 HW#7 Solutions
High
0.13 -200
Superlights -32.09 0.60 -20
** Medium
Pass -32.09 0.27 25
0.67 Low
-125
Basic
-62.75 High
0.64 -200
Superlights -131.82 0.30 -20
Medium
0.33 -125 0.06 25
No pass Low
** -125
Basic



The value with inIormation equals 75 . 62 $ − . the value without inIormation equals 65 $ − .
so the VOI equals 25 . 2 $ ) 65 $ ( ) 75 . 62 $ ( = − − − K.

f. Here is the diagram we would draw Ior this problem.

Superlights
or Basic
Superlight
repair cost
$


Some people will want to draw the Iollowing diagram instead.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 17 oI 27 HW#7 Solutions
Superlights
or Basic
Repair cost
$


Arrows represent complexity. but this is a rather simple problem. We can think about
what the Superlights will cost to repair regardless oI the alternative we choose.

Also. try to avoid inIluence arrows as much as possible; it makes it more diIIicult to
calculate value oI clairvoyance. We will discuss this more in DA II.

g.
Buy
clairvoyance?
Superlights
or Basic?
$
Superlight
repair cost
Report



Don`t conIuse this diagram (Ior the decision to buy clairvoyance) with the diagram Ior the
decision with Iree clairvoyance. which would look like the Iollowing.

Superlights
or Basic
Superlight
repair cost
$


h. The diagram Ior imperIect inIormation looks much like that Ior clairvoyance.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 18 oI 27 HW#7 Solutions
Buy
clairvoyance?
Superlights
or Basic?
$
Superlight
repair cost
Report
Test
indication


Again. don`t conIuse this with the diagram Ior a Iree test. which looks like this.

Superlights
or Basic?
$
Superlight
repair cost
Test
indication


Both oI these are in assessment Iorm (the observed distinction conditioned on the
distinction oI interest). To change them to inIerential Iorm. Ilip the arrow Irom the repair
cost to the indication.

2)
Analytically. you would set up the Iollowing trees. with p representing the accuracy oI the
test.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 19 oI 27 HW#7 Solutions
"Sun" Sun
Sun p 0.4p "Sun" 0.4p / (0.6 - 0.2p) 0.4p
0.4 1-p 0.4(1-p) 0.6 - 0.2p 0.6(1-p)
"Rain" Rain
"Sun" Sun
0.6 1-p 0.6(1-p) 0.4 + 0.2p 0.4(1-p) / (0.4 + 0.2p) 0.4(1-p)
Rain p 0.6p "Rain" 0.6p
"Rain" Rain
Sun
Outdoors 0.4p / (0.6 - 0.2p) 100 1.00
0 0.00
Rain
Sun
"Sun" Porch 0.4p / (0.6 - 0.2p) 90 0.95
0.6 - 0.2p 20 0.32
Rain
Sun
0.4p / (0.6 - 0.2p) 40 0.57
Indoors 50 0.67
Rain
Sun
Outdoors 0.4(1-p) / (0.4 + 0.2p) 100 1.00
0 0.00
Rain
Sun
0.4 + 0.2p 0.4(1-p) / (0.4 + 0.2p) 90 0.95
"Rain" Porch 20 0.32
Rain
Sun
0.4(1-p) / (0.4 + 0.2p) 40 0.57
Indoors 50 0.67
Rain

You could roll back this tree. iust like any other. However. it would be more diIIicult
because oI the presence oI the unknown p; Ior example. you don`t immediately know
which alternative is preIerred at any point in the tree.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 20 oI 27 HW#7 Solutions
Numerically. we could plug in diIIerent values Ior p and Iigure out the value oI
inIormation Ior each one. such as in the diagram below.

0
5
10
15
20
25
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Accuracy
V
a
l
u
e

o
f

I
n
f
o
r
m
a
t
i
o
n


So. we can read oII this diagram that the value oI inIormation equals $10 at two diIIerent
levels oI accuracy: 83° and 17°. Note that a detector that is almost always wrong is iust
as valuable as a detector that is almost always right.

3) Konstantin
a.
The way we see this problem. Konstantin has one decision (CCI or WWS) and Iour
uncertainties. all irrelevant to each other in any state oI inIormation: Driver / Bay area /
Japan. San Jose or San Francisco. Iinding a new client or not in SF. and being promoted
to Senior or Associate in Japan. So. the diagram looks like this.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 21 oI 27 HW#7 Solutions
CCI or
WWS
$
Driver. Bay
area. or Japan
Senior or
Associate in
Japan
San Jose or
San Francisco
Find a new
client in SF


Note that even though the English description oI the problem was quite long and a little
conIusing. the diagram makes it clear that the problem has a simple structure.

Again. some people will want to add arrows; Ior example Irom San Jose or San Francisco
to Iinding the client in SF. Their reasoning might be that iI Konstantin gets moved to San
Jose. he won`t then Iind a new client in San Francisco.

We want to be sure. thereIore. that we understand that I careIully constructed my
distinctions to avoid this; Ior example. the 'Iind a new client¨ distinction means that if
Konstantin gets assigned to SF. he will Iind a new client. I make a similar argument with
the 'senior or associate¨ distinction; it only comes into play iI Konstantin goes to Japan.
ThereIore. we can minimize the number oI arrows in the diagram. accenting the Iact that
this is actually a quite simple problem.

b.
First. there is the issue oI Konstantin`s risk attitude. Since he Iollows the delta-property
and is risk averse. we can set him up with a u-curve
ρ
x
e x u

− =1 ) (
where ρ is Konstantin`s risk tolerance. For each deal he assessed we calculated the risk
tolerance implied by his assessment. For the Iirst we calculated 10 . 100 $ = ρ K. and Ior
the second we calculated 83 . 97 $ = ρ K. so WE decided to use 100 $ = ρ K Ior the rest oI
the problem.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 22 oI 27 HW#7 Solutions
Driver
0.1 25.00 $ 0.22
San Jose
62.03 $ 0.5 42.00 $ 0.34
CCI 0.46 Bay Area 0.34
** 0.3 New Client
0.5 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
62.03 $ 0.462 Senior
0.6 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS


So his best decision is to go with the oIIer Irom CCI.

c.
From the tree above. we can see that WWS would have to oIIer at least $62.03K beIore
Konstantin would change his decision.

d.
Here is a sensitivity oI CE to risk tolerance Ior the two alternatives.

$-
$10
$20
$30
$40
$50
$60
$70
0 200 400 600 800 1000 1200
Risk Tolerance
C
e
r
t
a
i
n

E
q
u
i
v
a
l
e
n
t


The crossover (somewhat hard to see) happens when 68 $ ≈ ρ K.

e.
Now Konstantin has another decision to make. as well as a report Irom his clairvoyant
Iriend. Adding these to the diagram. we get the Iollowing.

CCI
WWS
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 23 oI 27 HW#7 Solutions


To solve this problem. we need to calculate the value oI clairvoyance on each uncertainty.

Driver
1 25.00 $ 0.22
San Jose
25.00 $ 0.5 42.00 $ 0.34
CCI 0.22 Bay Area 0.34
0 New Client
0.5 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
"Driver" 0.451 Senior
0.1 0 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
** 60.00 $ 0.45
WWS
Driver
0 25.00 $ 0.22
San Jose
40.93 $ 0.5 42.00 $ 0.34
CCI 0.34 Bay Area 0.34
1 New Client
0.5 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
73.36 $ 0.519815 "Bay Area" 0.451 Senior
0.3 0 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
VOC = 11.33 ** 60.00 $ 0.45
WWS
Driver
0 25.00 $ 0.22
San Jose
83.37 $ 0.5 42.00 $ 0.34
CCI 0.57 Bay Area 0.34
** 0 New Client
0.5 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
0.6 0.566 Senior
"Japan" 1 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS

CCI or
WWS
$
Driver. Bay
area. or Japan
Senior or
Associate in
Japan
San Jose or
San Francisco
Find a new
client in SF
Which inIormation
to buy?
Report
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 24 oI 27 HW#7 Solutions

Driver
0.1 25.00 $ 0.22
San Jose
62.43 $ 1 42.00 $ 0.34
CCI 0.46 Bay Area 0.34
** 0.3 New Client
0 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
"San Jose" 0.464 Senior
0.5 0.6 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS
62.03 $ 0.462233 Driver
0.1 25.00 $ 0.22
San Jose
61.64 $ 0 42.00 $ 0.34
VOC = 0 CCI 0.46 Bay Area 0.33
** 0.3 New Client
1 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
0.5 0.460 Senior
"San Francisco" 0.6 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS

Driver
0.1 25.00 $ 0.22
San Jose
64.36 $ 0.5 42.00 $ 0.34
CCI 0.47 Bay Area 0.38
** 0.3 New Client
0.5 0.41 1 53.00 $ 0.41
San Francisco 0 32.00 $ 0.27
"New client" 0.475 Senior
0.4 0.6 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS
62.03 $ 0.462233 Driver
0.1 25.00 $ 0.22
San Jose
VOC = 0 60.51 $ 0.5 42.00 $ 0.34
CCI 0.45 Bay Area 0.31
** 0.3 New Client
0.5 0.27 0 53.00 $ 0.41
San Francisco 1 32.00 $ 0.27
0.6 0.454 Senior
"No new client" 0.6 0.57 0.85 95.00 $ 0.61
Japan 0.15 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS


MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 25 oI 27 HW#7 Solutions
Driver
0.1 25.00 $ 0.22
San Jose
67.50 $ 0.5 42.00 $ 0.34
CCI 0.49 Bay Area 0.34
** 0.3 New Client
0.5 0.33 0.4 53.00 $ 0.41
San Francisco 0.6 32.00 $ 0.27
"Senior" 0.491 Senior
0.85 0.6 0.61 1 95.00 $ 0.61
Japan 0 35.00 $ 0.30
Associate
60.00 $ 0.45
WWS
66.34 $ 0.4849 Driver
0.1 25.00 $ 0.22
San Jose
35.68 $ 0.5 42.00 $ 0.34
CCI 0.30 Bay Area 0.34
0.3 New Client
0.5 0.33 0.4 53.00 $ 0.41
VOC = 4.31 San Francisco 0.6 32.00 $ 0.27
0.15 0.451 Senior
"Associate" 0.6 0.30 0 95.00 $ 0.61
Japan 1 35.00 $ 0.30
Associate
** 60.00 $ 0.45
WWS


So. the Iour values oI clairvoyance are $11.33K. $0. $0. and $4.31K. Konstantin should
choose to Iind out iI he will remain a driver. or be moved to the Bay area or Japan. His
certain equivalent Ior this new deal would be $73.36K. and he should pay up to $11.33K
to his Iriend the clairvoyant Ior this inIormation.

MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 26 oI 27 HW#7 Solutions

f.
Now. instead oI controlling the uncertainty that is revealed. Konstantin must leave it up to
chance. His new diagram includes that uncertainty.






You might think that we simply need to take an e-value oI the VOC`s we calculated in the
previous problem. Again. not quite correct (but close).


0.25 73.36 $ 0.520
0.25 62.03 $ 0.462
65.84 $ 0.482295 0.25 62.03 $ 0.462
0.25 66.34 $ 0.485
VOI = 3.80


We have Iour possible uncertainties that the clairvoyant might resolve Ior us (represented
by the Iour branches oI the tree). We already know the values with Iree clairvoyance Ior
each branch. We can thereIore Iind the value with Iree random clairvoyance like
Konstantin`s Iriend is oIIering. That equals $65.84K. and so the value oI the Iriend`s
services equals $3.80K.
CCI or
WWS
$
Driver. Bay
area. or Japan
Senior or
Associate in
Japan
San Jose or
San Francisco
Find a new
client in SF
Buy inIormation? Report
Uncertainty that
clairvoyant will
reveal
MS&E 252 Handout #28
Decision Analysis I December 4
th
. 2004

Page 27 oI 27 HW#7 Solutions