Decision Models and Optimization: Sample-Endterm-with solutions

Instructor: Milind G. Sohoni

email: [email protected]
Indian School of Business, Gachibowli, Hyderabad 500032

Term 2

Basic Information:

1. This exam document has 6 pages including this page. Please ensure that you have all pages before
you begin. You have two (2) hours to complete the exam.
2. There are 8 questions for a total of 42 points.

3. You can use the blank pages after the last page of this document for rough work. Your rough
work will not be graded.
4. Answers to the multiple choice questions must be marked next to the choices. Choose the best
answer by marking one item on the answer sheet. In case of numerical questions, if none of
the choices exactly match your calculations you may mark the choice which is closest to your
calculations.
5. Any scratch work on the exam will NOT be graded. Only answers written in the
legitimate writing space (in between the questions) will be marked.
6. The exam is completely self-explanatory. The instructor and the academic associates will not be
available to clarify doubts.
7. Please submit all your work, including the rough sheets, at the end of the exam.

Ground rules:

1. You are expected to obey the ISB honor code while taking this exam.
2. This exam is in-class, closed book, and closed notes. Only a one-sided A4-sized cheat sheet
is allowed. You can only use a non-programmable calculator. You cannot use your computer.

0. The version of this exam is:

A. Version A. B. Version B. C. Version C. D. Version D.
DMOP (Version A) Page 2 of 6 June

Mr. Jaggi and Tony own a popular cinema hall Theater Bollywood. The seating capacity of the hall
is 1000. They select movies that are action oriented, have a strong emotional content with many
lively songs and dances. Thus the demand does not vary much from movie to movie. They plan to
continue this tradition in future. Last year was a good in sales. It was estimated that the demand
is exponentially distributed with mean of 700. For simplicity, the price of ticket may be assumed
to constant at |200. This will remain the same for the next year. J&T typically pay a charge of
|120,000 per show to the film distributor. The distributor in turn makes up revenues when less than
500 seats are sold. In that case he effectively purchases tickets so that the total revenue to J&T
equals that obtained by selling 500 seats. For example if in a show 350 customers show up, then the
distributor pays back 150 × 200 = |30, 000 to Jaggi and Tony (J&T) so that his total charge equals
|90,000. This year J&T have conducted a massive advertising campaign on the music channels on
television. In addition, some of the biggest stars, super stars, mega stars and item stars visited their
hall leading to free publicity on news channels which of course covered this non-stop for weeks. J&T
believe that this has improved the profile of their cinema hall and expect a 20% increase in demand,
that is, they expect the demand to be 1.2 times the exponential distribution with mean 700 per
show.

1. (4 points) If D denotes an exponentially distributed random variable with mean 700 (a proxy for
typical demand last year), write down a single algebraic expression for
1. The profit (revenue-cost) earned by J&T in a typical show this year as a function of D.
2. The amount the distributor earns in a show this year as a function of D.

Solution:

J&T’s Profit = 200 × max {min {1.2D, 1000} , 500} − 120, 000.
Distributor’s Profit = 120, 000 − 200 × max {500 − 1.2D, 0} .

2. (2 points) We conducted 10,000 simulations for this show on @Risk. The results are shown below:

Figure 1: J&T’s profit distribution.

c Milind G. Sohoni ISB
DMOP (Version A) Page 3 of 6 June

Percentile Value Percentile Value

0 -20,000 90 80,000
10 -20,000 100 80,000
20 -20,000
30 -20,000
40 -20,000
50 -3,287
60 32,806
70 79,656
80 80,000

Table 1: Simulated profit distribution statistics.

3. (3 points) Based on the simulation results, what is approximately the probability that in a given
show the profits are between 17,000 and 25,000?
A. < 10%. B. 10%. C. 30%. D. 90%.

Solution:
As the probability of profits lying between -3,287 and 32,806 is 10%, the required probability
should be less than 10%.

4. (3 points) Based on these simulation results what is the estimated probability that the distributor
will receive less than |120,000 in a show?

Solution:
Distributor will receive less than |120,000 if the demand (D) falls below 500. In that case he
is required to pay for the (500 − D) seats, thus his profit falls below |120,000. This is same as
the probability of maximum possible loss to the theater owner. This loss is equal to (minimum
revenue – fixed cost) i.e. (500 × 200) –120, 000. Thus we are required to find the probability of
loss to theater owner of |20,000. This probability can be easily observed from the frequency
distribution of profits and can be approximated to be 45%.

5. (2 points) Approximately, how many samples are needed in simulation so that the 95% confidence
interval for the expected profits has absolute width equal to 500?

Solution:
The confidence interval width W is given by
σ
W = 2 × Z α2 × √ .
n

We are given W = 500, Z α2 = 1.96, and σ = 44, 933. This implies n ≈ 124, 097.

6. (3 points) If the capacity of the cinema hall was increased by 5 seats, by approximately how much
would the revenues increase?

c Milind G. Sohoni ISB
DMOP (Version A) Page 4 of 6 June

Solution:
The revenues increase by 5 × 200 since 5 extra seats can be sold now. The probability of this
happening can be approximated by the probability that the demand is greater than 1000. This
is approximately equal to the probability of maximum profit for the theater owner, which can
be inferred as 0.3 from the frequency diagram. So, revenues increase by 1000 with probability
0.3 or the expected revenues increase by 0.3 × 1000 = 300.
More rigorously,
X
Expected increase = (Pr{1000 < D < 1005} × (D − 1000))×200+Pr {D ≥ 1005}×(200 × 5) .
D

The first term denotes the expected increase if the demand lies between 1000 and 1005. This
term can be ignored as the probability of D lying between 1000 and 1005 is very small. The
second term denotes the expected increase if demand is greater than 1005. In this case the
increase in revenues will be fixed at the revenue fetched by those extra 5 seats (= 200 × 5).
Now, the probability that demand is greater than 1000 is equal to the probability of maximum
profit (200 × 1000 − 120000) for the theater owner i.e. the probability of profit = 80,000. This
probability can be observed from the frequency distribution and comes out to be 0.3. This
probability can be used as a proxy for the probability for demand greater than 1005. Thus, the
expected increase in revenues will be (0.3 × 1000) = 300.

7. Ashwini’s dog show is scheduled to appear in Pune on July 17. The profits obtained are heavily
dependent on the weather. In particular, if the weather is rainy, the show loses |28,000 and if sunny,
the show makes a profit of |12,000. (We assume that all days are either rainy or sunny.) Ashwini
can decide to cancel the show, but if she does, she forfeits a |1,000 deposit she put down when she
th
accepted the date. The historical record shows that on July 17, it has rained 41 of the time for the
last 25 years.
(a) (5 points) What decision should Ashwini make to maximize her expected net monetary return?

Solution:
She should go ahead with the show. In expectation she could make |2,000. Ashwini’s
decision tree is shown below.
25.0% 25.0%
Rains
-28000 -28000
TRUE Rain
Perform
0 2000
75.0% 75.0%
Doesn't rain
12000 12000
Show
Ashwini's DT.
2000
FALSE 0.0%
Cancel
-1000 -1000

(b) (5 points) What is the expected value of perfect information?

Solution:
If it rains, she should cancel and the lose her deposit of |1,000. If it doesn’t then she could
make |12,000. Her perfect information tree is as shown in the figure below.

c Milind G. Sohoni ISB
DMOP (Version A) Page 5 of 6 June

TRUE 25.0%
Cancel
-1000 -1000
25.0% Cancel or not?
Yes
0 -1000
FALSE 0.0%
Don't cancel
-28000 -28000
Rains?
EVPI Tree
8750
FALSE 0.0%
Cancel
-1000 -1000
75.0% Cancel or not?
No
12000
TRUE 75.0%
Don't cancel
12000 12000

The EVPI is |8,750 − |2,000 = |6,750.

(c) Ashwini has the option to purchase a forecast from Ashu’s weather wonder. Ashu’s accuracy
varies. On those occasions when it has rained, Ashu has been correct (i.e. predicted rain) 90%
of the time. On the other hand, when it has been sunny, he has been right (i.e. he predicted
sun) only 80% of the time.
i. (5 points) If Ashwini had the forecast, what strategy should she follow to maximize her
expected net monetary return?

Solution:
Let F + represent Ashu predicts rain and F − that he doesn’t (predicts it’ll be sunny.)
Similarly, let R+ represent if it actually rains, and R− if it doesn’t (i.e., it is sunny.)
We know that Pr {F + | R+} =0.9 and Pr {F − | R−} = 0.8. We also know that the
Ashwini’s priors on it raining are Pr {R+} = 0.25 and Pr {R−} = 0.75. We can
compute

Pr {F − | R+} = 1 − Pr {F + | R+} = 0.1,

Pr {F + | R−} = 1 − Pr {F − | R−} = 0.2,
Pr {F +} = 0.9 × 0.25 + 0.2 × 0.75 = 0.375 : Law of total probability,
Pr {F −} = 0.1 × 0.25 + 0.8 × 0.75 = 0.625,
0.9 × 0.25
Pr {R + | F +} = = 0.6 : Bayes’ rule,
0.375
Pr {R − | F +} = 1 − Pr {R + | F +} = 0.4,
0.1 × 0.25
Pr {R + | F −} = = 0.04 : Bayes’ rule,
0.625
Pr {R − | F −} = 1 − Pr {R + | F −} = 0.96.

If we draw the imperfect information tree, as shown in the figure below, her optimal
strategy is to get Ashu’s forecast. If Ashu predicts rain, cancel the show. Else, perform
it.

c Milind G. Sohoni ISB
DMOP (Version A) Page 6 of 6 June

FALSE 0.0%
Cancel
-1000 -1000
Do what?
Ashwini't DT
6125
25.0% 0.0%
Rains
-28000 -28000
FALSE Rain?
Perform
0 2000
75.0% 0.0%
Doesn't
12000 12000
60.0% 0.0%
Rains
-28000 -28000
FALSE Rain?
Perform
0 -12000
40.0% 0.0%
Doesn't
12000 12000
37.5% Perform?
Rainy Pr{F+}
0 -1000
TRUE 37.5%
Cancel
-1000 -1000
TRUE Ashu's report
Approach Ashu
0 6125
4.0% 2.5%
Rains
-28000 -28000
TRUE Rain?
Perform
0 10400
96.0% 60.0%
Doesn't
12000 12000
62.5% Perform?
Sunny Pr{F-}
0 10400
FALSE 0.0%
Cancel
-1000 -1000

ii. (5 points) How much should Ashwini be willing to pay to have the forecast?

Solution:
The EMV of Ashwini’s imperfect information decision tree is |6,125. She shouldn’t
be willing to pay more than |6,125 − |2,000 =|4,125 for weather information of the
“quality/accuracy” that Ashu provides. That is the expected value of imperfect infor-
mation. However, notice that it is more than half the value she could gain from perfect
information.

8. (5 points) Ten passengers get on an airport shuttle at the airport. The shuttle has a route that
includes 5 hotels, and each passenger gets off the shuttle at his/her hotel. The driver records how
many passengers leave the shuttle at each hotel. How many different possibilities exist?
   
5 10 14
A. 10 B. C. D. 510
5 4

c Milind G. Sohoni ISB