100 aGPSS Models: Cases and Exercises

Ingolf Ståhl, Stockholm School of Economics, 2018

Below are 100 aGPSS models. Most of them are meant to be

exercises for students, but many should also be seen as cases, i.e.
examples of applications of aGPSS.

The first 57 models in this booklet are meant to be studied

alongside with the study of the 50 different lessons in the aGPSS
textbook. Most of the remaining models are more advanced and
often require that all 50 lessons have been studied. None of the
models are from the aGPSS textbook, but are all new.

Since there is generally only one exercise in each of the 50 lessons

in this textbook, many students and teachers have declared a need
for more exercises. One of the aims of this booklet is therefore to
answer to this demand. After each exercise with a number lower
than 58, there is within parentheses a number that refers to the latest
lesson in the textbook that contains the information that is necessary
to solve the exercise.

Another purpose of this booklet is to show that GPSS is really a

general purpose system with at very wide area of applications. A
third purpose to give ideas for project work, which is often the final
part of a basic simulation course. Several of the fairly small model
in this booklet have later been expanded into a larger model
representing an actual situation in a company. In fact, only a few of
the models here are quite large with around 200 blocks. We have
found it better to present many smaller models than just a few very
large ones.

Among the issues studied, we have the establishment of optimal

equipment life (model 68), overbooking and price differentiation in
1
the airline industry (66), bidding on stocks of an oil company with
uncertainty about the success of an oil exploration process (69), a
basic example of the value of perfect information in decisions under
risk (92) and a model of stocking of perishable goods in a
supermarket (88). We also illustrate some basic ideas of continuous
simulation as applied to business problems with the aim of doing
market forecasts (73) and evaluation of the value of a high tech
corporation (99). We also deal with the evaluation of European
options, which for the case of constant volatility can be solved
analytically, but for the case of volatility varying over time requires
simulation (97). We also discuss simulation based costing (65) and
the use of laptop based simulation models for sales support (98).

We present many manufacturing examples (e.g. 63, 67, 75, 82, 83),
but also examples of logistics with trucks (100), shipping (77, 78)
and taxi operations (72). Some examples are from services like
restaurants (94), gas stations (62) and libraries (74). Among other
examples we have an IT-exhibition (81), systems with elevators (79,
80), administration of secretarial work (86), software development
(93), car maintenance (95), super market retail operations (54, 59,
89) and partner selection (71). There are also examples of oligopoly
theory models (85, 90) and of coordination of production, inventory
and pricing policy to improve the company’s cash flow. (87, 91).

It should finally be mentioned that several exercises are built on

programs in other books. The acknowledgements to these
exercises are given at the top of the suggested solutions, which are
available in electronic form (see www.aGPSS.com). It should here
be mentioned that the many of these examples are case studies
from T. Schriber’s famous “red book”, Simulation Using GPSS.
Y. Matsuyama’s careful review of the present booklet, leading to
many substantial improvements, must also be acknowledged.
2
1. You have started to work on the assembly line at a car factory to
make some extra money. You do riveting on the distributor box of
the newest model. You do piece-work, but you find it annoying
that you cannot determine your own production speed, since you
are dependent on the guy who works just in front of you on the
line and who mounts the boxes. The boxes arrive at his station
with a time distance of between 5 and 9 seconds. It then takes him
between 6 and 12 seconds to mount a box. It then takes between 3
and 5 seconds for you to do the riveting. (All times are uniformly
distributed.) What percent of the time is each one of you two busy
working, seen over a period of 10 minutes? (5)

2. Combine Models 08 and 13 so that you from the same program

get statistics on both waiting time and time spent in the store. (8)

3. Change Model11, in the first instance so that customers come

on average every 28 (instead of 18) minutes apart. Next change
also the half-width of the ADVANCE block from 5 minutes to 25
minutes, implying that service time varies between 0 and 50
minutes. Compare the table of the first run with the table in Figure
47. (8)

4. Two types of customer arrive at a one-chair barber shop.

Customers of the first type want only a haircut. Their InterArrival-
Time distribution is 35 + 10 minutes. Customers of the second
type want a shave as well as a haircut. Their InterArrival-Time
distribution is 60 + 20 minutes. The barber provides service to his
customers on a first-come, first-served basis. It takes the barber 18
+ 6 minutes to give a haircut. When he gives a shave, 10 + 2
minutes are required. Collect data on the waiting line that forms
ahead of the barber, both total and for the two types of customers.
Run the model through eight hours of simulated time. (9)
3
5. A port has two jetties, each of which can be used for unloading
by one boat at a time. Boats arrive at the port periodically and
must wait if no jetty is currently free. When a jetty is available, a
boat may dock and start to unload. When this activity has been
completed, the boat leaves the jetty and sails away. The port
authority has a pool of three tugs. A tug is required for docking
and when a boat leaves its jetty. On average boats arrive 10 hours
apart, but this can vary between 7 and 13 hours. It takes 14 + 6
hours to unload. Docking and leaving, using tugs, take 2 hours
each. Simulate for 28 full days of 24 hours. (11)

6. One doctor and two nurses work in a clinic. Patients arrive

every 15 + 10 minutes. Patients first go to one of the nurses
(whichever one is free) for tests, which takes 25 + 10 minutes.
They then proceed to the doctor for examination, which takes 15 +
10 minutes. Since some patients have complained about very long
waiting times, the doctor is considering a different arrangement.
He realizes that if the nurse, who has only made the tests on the
patient, also stayed with the patient during the examination, the
time for examination would decrease to 10 + 9 minutes. Write a
program for both the present and the proposed alternative.
Simulate for a seven-hour day to see if there will be fewer
patients, who have very long waiting times. (11)

7. Alter Model19 to allow for two lathes and two grinders. The
average processing times for the machines are now doubled, but
the half-widths are unchanged. (11)

8. Modify Model19, so that instead of starting production of a unit

of A every 25 minutes and one of B every 50 minutes, production
of a unit of A or B commences as soon as the lathe is free.

4
Compare the results of this new program with those of the old
program. (12)

9. A student owns two cars. Each day, he uses a car to get to school
and back. Car 1 is old and has probability 0.79 of starting on any
day, while car 2, even older, starts with probability 0.71. For either
car, once it is started, the chance that it completes a trip to or from
school is 0.95. Compute the probability that he will make it to
school and back on any particular day. First do the calculation
manually, then write a GPSS program to do it. (13)

10. Ambulances are dispatched at a rate of one every 10 + 7

minutes in a large metropolitan area. Fifteen percent of the calls
are false alarms, which require 12 + 2 minutes to complete. All
other calls can be one of two kinds: The first kind are classified as
serious. They constitute 25 percent of the non-false alarm calls and
take 25 + 5 minutes to complete. The remaining calls take 20 + 10
minutes to complete. There are three available ambulances, and
they are on call at any time. Simulate the system for 500 calls to
be completed. Give statistics on how long serious alarm cases wait
before an ambulance is on its way. (13)

11. People arrive at a self-service cafeteria at the rate of one every

40 + 20 seconds. Forty percent go to the sandwich counter, where
one worker makes a sandwich in 60 + 30 seconds. The rest go to
the main counter, where one server spoons the prepared meal on to
a plate in 45 + 30 seconds. Then all customers pay a single cashier,
which takes 25 + 10 seconds. For all customers, eating takes 20 +
10 minutes. After eating, 40 percent of the people return to buy a
coffee and pay the cashier for this. Drinking coffee takes 10 + 2
minutes.

5
Simulate a day of five hours. The output should include data on
average time spent in the cafeteria, the maximum number of people
at any time and average waiting time in front of the cashier. (13)

12. Modify Model26 to take into account that 40 percent of the

customers are impatient and balk if Joe is busy, while the remaining
60 percent balk if there are 4 or more persons waiting. (16)

13. We simulate a simple bicycle store for 250 days of 8 hours. The
store has space for 15 bikes in stock. The stock is replenished with
six bicycles every 10th day. If there is no space in the inventory
room, the delivery of the bike has to wait until there is such space.
Customers wanting to buy a bike arrive every x hours. Test with x =
10 and 20 to see if deliveries of new bikes are delayed or if
customers have to wait for a bike. (17)

14. In traffic systems, like that of Model31, it is possible, when the

light is green for a short time (i.e. there is a red light on the other
street), that all the cars that are waiting might not be able to cross
the street before the traffic light turns red again. Take this problem
into account by modifying Model31 according to the following
assumptions: Before crossing the street, each car will be in the
position of first car in the waiting line. It will be in this position for
1 - 3 seconds and from there cross the street, if there is not a red
light. (18)

15. We model a store, having two salespersons, each needing 4 + 2

minutes to serve a customer. To determine the IAT of the
customers, one has recorded the arrival times of various customers
during the day, leading to the set of numbers presented on the next
page, dealing with minutes after the opening of the store.

6
0.6, 1.6, 1.8, 3.4, 3.8, 5.2, 7.1, 9.3, 10.1,
12.8, 14, 16, 19.5, 24.9, 25.1, 29.2, 34.8,
35.6, 39.3, 39.8, 40.8, 45.1, 46.2, 46.3, 51.2,
52.4, 55.3, 57, 57.6, 62.3, 63.8, 64.9,70.1,
70.2, 72, 76.5, 77.7, 80.9, 81.6, 82.7, 85, 89.2,
89.4, 90.3, 91.7, 94.2, 95.1, 96.8, 99.4

On the basis of these numbers one can calculate the time between
the arrivals and then distribute them into six bins for 0 - 1, 1 – 2, .. ,
5 - 6 minutes etc.

Next one can formulate an empirical IAT function and use this
function in the model. Simulate the operations of the store for 10
hours. (20)

16. In a store, sales only refer to one product, but with different
prices for two types of customers, arriving 18 + 6 minutes apart:
Members, accounting for 60 percent, pay $ 20, but the rest must pay
$ 30. If there are no units in stock, when a customer arrives, the
customer waits until new stocks arrive. Every day 25 new units are
delivered provided at least one customer is waiting and the store has
enough cash to pay $ 15 for each unit. When we start, the company
has 400 in cash. Simulate an operation of 30 days of 8 hours. (22)

17. A store in central Stockholm is handled by two salespersons,

who serve the customer who has waited the longest time. The
InterArrival Times of the customers are exponentially distributed
with a mean of 1.5 minutes. The service times are normally
distributed with a mean of 8 minutes and a standard deviation of
one minute. Customers balk if there are already a total of 6
customers waiting. The store is open from 10 AM to 7 PM. After 7
PM no new customers are allowed into the store. The store is closed
when all customers who have entered have been served. (23)
7
18. At a one-car wash facility, it must be decided how many spaces
to provide for cars waiting to use the facility. Cars arrive in a
Poisson stream (= according to an exponential distribution) with an
average InterArrival Time of 5 minutes. Car-washing time is
exponentially distributed with a mean of 4 minutes. Potential
customers who find no waiting space available go elsewhere to have
their car washed.

Build a model of this simple system, then use the model to observe
system behavior for the alternatives of one, two and three waiting
spaces. For each configuration, simulate for one eight-hour day of
operation. Basing your answer on model output, estimate the
fraction of potential customers who actually remain at the car wash
to be served. (23)

19. Modify Model08 so that Average time/trans, printed in the

queue statistics, is also obtained as a savevalue. Total waiting time
is calculated by subtracting CL, i.e. the value of the present clock,
from this total time at each entry into the queue and by then adding
CL to this time at each exit from the queue. At stop time, the
product of CL and the number of customers remaining in the queue
(in the block prior to SEIZE) is also added to the total time. This
total waiting time is then finally divided by the number of entries
into the queue. (25)

20. The quality of a pot in Exercise 20 in the textbook, and hence

the price to be charged, varies in a random fashion. 30 percent of
the pots are of low quality and can be fetch a price of $100; 60
percent are of medium quality, fetching a price of $ 150, while 10
percent are of high quality, fetching a price of $ 200. Have the
simulation also calculate the revenues of the production during the
day. (25)
8
21. Modify Model39 as follows: Customer arrivals now follow an
Erlang distribution with n = 3 and service times follow the normal
distribution. Average times are not changed, but service times are,
so that 95 percent of all times lie between 15 and 35 minutes. Run
for one and two salespeople. (26)

22. A shop buys and sells antiques. Persons selling antiques arrive
at the shop according to the exponential distribution with an average
IAT of 2 hours. The price paid for an antique item is on average $
100, but the prices vary according to a normal distribution with a
standard deviation of $ 20. The shop always pays cash. The buyers
arrive at the shop also according to the exponential distribution, but
with an average of 5 hours. The shop sells antique items for $ 150
on average, but prices vary according to a normal distribution with a
standard deviation of $ 25. The customers obtain credit for 4 days,
but the actual times until payment follow an Erlang distribution
with n = 3 and with an average time of 6 days.

The store is open between 9.00 a.m. and 5.00 p.m. each day, also
Sundays. When the store is closed at 5.00, all customers leave. At
simulation start the inventories are valued at $ 10,000 and there is $
50,000 in cash. How much cash will the shop have and what is its
inventory worth after a month of 30 days? Every item in the
inventory is valued at $ 100, irrespective of the purchase price. (26)

23. Change Model52 in the following way: The initial price is to be

read in as $ 33,000, but if cash falls below $50,000 the price shall
be increased by x percent, where x is a value read from the
keyboard. The price is, however, not allowed to increase above
$50,000. Test the effect of different values of x. The aim is to be
able to do three runs in a row without getting a negative cash
balance. (30)
9
24. In a one-line, one-server queuing system, arrivals occur in a
Poisson pattern (i.e. according to the exponential distribution) with
a mean rate of 14 arrivals per hour. Service times follow the
exponential distribution, but the mean service time depends on the
content of the waiting line ahead of the server, as follows:

Contents of the waiting line Mean service time

(minutes) (minutes)
0 5.5
1 or 2 5.0
3, 4 or 5 4.5
6 or more 4.0
Build a model of the system. If the arrival rate increases by one
additional arrival per hour, will the server still be able to handle
the flow of customers reasonably well, or will the waiting line
become much longer? Run the model for 500 customers! (30)

25. For Model40, produce graphs over the development of cash

during the day, for various values of the starting position on cash in
the morning. Since cash might become negative, the program must
be rewritten, so that cash is not handled as a storage, but as a
savevalue, e.g. X$cash, which may assume negative values. It
should be possible to change the starting value of cash interactively,
i.e. without having to change the program. Note that the graph
might be illegible, if more than five days are simulated. (30)

26. Customers arrive at a store every 25 seconds on average,

following an exponential distribution. Service time is likewise 25
seconds on average, also following an exponential distribution. A
customer buys on average for $10.00. The store owner contemplates
whichever of one or two salespersons, each costing $ 100 per day,
leads to the higher daily profit. Run 40 times. (31)
10
27. Modify Model15 so that the cost of waiting for the customers
can be calculated and printed, assuming that they value their time at
an average of $ 6.00 an hour. (32)

28. Rewrite Exercise 12 in the textbook, so that one can calculate

the total costs of lost production, repairers and reserve machines,
assuming a repairer costs $13.75 per hour, a reserve machine $30
per day and the hourly cost penalty for having fewer than 50
machines in production is $20 per machine. Allow for a change of
the value of both the number of repairers and the number of reserve
machines. Run the program for all combinations of values on these
two variables from 1 to 3 to study variations in costs. (34)

29. Modify Model21 so that one of the two salespeople (or barbers)
only works part time - for three hours a day, starting at noon and
stopping at 3.00 p.m. The store opens at 9.00 a.m. and closes at 5.00
p.m. Only statistics on waiting time are required. (36)

30. Messages are created at the rate of one every 30 + 10 seconds.

They are sent over a communication channel, one at a time. Twenty
percent of them require a reply, which returns over the same
channel after a delay of 60 + 30 seconds. Assume the transmission
time for the original message to be 25 + 15 seconds and for the
reply 15 + 5 seconds. Assume also that both messages and replies
can be stored for transmission, if necessary. Compare the time it
takes to complete 100 messages with replies: (1) when the replies
have priority and (2) when the messages have priority. (36)

31. Modify Exercise 36 in the textbook. We assume that all arrivals

follow the exponential distribution, that stay and operation times
follow the Erlang distribution with n = 3, and that the average IAT
is 9 h. for the general patients and 8 h. for the operation patients.
11
Furthermore, we assume that the operation theater does not close
when an operation is in progress, but that no new operation starts, if
a total of 5 operations have already been carried out since the
operation theater was opened in the morning. We now assume that
patients requiring an operation have priority for beds over patients
who are there for a short hospital stay without operation. (36)

32. In a factory, a tool crib is manned by a single clerk. The clerk

checks out tools to mechanics, who use them to repair failed
machines. The IATs and the times to process a tool request (both in
seconds) depends on the type of tool, as follows:
Category Mechanic Service
of tool IAT time
1 420 + 360 300 + 90
2 360 + 240 100 + 30
The clerk has been serving the mechanics first-come, first-served,
irrespective of their requests. Owing to lost production, it costs $9
per hour when a mechanic waits for service at the tool crib,
regardless of the tool to be checked out. Management believes that
the average number of waiting mechanics can be reduced, if
Category 2 requests are serviced at the tool crib before those in
Category 1. Model the tool crib for each of the two queue
disciplines, simulating each case for an eight-hour work day. Does
‘first-come, first-served, within Priority Class’ reduce the average
number of mechanics waiting in line? Calculate also the total daily
cost of lost production for each priority discipline. (36)

33. An office has two copying machines. Secretaries from the

general office arrive to use a machine at the rate of one every 5 + 2
minutes. They choose a machine at random, whether it is busy or
not, and they stay with their choice.
12
A secretary from the president's office arrives at intervals of 15 + 5
minutes. If a machine is free, it is used. If not, the secretary goes to
the head of the line for machine number 1. Assuming that all jobs
take 6 + 3 minutes, determine the time needed for the completion of
100 jobs of all types. (37)

34. Modify the program of exercise 32 above so that, in a single

batch run, 40 different days of operation will be simulated for the
tool-crib problem. Arrange the model so that both the ‘no priority’
and ‘priority’ queue disciplines are investigated with the same
program. (Note that a savevalue determined initially can be used as
priority.) The resulting output will provide two samples, each
consisting of 40 different ‘total waiting costs’ estimates for the
corresponding disciplines. Let aGPSS compute the mean and
standard deviation for each of these two samples. How much better
is the priority discipline based on these samples? (37)

35. Customers pass through the following servers: a weighbridge

(on the way in), an unloading bay, a washer, a loading bay and the
weighbridge (on the way out). Queues q1 to q5 form in front of
these five servers. q1 and q5 both form in front of the weighbridge.
q5 (customers about to leave) has priority over q1.

The InterArrival Time between successive customers is exponential

with 17 minutes as the average. The first customer arrives at time 5.
No customers are accepted after time 360, but customers already in
the system finish being served. Service times are all uniform
between the following limits: for the weighbridge 3 – 7, the washer
5 – 7, the unloading bay 8 – 16 and the loading bay 15 - 20.
Investigate the reduction in average customer waiting time obtained
by acquiring a second loading bay. (37)

13
36. At an attraction at an amusement park, with a capacity for 8
children, 50 per cent of the children, who have taken a ride, want to
repeat the ride. Those who already have taken the ride have to wait
in line after those who have not yet taken the ride. Hence, a child
will have a higher priority, when newly arrived, than when having
already taken the ride. Model this for the case that the children
arrive 10 + 2 seconds apart and a ride takes 50 seconds. Simulate
for a time of 5 minutes. (37)

37. Modify Model73 so that correct statistics for server utilization

are also obtained. (38)

38. Modify Model39 to take into account a varying intensity in

customer arrivals during the day and the possibility of hiring a
part-timer as the second salesperson. Assume that the store is now
open between 9 a.m. and 6 p.m. Average IAT is 25 minutes 9 a.m.
- 11 a.m.; 15 minutes 11 a.m. - 1 p.m.; 30 minutes 1 p.m. - 4 p.m.
and 10 minutes 4 p.m. - 6 p.m.. The specific arrival times are
exponentially distributed. The part-timer is willing to work any
four hours in a row. Should the part-timer be hired to work
between 11 a.m. and 3 p.m. or between 2 p.m. and 6 p.m., if the
objective is to sell as much as possible? Assume that all other
factors than those mentioned here are as in Model39. (38)

39. In a foundry, finishing work on castings is done by a machine.

Only one worker is needed to operate such a machine. The work
consists of a sequence of two processes, Process 1 and Process 2.
The steps to do the work are:
1. Perform Process 1.
2. Re-position the casting on the machine.
3. Perform Process 2.

14
4. a) Unload the finished casting from the machine.
b) Store the finished casting.
c) Fetch the next rough casting from the storage area.
d) Load this rough casting onto the machine.
5. Return to step 1 above.
An overhead crane is needed at each step involving movement of
the casting, i.e. steps 2 and 4 above. Whenever a crane is not being
used for one of these steps, it is in an idle state. The finishing
machine goes repeatedly through a single closed cycle consisting of
the four steps listed above. The crane, on the other hand, can go
through each of two distinct cycles, depending on whether it is
being used for step 2 or step 4. The times (in minutes) required to
perform steps 1 and 3 in the finishing process are as follows:
Time for Cumulative Time for Cumulative
process 1 frequency process 2 frequency
< 60 0.00 < 80 0.00
70 0.12 90 0.24
80 0.48 100 0.73
90 0.83 110 1.00
100 1.00

The handling of the casting in step 2 requires 15 + 5 minutes. The

sequence described by step 4 takes 30 + 5 minutes. Management
wants a simulation study of the finishing department. It wants to
determine what utilizations would result if the number of machines
served by a single crane were three, four, five or six. Secondly, it
wants to know what the effect would be if two cranes or three
cranes were used. For each condition, which is input interactively,
simulate for five 40-hour work weeks. In case of a conflict, step 2 is
to be given priority over step 4. (38)

15
40. Customers come to a store with two salespeople on average
every 10 minutes following an exponential distribution. The
average service time is 15 minutes. Hardly any customer takes less
than five minutes or more than half an hour. It appears that the
triangular distribution could be a convenient approximation of
service times. Simulate for a day of nine hours. (39)

41. Run Exercise 24 above again for 500 customers, but now with
a print-out after every 100th customer. Hint: The decimal fraction
of n/100 is = 0 for every such customer. Use FN$dec. (39)

42. Modify Model22 in such a way that each worker finishes work
for the day and goes home as soon as he has produced 14 pots. (40)

43. A party of 100 people has gone to a football game in four buses,
each with a capacity of 25 people. When the game is over, each
person returns independently to the bus that brought him to the
game. The time taken to reach a bus is normally distributed with
mean values of 10, 12, 15 and 18 minutes for the four buses. In each
case the standard deviation is 2 minutes. When a bus is full, it
leaves, and arrives home after a drive that is normally distributed
with a mean of 80 and a standard deviation of 5 minutes. Begin a
simulation from the time the game finishes and find the time at
which the last bus arrives home. (40)

44. A small grocery store consists of three aisles and a single

checkout counter. Shoppers arrive at the store in a Poisson pattern
with a mean inter-arrival time of 75 seconds. After arriving, each
customer takes a basket and may then go down one or more of the
three aisles, selecting items as he proceeds.

16
The probability of going down any particular aisle, the time
required to shop an aisle, and the number of items put into the
basket in each aisle are as follows:
Probability of going Seconds to go Number of
Aisle down the aisle down the aisle items selected
1 0.75 120 + 60 3+1
2 0.55 150 + 30 4+1
3 0.82 120 + 45 5+1
When shopping is complete, customers queue up first-come, first-
served at the checkout counter. Here, each customer puts an
additional 2 + 1 “impulse items” into the basket. The checkout time
is 3 seconds per item in the basket. After checking out, the customer
leaves the basket and departs. Simulate the store for an 8-hour day.
Measure the utilization of the checkout counter and the maximum
length here. Determine the maximum number of baskets in use. (41)

45. The IATs for parts needing processing is defined as follows:

InterArrival Time Proportion
(seconds)
10 - 20 0.20
20 - 30 0.30
30 - 40 0.50
There are three standard types of part: A, B and C. The proportion
of each part, and the mean and standard deviation of the normally
distributed processing times, are as follows:
Part type Proportion Mean Standard Deviation
(seconds) (seconds)
A 0.5 30 3
B 0.3 40 4
C 0.2 50 7

17
Compare the waiting times for parts when one, two or three
machines work in parallel, where each machine processes any type
of part, one part at a time. Simulate for one hour. (42)

46. Simulate a single-server queuing situation, where customers

arrive every 60 seconds. The service times of successive customers
are uniform random variables between 10 and 90 seconds. Simulate
50 cycles, where one cycle begins and ends with an idle period for
the server. For each cycle, calculate the sum of individual waiting
times for all customers served during the cycle and the number of
customers served during the cycle. (42)

47. Change Model49 as follows: The orders arriving no longer refer

to only one unit. Only 40 percent concern a single unit. 35 percent
of the orders are for two units and 25 percent of them are for three
units. The accounts due will hence refer to different amounts and
must be represented by a parameter, since it is possible that e.g. a
claim regarding 3 units and hence for a total of $ 75 K ($75,000)
will be paid later than a claim regarding 1 unit and for $ 25 K, even
though the order for the 1 unit arrived later. The program shall, just
like Model49, produce a quarterly report, but also a graph over the
development of cash. Run the program five times to study, if there
is any substantial risk of a negative cash balance. (42)

48. Modify Model45 in the following ways: Customers arrive

following an exponential distribution, with an average InterArrival
Time of 2 minutes at 10 a.m., of 1.5 minutes at 3 p.m. and of 1
minute at closing time 6 p.m. (8 hours after start), i.e., more and
more customers arrive as they day goes on. Normally seven people
work full-time (eight hours a day) in the store. Each salesperson has
a separate waiting line. The store is closed in a ‘proper manner’ as
in Model30.
18
Simulation is to be used for deciding whether to hire a part-timer to
work towards the end of the afternoon. The part-timer is paid $ 10
per hour. All other factors are as in Model45. Carry out a
simulation with a part-timer, starting at what you would consider to
be a suitable time. Make a comparison between two starting times
based on the profits from the operations. (43)

49. Customers arrive at a bank in an exponential pattern at a mean

rate of 200 per hour. Eight teller windows are open at the bank at all
times. A separate waiting line forms ahead of each teller. If a teller
is free when a customer enters the bank, the customer immediately
goes to that teller. Otherwise, the customer joins whichever waiting
line is the shortest. Customers then remain in that line. The various
types of business that customers transact fall into five categories. In
each category service time is exponentially distributed. The relative
frequencies of these categories and the corresponding mean service
time requirements, are as follows:
Category of Relative Mean service
business frequency time (seconds)
1 0.10 45
2 0.19 75
3 0.32 100
4 0.24 150
5 0.15 300

Another bank in the vicinity has introduced a ‘Quick line’ queuing

system. In this system, customers entering the bank form a single
line. Whenever a teller becomes available, the customer at the head
of the line goes to that teller. Build a model, which gathers waiting
line information for the bank's operation, both as it currently exists,
and for a change to a Quick line system. Run the model for a five-
hour day in each case. (43)
19
50. Orders arrive at a factory at the rate of 20 every hour, following
an exponential distribution. There are orders for three types of
goods: 40 percent are of type A, 35 of type B and 25 of type C. They
each require processing in a special machine. Processing in the A
machine takes 6 + 4 minutes, in the B machine 9 + 3 minutes and in
the C machine 12 + 5 minutes. Simulate for 50 hours to study to
what extent orders are held up at the various machines. (45)

51. Modify Model96, so that it provides data on how many times

the service of Joe has been interrupted by other transactions. (46)

52. A computer center substation has two workstations. On arrival,

users choose the workstation with the shortest waiting line. Students
arrive at a rate of one every 8 + 2 minutes. They can be interrupted
by professors, who arrive at a rate of one every 12 + 2 minutes.
There is one systems analyst, who can interrupt anyone, but
students are interrupted before professors. The systems analyst
spends 6 + 4 minutes on the workstation and arrives every 20 + 5
minutes. Professors and students spend 4 + 2 minutes on the
workstation. If a person is interrupted, the person joins the head of
the queue and resumes service as soon as possible.

Simulate 50 professor or analyst jobs. Estimate the mean length of

the student waiting line and the average waiting time at the two
terminals. (46)

53. Motorists wishing to cross the strait between the mainland and a
small island have to use a ferry. The ferry is moored on the
mainland overnight and starts work promptly each day at 7 a.m. It
shuttles to and fro between the mainland and the island until
approximately 10 p.m. when the service closes down for the night.
The ferry has a capacity limit of six cars.
20
When the ferry arrives at a quay, the cars on the ferry are driven off,
and then any waiting cars (up to the maximum of six) are driven on.

When the ferry is fully loaded, or the quay queue is empty, the ferry
leaves that side of the strait and starts another crossing, When the
ferry has completed a round trip and deposited any passengers on
the mainland, the captain checks the time. If it is 9.45 p.m. or later,
he closes down the service for the night.

Both on the mainland and the island, cars arrive on average at a rate
of nine per hour, following an exponential distribution. Crossing
times follow the normal distribution with a mean of 8 minutes and a
standard deviation of 0.5. It takes a car 0.5 minutes to drive on or
drive off the ferry.

Run the model for one working day on which there are three cars
waiting on the mainland and one on the island at 7.00 a.m. Produce
a table of the number of cars on the ferry at each crossing. (47)

54. In a retail store, the daily demand for a given item is normally
distributed with a mean and standard deviation of 10 and 2 units,
respectively. Whenever the retailer's stock-on-hand drops to or
beneath a predetermined point, called the re-order point, he places a
stock-replenishment order with his supplier regarding a specific
replenishment amount, called the re-order quantity, provided there
is no unfilled order.

The replenishment arrives at the retail store 6 to 10 days after the

order was placed. This time between placing the replenishment
order and the arrival of the goods at the retail store is termed lead-
time. The lead-time distribution is as follows:

21
Lead time Relative
(days) frequency
6 0.05
7 0.25
8 0.30
9 0.22
10 0.18

Demand that arises when the retailers is out of stock is lost, since
customers whose demand cannot be satisfied immediately go
elsewhere to carry out their business. Initially there are 100 units in
stock. The retailer wants to know how his experience of stocking
the item will vary, depending on where he sets the re-order point.

Build a model for the retailer's situation. Design the model to obtain
a table over ‘lost daily sales’ and ‘number of units carried in
inventory’. Run the model to estimate these two distributions, when
the re-order quantity is 100 and the reorder point is 80. Shut off the
model after a simulation of 1000 days. Ignore any weekend
problems, by assuming that the retailer does business seven days a
week and that a replenishment order continues to proceed also on
Saturdays and Sundays. (47)

55. A bus is scheduled to arrive at a bus stop every 30 minutes, but

it can be up to 3 minutes late. All delays between 0 and 3 minutes
are equally likely. If a bus is late or not is completely independent
of whether or not the preceding bus was late. A delay will not affect
succeeding buses either. People arrive at the bus stop according to
the (negative) exponential distribution with an average of 12
persons each 30 minutes. The bus, which has a capacity of 50
passengers, will have between 20 and 50 passengers on board when
it arrives, with every number between 20 and 50 equally likely.
22
Between 3 and 7 persons (with 3, 4, 5, 6 and 7 equally likely) get
off at the bus stop through the only door for exit from the bus.

After this, as many as possible of the waiting passengers will get on

the bus. Those who cannot get space on this bus will walk away and
will not return to the bus stop. It takes a time of 4 + 3 seconds for a
passenger to get off and a time of 8 + 4 seconds to get on. Boarding
of the bus cannot start until everyone who wants to get off the bus
has done so. The passengers get off, and get on, the bus, one at a
time. A person, who arrives while the bus remains at the bus stop,
can get on, provided there is space left. The bus leaves as soon as
either every one waiting has boarded the bus or the bus is full.

Write a simulation program to model this. The program should

provide information on the distribution of passenger waiting time,
divided into intervals of five minutes, and how many of the waiting
passengers did not get on the bus. Simulate 25 bus departures from
this bus stop. (47)

56. Alter Model44, so that it produces statistics referring to the

steady state conditions. First write one version of the program to
determine when steady state conditions begin, by allowing a study
of the changes in the length of waiting for new books to arrive,
measured month by month (of 25 working days). Next write a
version producing steady state statistics, simulating a year. (48)

57. Modify Model60, so that it, for one order quantity, produces a
report in a matrix table for each quarter (every 62.5 working days)
on direct sales, delayed sales, inventory costs, ordering costs and
profits. For simplicity, we use the size of the storage at the end of
the quarter as the average contents of the storage of the quarter. (49)

23
58. An oil storage depot distributes three grades of fuel: home
heating oil, light industrial fuel oil, and diesel fuel for road vehicles.
There is one pump for each grade of fuel, and the demand for each
is the same. Orders for fuel oil vary between 3000 and 5000 gallons,
in increments of 10 gallons, evenly distributed. The time required to
fill fuel trucks is a function of the pumping rate (6, 5 and 7 minutes
per 1000 gallons respectively), the order size, the number of
vehicles in the depot (30 seconds extra per vehicle) and the setup
time, which is a fixed time of two minutes. The depot can hold a
maximum of twelve trucks. The mean arrival rate of trucks is 18
minutes, modified by the following data:

Frequency 0.20 0.40 0.25 0.15

Ratio to mean 0.45 1.0 1.5 2.0
Simulate the operation of the oil storage depot for 5 days to find the
distribution of transit times of trucks and the total quantity of fuel
sold each day.

59. A super market has three checkout counters at its northern exit
and two checkout counters at its southern exit. Customers arrive
every 0.7 minutes, following an exponential distribution. 45 per
cent of them go to the southern exit. At each exit, the customers go
to the checkout counter which is free and, if all are busy, to the one
with the shortest waiting line. They here spend 2.6 – 3.4 minutes.
The manager wants to determine how the longest of all the waiting
lines at the five counters varies over the day. He wants a report on
this every ten minutes during the eight hour day.

60. Modify Model44 by incorporating the possibility of influencing

sales by changing the price, assuming that the number of students
per year q asking for the book is a function of price, q = ap-b, where
p is the price, a is a scale factor and b the price elasticity.
24
Let a = 30000 and b = 1.5 (implying that sales go up by 1.5 percent,
if price goes down by 1 percent). Simulate both for a price of 25
and a price of 30. All other factors are the same as in Model44.

61. Extend Model60 to allow for the simultaneous determination of

the best combination of order quantity and replenishing level. We
shall here limit ourselves to the simple case that each of these
decision values can be 10, 15 or 20, i.e. we have a total of 9
combination cases. One way is to have the number of the case as the
decision variable and two functions that relate the number of the
case to the value of the order quantity and the replenishing level.

62. The InterArrival Times of cars approaching a gas station and the
service times are distributed as follows:
Interarrival Time Cumulative Service time Cumulative
(in seconds) frequency (in seconds) frequency
0 0.00 100 0.00
100 0.25 200 0.06
200 0.48 300 0.21
300 0.69 400 0.48
400 0.81 500 0.77
500 0.90 600 0.93
600 1.00 700 1.00

A car stops for service only if the number of cars already waiting
for service is less than or equal to the number of cars currently
being served. Cars that do not stop go to another gas station. The
gas station is open from 7 a.m. until 7 p.m. Cars arriving later than 7
p.m. are not allowed in, but cars already waiting at 7 p.m. are served
before the attendants leave for the night. It is estimated that the
profit per car served averages $2.00, excluding attendants' salaries
and other fixed costs.
25
Attendants earn $6.50 per hour and are paid only for a 12-hour day,
even if they stay beyond 7 p.m. Other fixed costs amount to $75 per
day. The station's owner wants to determine how many attendants
(1 - 4) he should hire to maximize his daily profit. Build a model of
the operation of the station. Simulate with each configuration for
five different days.

63. A certain machine uses a type of part that is subject to periodic

failure. Whenever the part fails, the machine must be turned off.
The failed part is then removed, a good spare part is installed, if
available, or as soon as one becomes available, and the machine is
turned on again. Failed parts can be repaired and reused. The
lifetime of a part is normally distributed, with a mean of 350 hours
and a standard deviation of 70 hours. It takes 4 hours to remove a
failed part from the machine. The time required to install a
replacement part is 6 hours. Repair time for a failed part is normally
distributed, with mean of 8 and a standard deviation of 0.5 hours.

The machine operator is responsible for removing a failed part from

the machine, and installing a replacement part in its place. There is
one repairer who is responsible for repairing failed parts. The
repairer's duties also include repair of items routed from another
source. These other items arrive according to a negative exponential
distribution with a mean arrival time of 9 hours. Their service-time
requirement is 8 + 4 hours. These other items have a higher repair
priority than the failed parts used in this particular machine.

Build a model for this system, to estimate the fractional utilization

of the machine as a function of the number of spare parts provided
in the system. Study the system under the alternative assumptions
that zero, one and two good spare parts are provided initially. Run
each simulation for the equivalent of five years, assuming 40-hour
work weeks.
26
64. We shall here modify Model07 to allow for reneging.
Customers leave the waiting line if they have waited 40 minutes
without being served. To handle this in aGPSS, a SPLIT block is
used. With this block we divide each customer into two fictitious
"half-persons"; the left half goes as the normal customer into the
waiting line; the right half stays outside of this waiting line, waiting
to renege after a certain time. When it is time for the left half to
leave the waiting line and go into service, it checks if the right half
has already reneged, in which case it just leaves the system.

If the customer instead gets served before reneging time, it sends a

signal to the right half that it has left the waiting line. When it is
time for the right part to renege, it checks, if the left half has already
gone into service, and if this is true, it just leaves the system; else it
signals to the left half that it has reneged and then leaves.

The question is how the two halves can signal to each other. Since
several signals, all for different customers, might exist at the same
time in the system, each customer must have its own signal. Hence
we must have a parameter, e.g. p$cust that gets the values 1, 2, 3,
etc. for the different customers.

In Model89 we used Q(p$bestq) to denote the length of the AD set

for the different counters. Likewise, we can create savevalues using
this kind of indirect addressing. X(p$cust) can be used to denote
whether a customer has stopped waiting, i.e., has either reneged or
gone into service. In that case X(p$cust) = 1. Otherwise, i.e. if the
customer is still waiting, X(p$cust) keeps its initial value 0.

65. This application deals with a very simple case of simulation

based costing, but it exemplifies some of the basic ideas why
simulation based costing is important.
27
Stochastic variations as regards e.g. machine time and arrival rates
of one product can also influence the cost of other products. We
study a factory where two types of products, A products and B
products, are made in the same machine. The A products arrive
exactly every 10 seconds and require a time of exactly 5 seconds.

The B products arrive every 20 seconds on average and require on

average 10 seconds of machine time. For the B products we study
two cases. In the first case there is no uncertainty and, just like for
the A products, the sampled times are the average times. In the
second case we have uncertainty, in which case the sampled times
follow the negative exponential distribution of the average times.

We can study the two cases by inputting either 1 for certainty or 0

for uncertainty. For each case we simulate for 30000 seconds and
print the number of A and B products and the cost of each product.
The cost per unit is equal to the average time used in the machine
times the cost of one second of machine time, which in turn is
calculated as 100000 divided by the total number of machine
seconds used.

66. A regional airline flying Dash Q400 planes with a capacity of

75 seats is contemplating its policy as regards how many days in
advance of the flight they shall require passengers to book to get a
low air fare.

It sets up the policy 100 days in advance for a flight. It knows that
the expected number of passengers wanting to fly on a specific day
will increase dependent on the distance from this start date. The
closer one comes to the flight on day 100, the more passengers will
want to fly. The average time between two bookings is estimated to
be 1-CL/200, where CL is the number of days since the simulation
28
start 100 days before the flight. A great variation, day from day, is
however likely, so that the sampled date between two bookings is
estimated to follow the negative exponential distribution.

60 percent of the passengers want to fly economy class. They pay

$75 per passenger. If no cheap flights are available, they turn to
another airline. 3 percent of the economy passengers do not show
up, but they cannot rebook. Of the business class passengers, each
paying $ 150, 20 percent do not come to this flight, but can rebook
to another flight and their payment is not included in the income of
this flight. If there are more than 75 passengers showing up, the
airline has to pay some passengers $300 to get them to give up their
seats.

The airline’s main decision is the last date of booking flights at the
economy price, since one does not want to turn away business
passengers. Calculate and print the profit as income minus total
costs.

67. A production shop is comprised of six different groups of

machines, as follows:

Group number Machines in group

Kind Number

1 Casting units 14
2 Lathes 5
3 Planers 4
4 Drill presses 8
5 Shapers 10
6 Polishing machines 4

29
Three different types of job, Type 1, Type 2 and Type 3, move
through the production shop. Each job-type requires that operations
be performed in a specified sequence, as shown below.

Job type Machine sequence Mean operation time (min.)

Casting unit 125
1 Planer 35
Lathe 20
Polishing machine 60
Shaper 105
2 Drill press 90
Lathe 65
Casting unit 235
Shaper 250
3 Drill press 50
Planer 30
Polishing machine 25

The operation times are all exponentially distributed, with the mean
time given above.

Jobs arrive at the shop in a Poisson stream at a mean rate of 50 jobs

per 8-hour day. 24 percent of the jobs are of Type 1, 44 percent of
Type 2 and the rest of Type 3.

Simulate the production shop for five 40 hour weeks. At the end of
each week, print the average time in shop for each job type for this
week.

30
68. The establishment of optimal equipment life is important in
managerial economics, accounting (e.g. for determining the
depreciation rate) and finance. A company does not want to spend
too much on new expensive equipment, but, on the other hand, it
does not want to have obsolete equipment leading to high costs.
One method for determining optimal equipment life is the MAPI
method, which determines equipment life taking technical
development into consideration. Its critical assumption is that new
equipment leads to lower production costs than old equipment. A
machine that is new this year leads to inFact lower costs than a
machine that was new last year. Hence, a machine that is 5 years
old has 5*inFact higher costs than a machine that is new. Hence, for
a machine with the age cl, the extra cost, xcost, this year, due to not
having the latest machine model, is cl*inFact.
One next calculates the present value, prVal, of this extra cost as
xcost*pvFact, where pvFact is the present value of getting one
dollar y years from now. When we have an annual interest rate
irate, we can, using discounting, define the present value factor
pvFact as 1/(1+irate)^y. For a machine that is kept y years, this
value is next summed up for all years, going from 1 to y. To this
total sum of present values sumPrV of the extra cost of having the
equipment, we also add the initial investment payment invest.

On the basis of this total cost of having the machine for y years, we
next calculate the annual cost annui by multiplying this sumPrV
with the annuity factor anFact needed to distribute the cost over y
years. This calculation of the annual cost makes it possible to
compare machines with different lengths of life. The annuity factor
anFact is 1 divided by the sum of present value factor smPvFa, i.e.
anFact = 1/smPvFa, where smPvFa, in turn, is (1-pvFact)/irate.
31
In order to be able to investigate also the stochastic case when the
inferiority factor is not fixed, we assume that it varies following the
normal distribution. We hence set infact = infaAv+fn$snorm*stDev.
In order to have the lowest cost as a maximum we use a minus sign
for the annuity.

Using EXPERI, we can calculate the annual cost of having a

machine differently many years, first for the case without stochastic
variations, setting stDev to 0, then for the case of stochastic
variations with stDev > 0. We input the interest rate in the usual
terms, e.g. as 10 (percent), which would lead to irate = 0.1. We also
input invest, infaAv and stDev.

69. Samuelson and Bazerman (1985) provided a bidding problem

in the energy sector, for which aGPSS can provide a powerful and
illustrative solution. The problem is as follows: Company A is
considering acquiring company T by means of a tender offer in cash
for 100 percent of T’s shares. The value of T depends on the
outcome of an oil exploration project that T is currently
undertaking. If the exploration fails completely, T will be worth $0
per share, but in the case of a complete success, a share of T under
current management will be worth $100. All values between $0 and
$100 are equally likely.

Regardless of outcome, T will be worth 50 % more under A’s

management than under its current management. The price of the
offer to T must be determined before A knows the outcome of the
drilling, but T will know the outcome when deciding whether or not
to accept A’s offer. T is expected to accept any offer from A that is
greater than the per share value of T under current management.
What price should A offer?

32
A typical student view is that the expected value of T to its current
owner is $ 50 and that is worth 50 % more to A and that A should
bid in the interval of $ 50 to $ 75. Investigate what would be a
suitable price, by building a model that calculates the expected
value for A for a given price offer. Finally, use EXPERI to
determine the optimal price bid to be made by A.

70. In Lesson 17 we studied the WAITIF block. This block can as A

operand use only a server name and as B operand only a symbol (U,
NU, E, NE, F or NF). In contrast to the IF block, it is not possible to
use SNAs or constants as A and B operands. The reason for this is
that the processor should make a specific check every time that it is
possible that a change has occurred that could affect the WAITIF
block. With a WAITIF block referring only to a specific server, the
required amount of such checking is very small. If the WAITIF
block instead could refer to any SNA, such checking would have to
take occur in connection with every event. This could possibly lead
to extremely long running times.

The lack of a specific block allowing for waiting with regard to any
SNAs implies that one in some cases must add a logical SEIZE
block to allow for waiting. We shall exemplify this by modifying
Model44. Instead of every customer buying just one item, a
customer buys 1, 2 or 3 items; 40 percent buy 1 item, 35 percent 2
items and 25 percent 3 items. This number is thus sampled from a
random function and then stored in a parameter, e.g. called
P$number. If this number of items is not in stock at arrival, the
buyer will not buy any items directly, but go into a waiting line at
the address NODIR until this number is available.

We here indicate that there is a shortage by putting a logical facility,

e.g. called SHORT, into use. The customer will now wait if, and as
long as, this facility SHORT is in use.
33
This shortage can possibly be removed by the arrival of goods in the
reordering segment. A customer can then leave the WAITIF block
and try to take P$number units out of the replenished stocks.

Since there might be many customers waiting in the WAITIF block,

it might very well happen that stocks are depleted again before all
waiting customers have been able to get their P$number units.
Hence, when each customer leaves the WAITIF block, we check
that there are at least P$number units in stock.

If this is not the case, there is a new shortage and the customer is
sent back, e.g. to an address LACK, where he again puts the logical
facility SHORT into use. The customer then goes back into the
waiting line. In order to assure that he remains at the first place in
the waiting line, he is assigned the highest possible priority, e.g.
through a block LET Priority=250-CL, before going back to LACK.
He will now have to wait at LACK for another delivery of goods.

The only change in the reordering segment compared to Model44 is

that the facility SHORT is released on arrival of new products as a
signal to the WAITIF block in the customer segment. Besides the
new assumptions above, all other assumptions behind Model44 hold
also for this modified model.

71. We shall look at a problem that in decision theory is sometimes

called the “partner selection” problem. A young man will in his
search for a partner meet x young ladies. He gives each lady a score.
We assume that this score, with an average of 1000, follows a
statistical distribution, e.g. the negative exponential, the normal or
the uniform distribution. If he does not choose her, he cannot go
back to her, but proceeds to the next lady. He knows, however, after
meeting each lady what the highest score this far is.

34
One decision rule is now that he, after meeting y ladies, determines
that the critical score is the highest score this far. He then selects
the next lady that he gives a higher score than this critical value. If
he after having met all x ladies, has not found anyone with a higher
score than the critical value, he will remain single, leading to a
score of 0. What is the optimal value of y for a given value of x?
Write a program to try to answer this question for different values
of x, choosing one of the three distributions mentioned above.

72. Some passengers arriving at the central railway station in

Magdeburg, Germany, want to get home by cab. The driver is told
the destination. For this, a code 1 – 8 is used. The frequency of
travelers and the distance to go is as follows:
Destination Percent Time distance
1 20 17
2 11 13
3 5 25
4 7 17
5 15 16
6 5 18
7 16 13
8 21 17
The average distance in time is the time distance given above. The
sampled time is then this average time + 5 minutes. After bringing
the passengers to their homes, the taxis all return to the station. The
time back to the station is equivalent to the time from the station.
Passengers arrive with an IAT of 2.8 + 0.5 minutes. We simulate for
100 passengers arriving at their home. The simulation should be
used to investigate how waiting times for passengers change if the
number of cabs increases from ten to eleven to twelve.

35
73. We shall in this model use aGPSS for a type of very simple
continuous simulation. Even if aGPSS is a tool for discrete events
simulation, one can in certain cases, e.g. when we make forecasts
over many years, but at most a century, and look at each year as a
whole, use aGPSS for a basic type of continuous simulation.

Continuous simulation mainly deals with differential equations. We

shall base our model on the following solution technique. We shall
here use y’= dy/dt for three different types of equation:

1. y’ = f(y) = ky for exponential growth

2. y’ = f(y) = k(ymax – y) for modified exponential growth

3. y’ = f(y) = ky(ymax – y) for the logistic curve, which can

be seen as a combination of equations 1 and 2.

We next approximate the differential equation with a difference

equation, i.e. we approximate dy/dt with y/t to get y/t = f(y).
For the case that t = 1, e.g. when we step a whole year at a time,
we set y=f(y). With yj+1 = yj + f(y), and starting with a value y0, we
can next, step by step, calculate a value yj for any year j.

We can next apply these ideas to a model that is meant to forecast

the sales of a new equipment to monitor the heating of one-family
houses in a country with initially 3 million house-holds. The
number of households is growing annually at a certain rate,
100*HoldGr percent

Today 40 percent of the households live in a one-family house.

Most, but a maximum of 90 percent, of the remaining households
will want to move from an apartment to such a house.
36
Every year 100*HousGr percent of these remaining household
move to their own house This development is thus in line with the
modified exponential growth.

Now, a new type of computer system has appeared on the market

that by a web-based app can monitor several functions, like heating,
in the house. When we start the simulation, there are only 2000 such
computers in use.

The growth of these computers follows the logistic curve, implying

that, when the sales start, the usage increases almost exponentially,
but after some years, as the market becomes more saturated, the
development is more in line with the modified exponential growth.
In function 3, ky(ymax – y), we set k to 0.0005 and ymax to 80 percent
of number of single homes.

We shall make three graphs, one for development of the number of

households, one for the number of single homes and one for the
ownership of these new computers.

We run the simulation for 50 years.

In order to see how each of these three functions work, we first

input both of HoldPr and HousPr as 0 to look at the third graph
showing the logistic curve. Next we input HousPr as 3, but keep the
value of HoldPr as 0, and look at the second graph, showing the
modified exponential curve, which illustrates how the growth in
house ownership levels off. Finally, we input HoldPr and HousPr
as 1, to see how ownership in computers develops also because of
changes in the number of households and the number of one-family
homes.

37
74. At a certain library, anyone wanting a book must present a
checkout slip to a clerk working behind the checkout desk. This
clerk then goes into the stacks to find the book and returns to the
desk with it. Build a model of the checkout procedure under the
following conditions:

1. People wanting to check out books arrive at the desk in a

Poisson stream at a mean rate of 30 per hour.
2. Each person wants to check out just one book and the
wanted book is always available.
3. The number of clerks working at the desk is a decision
variable in the model.
4. Whenever a clerk becomes available, the clerk is willing to
pick up checkout slips from as many as four people at a
time, if that many are waiting for service.
5. The time required to pick up a checkout slip is negligible.
6. A one-way trip between the checkout desk and the stacks
requires 1 + 0.5 minutes.
7. The time to find one, two, three, or four books is normally
distributed with mean values of 3, 6, 9 and 12 minutes,
respectively, and with a standard deviation of 20 percent of
the mean.
8. When a clerk returns from the stacks, the time required to
complete the rest of the checkout procedure is 2 + 1 minutes
per person.
Estimate the distribution of time spent by each person at the
checkout desk, the average number of slips that a clerk picks up and
the utilization of the clerks. Gather this information for the
alternatives of 3, 4, and 5 clerks. For each alternative, simulate until
100 persons have been completely served.
38
75. Extend the model of Exercise 63, which dealt with a machine
that used a single part, which we now call the A part. We now allow
the same machine to use also another type of part, the B part. Just
like the A parts, the B parts are subject to periodic failure.
Whenever the A or B part in use in the machine fails, the machine
must be turned off. The failed part is then removed, a good spare is
installed, if available, or as soon as it becomes available, and the
machine is turned on again. Note that, when one part fails, the time
until the other one fails changes. If, for example, initially the A part
had 330 h. left and the B part 415 h. left, then when the A part fails
at time 330, the B part will only have 85 h. left before failure.

Like the A parts, the B parts can be repaired and then reused. The
lifetime of B parts is normally distributed, with a mean of 450 and a
standard deviation of 90 hours. It takes 4 hours to remove a failed B
part from the machine and 6 hours to install a replacement part. The
time required to repair a B part follows the distribution below.

Repair time in hours Probability

5 0
6 0.22
7 0.35
8 0.26
9 0.17

Except for the added feature of the B parts, the conditions under
which the machine operates are identical to those in Exercise 63.
Use the modified model to estimate the utilization of the machine as
a function of the number of A and B spare parts in the system.
Simulate the system for each combination in which 0, 1 or 2 spares
of each type of parts are provided. Run each simulation for 5 years
of 40 weeks each.

39
76. Up to four barbers work in a barber shop. Intervals between
successive customers are exponential, with an average of one every
10 minutes during time 0 - 120 and 225 - 390 (minutes after
opening) and one every 5 minutes during time 120 - 225. No
customers are generated after 390, but customers already in the
system are served until ten hours after opening time. Lunch hour is
as follows: The first barber goes to lunch at time 60 for a lunch of
30 minutes. The other barbers can then go to lunch, also for 30
minutes, one after the other. A barber goes for lunch first when he is
idle, e.g. when he has finished serving the customer.

There are 3 waiting chairs. If an entering customer cannot be served

at once he will leave immediately with the following probabilities:
0.1 if 3 waiting chairs are empty; 0.3 if at least 2 waiting chairs are
empty; 0.5 if one waiting chair is empty and 1 if no such chair is
empty. After having waited 15 minutes, a customer leaves unserved
with a probability of 0.5. Service time is normal with a mean of 20
minutes and a standard deviation of 4 minutes. Determine the
number of customers served per day for 1, 2, 3 or 4 barbers.

77. Tankers arrive at a harbor every eight hours on the average,

exponentially distributed. The first tanker arrives at time 0. They
then discharge their cargo into shore tanks. When a shore tank is
full, or nearly so, i.e. has less than 20 K (20,000) tons capacity free,
its contents are automatically transferred to the refinery. While this
transfer is taking place, a shore tank may not be filled by a tanker.

The time to set up a pump is 0.5 hours. The pumping rate is 1 K

tons per hour and the discharge rate is 4 K tons per hour. Tanker
loads are either 15, 20 or 25 K tons, all equally likely. All shore
tanks have a capacity of 70 K tons.

40
Run the simulation for 1000 continuous hours with the five shore
tanks. Initially, two shore tanks are empty and free; one is currently
discharging and will be free after 8 hours, and the other two are
currently being loaded and will be freed after 12 hours (with 45 K
tons contents) and 3.5 hours (with 25 K tons contents) respectively.

78. A port in Africa is used to load tankers with crude oil for
overwater shipment. The port has facilities for including as many as
three tankers simultaneously. The tankers, which arrive at the port
every 11 + 7 hours, are of three different types. The relative
frequency of the three types and their loading-times are as follows:
Type Relative frequency Load time (hr.)
1 0.25 18 + 2
2 0.55 24 + 3
3 0.20 36 + 4
There is one tug at the port. Tankers of all types require the services
of this tug to move into a berth, and later to move out of the berth.
The area experiences frequent storms, and no berthing or deberthing
can take place when a storm is in progress. When storms occur, they
last 4 + 2 hours. The time between the end of one storm and the
start of the next one follows the exponential distribution with a
mean of 48 hours. When a tug is available and no storm is in
progress, berthing or de-berthing takes 1 hour.

A shipper is considering bidding on a contract to transport oil to the

UK. He has determined that five tankers are needed for this task.
They enter the port 2 days apart and then require 21 + 3 hours to
load oil at the port. After loading and de-berthing, they would travel
to the UK, offload the oil, return to the port for reloading, etc. Their
round-trip travel time, including offloading, would be 240 + 24
hours.

41
Simulate the operation of the port if this plan goes through for one
year. Measure the distribution of the in-port residence time of the
proposed additional tankers, as well as of the three types of tankers
that already use the port.

79. A small elevator has space for only one person and goes
between two levels. People wanting to go up arrive at the rate of
one every 10 minutes, following an exponential distribution, while
people wanting to go down arrive only every 20 minutes (also
following an exponential distribution), since many prefer to walk
down the stairs. The elevator takes 2 minutes in each direction.

On the ground-level there is only a button for going up and on the

top-level only a button for going down. If the elevator is on the top-
level and someone on the ground-level is pressing the button for
going up and there is no one waiting to go down, the elevator goes
down empty.

Likewise, the elevator goes up empty, if someone on the top- level

is pressing the go down button and the elevator is on the ground-
level and no one wants to go up. In the morning the elevator is on
the ground-level. Simulate a day of 20 hours to determine the
utilization of the elevator.

80. A wholesaler has a depot, where goods are delivered by truck.

The goods are brought to the customers by van. The depot opens at
time 0 and closes at time 360 (time in minutes). After time 360, no
additional vehicles (trucks or vans) are allowed into the depot, but
the trucks and vans already in the depot finish their tasks. The time
between two trucks or two vans arriving follows the exponential
distribution with the average times of 30 minutes for trucks and 5
minutes for vans.

42
The serving of the trucks and vans takes place on the second floor
in the depot. An elevator is used to get there. This elevator has
space for one truck or one van. The time to transport a vehicle up or
down varies between 1.5 and 2.5 minutes, equally distributed. This
time includes driving the vehicle in and out of the elevator. If the
elevator travels empty, it takes exactly 1 minute up or down.

The servicing of the trucks or vans takes place at special loading

bays. There are at present two bays for trucks and three for vans.
(Each kind of vehicle must use its own type of bay.) The servicing
is carried out manually by a team of workers. A team can service
either a truck or a van. The time per truck follows the normal
distribution with an expected value of 30 and a standard deviation
of 5 minutes. The time for a van is also normally distributed with an
expected value of 10 and a standard deviation of 3 minutes. The
teams work all day without a lunch break.

The following rules are used to control the traffic: The trucks have
priority over vans, both when waiting at the elevator and at the
bays. The vehicles wait in two lines at the elevator, one for trucks,
one for vans, until there is a bay free for the particular type of
vehicle. They are then brought upstairs to the bay. (This time is
included in the 1.5 - 2.5 minutes for the elevator.) The vehicle
might then have to wait until a team is free to give service. The
elevator goes up or down empty if there is some vehicle waiting on
the other floor.

Management has observed considerable waiting times, in particular

for the vans. In order to reduce these times, management
contemplates an increase in the number of teams from 3 to 4 and an
increase in the number of van bays from 3 to 4. Simulate the effect
of these proposals! (Hint: Do exercise 79 above first.)
43
81. A new type of IT-exhibition is planned to be located in the
center of Stockholm, the capital of Sweden. The exhibition is to be
open 7 days a week, but the number of visitors will depend on the
day of the week. The attendance on week-days will only be 60 per
cent of average attendance, while on Saturdays and Sundays
expected attendance is 200 percent of average attendance. The
average attendance will be dependent on the entrance fee. Only four
different entrance fees are contemplated: 20 SEK, 40 SEK, 50 SEK
and 100 SEK. For these fees the average number of visitors per day
is expected to be as follows: 20 SEK: 370; 40 SEK: 330; 50 SEK:
300 and 100 SEK: 230 visitors.

The main entrance, towards a busy street, will be open each day
between 10.00 a.m. and 4.00 p.m. During this time, visitors are
expected to arrive in a random fashion with InterArrival Times
distributed according to the negative exponential distribution. At
4.00 p.m. this entrance door will be locked, but those already inside
will allowed to stay until they are content and all attractions will run
until the last visitor has left the exhibition.

In case there are already 120 persons in the real exhibition area, no
new visitors will be allowed to enter this area through an inner
entrance, where visitors pay the entrance fee. They can then wait
e.g. in a cafeteria inside of the entrance to the street, but outside of
the actual exhibition area.

A visitor can either follow a guided tour or walk on his/her own. 25

per cent of the visitors are expected to go on their own, regardless
of when a guided tour begins. Each guided tour runs with exactly 10
visitors. A guided tour will normally start first when 10 visitors
have gathered for such a tour. There will be only one guided tour at
a time.
44
In case there are already 10 persons waiting for a guided tour to
start, other newly arrived visitors, who originally had wanted to
follow a guided tour, will go on their own.

The guided tour starts with a walk through the Hall of Introduction,
which takes exactly 10 minutes. For those who are on their own this
visit to the Hall of Introduction will take between 3 and 11 minutes,
with all times equally likely.

Visitors next come to a hall with demonstration computers. There

are five such computers. Each visitor uses one such computer and
runs through a multimedia program. Those who follow the guided
tour get advice from the guide and work seriously with the program.
95 per cent of them are expected to spend between 6 and 14 minutes
using this computer. For those who walk on their own the
corresponding time is assumed to be between 4 and 8 minutes. All
these times on the computers are assumed to be normally
distributed.

If some computers are being used, while others are free, a newly
arrived visitor will self-evidently go to one which is free. In case all
computers are being used, a visitor will go the computer that has the
shortest waiting line in front of it. The persons in a guided tour will
in each waiting line always be allowed to go ahead of the persons
who walk on their own. They are, however, not permitted to
interrupt someone who has already started running a program.

After all members of the guided tour are finished with the
multimedia program, the guided tour is over and a new one can
start. The main tour is at this stage also over for the persons going
on their own.

45
Before leaving the exhibition visitors will, however, spend between
2 and 6 minutes (with all times uniformly distributed) browsing in
an IT-shop. At the end of this time 60 percent of the visitors will
buy a heavily discounted computer diskette for 100 SEK.

There are two types of program diskettes, one for young people and
one for adults. 40 percent of the diskette buyers will choose the
youth diskette. It is not certain that there is a ready-made diskette
available when a customer asks for it. One person works full time
copying diskettes and a customer will have to wait until there is a
copy of the desired diskette available. After this the customer leaves
the exhibition.

The person producing the diskettes starts with this work at 10.00
a.m. and follows the principle that he will at each instance produce
the diskette of which there at the moment is the smallest stock. In
the case of equally large stocks, he will produce the adult diskette. It
takes him between 2 and 4 minutes (uniformly distributed) to
produce a diskette, independently of the type. When he has finished
with one diskette he starts to produce the next one. He continues in
this way until the last visitor has left the exhibition. Diskettes
remaining at the end of the day are sent to another shop. Hence,
there are no diskettes of any type at the start of the day.

The planners of the exhibition wish to simulate one day at this

exhibition. They want to have a program which from the keyboard
can read the day of the week (e.g. with 1 for Monday and 7 for
Sunday), as well as the entrance fee. They want the program to
provide information on the average time spent by the visitors at the
exhibition, counted from the time they enter the inner entrance, and
the total revenues from entrance fees and sales in the IT-shop.

46
82. A corporation faces the following demand on the market as
regards expected annual sales of a certain product at different prices
of the product:
Price in Annual sales
$ 1000 in units
10 310
20 110
30 60
40 40
50 30
This sales quantity per year determines the average number of days
between each order. The real times between each order will,
however, follow the (negative) exponential distribution.

30 per cent of the customers belong to a category that receives a

discount of 20 percent of the price. In case the demanded product is
in stock it will be delivered. Otherwise, the customer will wait for
delivery, but only if the corporation can promise a delivery within a
month. The corporation will give this promise only if the planned
monthly production is at least as large as the number of orders that
at this moment are waiting for deliveries. After the product has been
delivered the customers will on average pay after 45 days. Around
1/3 pay within a month, but around 20 per cent pay later than after 2
months and around 5 per cent pay later than after 3 months.

The products are made in a special machine. The corporation

decides on a fixed monthly production in units and starts the
production of a unit in a steady stream, so that the total number of
production starts corresponds to the total planned production of the
month.

47
At the start of the production of one unit, costs of $ 4,000 are
incurred. The costs also lead to an immediate corresponding
decrease in cash position. The time needed for the production of one
unit follows the normal distribution and is such that in 95 percent of
all cases the time will be between 2 and 3 hours. After this, the
products are put into inventory. The person handling the machine
receives a total of $ 4,000 per month at the end of each month.

A problem can arise as regards production due to the fact that the
machine, now fairly old, will break down on average once a month,
i.e. once every 200 hours. The actual time will, however, follow the
exponential distribution. At machine break-down the production of
the unit is stopped. The production of the unit is restarted when the
machine has been repaired. The time to repair the machine is
between 2 and 4 hours (uniformly distributed). The firm doing the
repair charges $1000 per hour.

At the start of the simulation there are 5 finished units of the

product in stock and $ 100,000 in cash. The production is run in one
shift five days a week. For the sake of simplicity we assume that a
month consists of 200 work hours.

Write a program to test various monthly production plans and to

provide a report on how this will affect the cash position and the
total contribution of this product to the profits of the corporation.

Assume that there are no other costs than repair costs as regards the
machine. The cash position should be followed in a graph, covering
all changes, while the report on the total contribution should come
at the end of the year for which the simulation is to be done. The
program should have both price and monthly production read from
the keyboard.
48
83. We study a machine shop where work is done on parts that are
purchased for $ 15,000 per batch of 10 units. The work is done in a
machine of which one can have several working in parallel. The
work is done in the following steps, in which one worker follows
the product the whole time:

(1) The worker carries out process 1 in a machine that is idle. The
machine is then used according to the following time distribution:
Minutes Percent
60 5
70 7
80 36
90 35
100 13
110 4

(2) The worker moves the product with the aid of a crane, as soon
as the crane is free, to a chemical bath and then back again to the
machine. This will in 95 per cent of the cases take between 10 and
20 minutes, with all times following the normal distribution.
(3) The worker next starts carrying out process 2 in a machine,
provided it is idle. The machine is then used according to the
following distribution:
Minutes Percent
80 2
90 7
100 28
110 33
120 24
130 6

49
(4) The worker moves the product with the aid of a crane, as soon
as the crane is free, to a container. This takes between 25 and 35
minutes, with the times uniformly distributed.

After step 4, the product is ready and the worker can go back and
start with step 1 again. In case step 2 and step 4 compete for the
crane, step 2 shall have priority. When the crane has taken care of a
product, the machine is free to be used for another product, handled
by another worker.

At the end of the production process there is a container, which can

hold 10 units of the product. Each finished product hence increases
the number of products in the container by 1. When there are 10
products in the container, the container is brought over to
inventories of finished goods, ready for sales, and one starts to fill
up another container. Orders are for full containers with a batch of
10 units. Demand, i.e. sales per month (of 30 days), for these
containers depend on price, according to a continuous demand
curve, for which we give some point values below:
Price of unit Monthly
in $ 1000 sales in units
10 410
20 140
30 80
40 60
50 40

The time between each order follows, however, the negative

exponential distribution. If there is not any container in stock when
an order arrives, 60 percent of the customers turn to a competitor,
while the remaining 40 percent wait, in a FIFO-order, until there is
a container.
50
Each worker costs $ 3,500 per month in wages + social costs, while
the machines are rented from another company within the company
group for $ 25,000 per month per machine. Each crane is rented for
$ 45,000 per month. The company runs continuously, 24 hours a
day, 30 days a month, in 4 shifts. The moment one worker leaves
his shift another worker continues the activity in which the first
worker was just engaged.

The company wishes to use the simulation model for determining a

suitable number of workers per shift, machines and cranes.
Management wants to simulate for one month (of 30 days), at the
start of which there are 3 containers of finished goods in stock.
Management also wants a simple report on the monthly profits of
this activity for a certain combination of price, workers, machines
and cranes.

84. We simulate the activities of a new computer company. We first

simulate the production of each single unit and estimate the time
between the start of the production of each such unit as the number
of days per year, 360, divided by the annual rate of production,
qprod. The value of qprod is to be read from the keyboard, e.g. as
50. Then we increase the total number of units produced this far,
totpro, by one unit and move it into the inventory.

We next calculate the cost of producing this unit. This cost will,
thanks to learning, decrease exponentially with a factor learn (set to
0.001) and the total number of produced units, so that the actual
cost per unit accost is cost0*exp(-totpro*learn). cost0 is a constant,
set to 20, representing production cost per unit, when no unit has yet
been produced. exp represents the exponential function FN$EXP.
We next increase total costs and decrease cash by unit cost acost
Cash is at simulation start set to 4000.
51
We check that cash does not become negative. In such a case the
simulation run is interrupted.

Sales activity. There are two types of sales: New sales, i.e. sales to
first time customers, and repeated sales. As regards new sales we
distinguish between two types of customer: innovators and
imitators.

For imitators we define annual demand imisal = imimar*demand,

where imimar in turn is scale*newsal*(matur-newsal). newsal is the
accumulated number of computers sold up to now to first time
customers, matur is the saturation level and scale is a constant (set
to 0.0001).

The saturation level matur is allowed to grow from an initial value

satur (set to 200) over time by a factor dgrate, which is the growth
rate per day (set to 0.00001), i.e. matur=satur*exp(t*dgrate), where
t is time in days.

demand is a linear demand function such that annual sales would be

100 units if price was 0 and 0 if price was 100. Sales are assumed to
vary continuously between these two values. Price is read from the
keyboard at the start of the simulation. This price is not changed
during the simulation run. 50 might be a suitable value on price.

For innovators we assume the following demand equation:

inosal = inomar*demand, with inomar = scal1*(matur-newsal).
scal1 is a constant set to 0.002.

We start the simulation with a low value of accumulated sales to

new customers, 10. This value 10 is also given to accumulated new
sales to all types of customer, totsal.
52
Next we generate, in one statement, each individual sale to an
imitator and, in another statement, each individual sale to an
innovator. For each type of customer, we determine the average
time between two orders as the number of days per year divided by
the annual sales, imisal or inosal, to this type of customer. The
actual sampled time between two orders is next determined by
multiplying this average time with a value sampled from negative
exponential distribution.

As regards potential new sales, we first check that accumulated

sales to new customers, newsal, do not exceed the saturation level,
matur. In such a case we terminate the transaction immediately.

As regards repeated sales we assume that 80 percent of all buyers

wish to make a repeated purchase, by going to that part of the
program that deals with repeated purchases. The remaining 20 per
cent leave the system. A repeated purchase will occur after a
certain time, which is stochastic. Nobody buys a new computer
within two years = 720 days, but the time after this until a new
computer is demanded will follow an Erlang distribution with three
drawings from the negative exponential distribution and with an
average of 600 days.

In this way we simulate the arrivals of orders, both regarding new

sales and repeated sales. These orders will, however, not always
lead to an immediate sale. In case the firm does not have any
computer in stock, the customer must wait until a newly produced
computer comes into stock. The firm will then first take care of the
order from the customer, who has waited the longest. Hereby the
total accumulated sales, totsal, is increased by one unit and, in the
case the sale was to a first time customer, the accumulated sales to
such customers, newsal, is also increased by one unit. The sale has
an immediate positive effect on both cash and profits.
53
In connection with the sale, the sold unit is taken out of stock. In
case stocks thereby become 0, the firm is not producing enough.
The firm then increases the annual production rate qprod by a factor
beta, set to 0.01. If, on the other hand, the number of units in stock
on this occasion is still larger than the desired level of a buffer of 10
units, the firm decreases the production qprod by multiplying it by
1-beta

The program shall produce a graph of the development of cash over

time as well as a graph over total accumulated sales, totsal, and
accumulated sales to new customers, newsal. The program shall
also at the end of each year provide information about the revenue,
costs and profits of the year.

Both cash and profits shall be increased at the end of the year, due
to the firm obtaining a payment of interest, calculated as 5 percent
of the lowest value of cash during the year.

Simulate for a total of 5 years.

85. We study a market for identical products, e.g. wheat. There are
two producers. Demand can be described by a linear demand
function p=a-bQ, where Q=q1+q2. Both producers have the same
constant variable unit cost c.

Firm i’s profit vi =qi(p-c)=qi(a-bQ-c)=qib((a-c)/b-Q)=qib(A-Q),

for A=(a-c)/b. We write this on a new scale as Vi=vi/b=qi(A-Q).
Hence V1= q1 (A- q1- q2)= A q1-q1q1-q1 q2 and V2 = A q1-q2 q1-q2 q2.
The non-cooperative solution, found already by Cournot, is
obtained when each firm regards the competitor’s quantity as given.
The optimal quantities are then determined by

54
V1'(q1)= A - 2 q1 - q2. = 0 and V2'(q2)= A - q1 - 2q2. = 0
-> 2 q1+q2. = A = q1+2q2 -> q1 = q2
-> A-3q1=0 -> q1= q2=A/3.
We now want to simulate a market, where the two firms know
neither demand nor costs, but only their own profits and own
quantities offered in each period. In experiments, one has found
that, under certain conditions, the Cournot solution will be obtained
after some time also under these assumptions of no knowledge
about demand and costs. The question is under what general
conditions this can occur. In order to investigate this, a simulation
model is necessary, since one in contrast to costly experiments can
afford to test a great number of conditions by inexpensive
simulations.

For each of the two firms we first input two starting quantities
oneOld, oneNew, twoOld and twoNew. We also input the intercept
of the linear demand function aVal and a change factor chg. We set
the initial value of size to 1
A first single event is generated at time 0. This puts the starting
quantities of the two firms, oneOld and oneNew, into a graph. Next
we calculate the initial changes in quantity, qch1 and qch2, where
e.g. qCh1 = oneOld – OneNew. Another single event is then
generated, at time 1, which puts oneNew into the graph at time 1.

Firm 1 next makes its moves in periods 2, 4, 6, etc., while firm 2,

makes its moves in periods 3, 5, 7, etc. The parties will hence take
turns making their moves. This sequential quantity setting is
fundamental for the model, since it ensures that each party can see
the effect of one’s own latest move.

55
For firm 1’s periods we start by calculating firm 1’s old profit oldP1
based on the most recent quantities, oneOld and twoOld. We next
calculate the new quantity for firm 1, oneNew, by adding qch1, to
oneOld. On the basis of oneNew and twoOld we next calculate the
profit of firm 1 in this period. If this new profit is higher than the
previous profit, we let the new change for firm 1, qch1, be in the
same direction as the old change, namely as (oneNew-oneOld)*size.
If the new profit is lower than the previous profit, we let the next
change be in the opposite direction of the old change, namely as
(oneOld-oneNew)*size. In case of equal, i.e. unchanged profits, we
go with 50 percent probability to repeat the change and with 50
percent probability reverse the change.

Before we move into the next period, we give the value of oneNew
to oneOld. We also update the value of size by multiplying it with
the change factor chg, e.g. 0.995, so that size constantly decreases.
For firm2’s decisions in periods 3, 5, 7, etc., we have a similar
segment to determine a sequence of values on qCh2, twoNew and
twoOld.

Build a model for this Cournot process. The model should produce
a graph, by inputting every new quantity for each party when it has
been computed.

When a user runs this model, aGPSS will first ask for the values of
oneOld, oneNew, twoOld, twoNew, aval and chg. Let us exemplify
with the values 1, 2, 20, 19, 24 and 0.995. In this case, with the
initial values of q1 and q2 fairly far apart, we will after around 40
periods get a convergence on the quantity 8. In this case with aval =
A = 24, this is equivalent to the theoretical Cournot equilibrium of
24/3 = 8. If we input other values, convergence might not be so fast
and in several cases there will not be any convergence.
56
86. The issue is the administration of secretarial work in a company.
At present, the president has a secretary, called PSEC, who does
tasks only for him and no other jobs. The secretarial tasks of the
other employees are handled by three secretaries, who work in a
secretarial pool. Hence all these general tasks go to this pool and are
handled by the first secretary available in the pool.

The present situation has led to some problems. While there is often
a long waiting line for the general tasks at the secretarial pool, the
utilization of the president's secretary is not very high (less than 50
percent). A consultant has therefore suggested an arrangement,
whereby also PSEC can to some extent be connected to the pool to
do general tasks, when idle. Furthermore, the president should be
able to use the pool, if PSEC is busy.

The president is, however, afraid of getting worse service than

earlier. The consultant therefore outlines the following specific
proposal: The president's tasks would first of all go to PSEC. If she
is busy (with whatever job), the task would go to the pool. If the
pool is also fully occupied, the task would go back to PSEC. She
would in this case interrupt her present work, if this is a general task
and instead do the president's task. She would resume work on the
interrupted general task first when the president's task is finished.

A general task would first of all go to the pool. If the pool is

completely occupied, the task goes to PSEC. If she is busy, the
work goes back to the pool to go into a waiting line there.

The question is now how service would change, if this proposal is

implemented. Therefore the consultant wants to do a simulation
study, comparing the present system with the proposed system.

57
The inter-arrival times for the president's tasks vary between 20 and
50 minutes, while the IATs for the general tasks vary between 6 and
18 minutes. The president's tasks take between 15 and 25 minutes,
when done by PSEC, and between 16 and 30 minutes, when done
by someone in the pool, since the secretaries in the pool are less
experienced. The general tasks take between 20 and 40 minutes
when done in the pool and between 17 and 33 minutes when done
by PSEC. All times follow the uniform distribution.

Simulate both the present system and the proposed system for a day
of 400 minutes with 50 runs to see which system is best with regard
to the time needed to finish the president’s jobs.

87. The simulation concerns a company that buys a product from a

manufacturer and then resells it on the market. Demand for the
product is dependent on both price and product quality. Demand is
also a function of earlier sales, since some customers, after varying
times, return to buy an additional unit of the product. Demand is
characterized by a continuous function for first time buyers,
exemplified by the following data:

Price Annual demand

in units
10 60
20 40
30 25
40 20
50 15

The sales conditions are similar to those of Model62. The part of

the segment that deals with the customers who return for repeat
purchases is the same, also with the same relationship between unit
cost and the probability of no repeat sales.
58
If there are no units left in inventory, the customer goes to a
competitor and the order is lost. The number of lost orders is to be
recorded.

The financial consequences are, however, different from those of

Model62. This refers to the blocks after the SPLIT block. First the
customer decreases the contents of the storage with one unit. Then
we also add up sales and costs of goods. These costs are, however,
not connected with immediate payments. The payments (= cash
outflow) for purchased goods are now to be handled in the
inventory segment (see below).

Sales first lead to an increase in accounts receivable. The time after

which customers pay back follows the Erlang distribution with an
average of 45 days and n=3. The first 30 days are interest free. We
hence calculate the payment delay time, then subtract 30 from this
number of days. If this subtraction leads to a negative number, we
set the value for the time of interest payment to 0.

Assuming that the daily interest that we charge our customers is

24/36000, i.e. 24 percent annually, we next calculate the interest a
customer has to pay when paying for the product. This interest cost
of a particular purchase is added to net interest. At the same time
the total cumulative cash out-flows increase by this interest
payment. Finally, accounts receivable decrease and cumulative cash
inflows increase by the payment for the product.

After inputting price, unit cost and cash, we input replenishing level
and order quantity, i.e. when to reorder and how much to reorder
each time. The inventory segment is similar to that of Model44, but
for each replenishment purchase we calculate the total purchase
expenditure as the ordered quantity times the purchasing cost per
59
unit. This value is the borrowed amount and we increase debts by
this amount.

The company's pays the supplier after 50 days. The company pays
an interest on the borrowed amount. The daily interest rate is
20/36000, i.e. 20 percent annually. The first 20 days are, however,
interest free. These interest costs of a particular purchase are
deducted from the net interest results. At the same time the total
cumulative cash out-flows decrease by these interest costs. Finally
as the borrowed amount is paid back, debts decrease and cumulative
cash outflows increase by this amount.

The program should generate a report every 90th day, i.e., at the end
of each of six quarters. In this report segment, the number of the
quarter should first be increased by 1 and written. Net cash-flows is
calculated as cash inflows – cash outflows and added to total cash.
The value of inventories = the number of units in store times
purchasing cost per unit. The month's closing balance of assets is
cash + accounts receivable + inventories and profits are sales - total
costs + net interests.

The results are to be presented in a result matrix table with a short

explanation of the contents of the rows. The columns are
determined by the number of the quarter and the rows, 1 - 13,
represent, in order: sales revenues, total (purchasing) costs, net
interests, profits, cash inflow, cash outflow, net cash flow, cash
balance, accounts receivable, inventories, total assets, debts and
equity = assets - debts. Note that sales, cash inflow, cash outflow,
total costs and net interests are all quarterly, not cumulative, values.

We also want a graph over the development of cash over time.

60
88. We shall here study a problem that is of importance for the
managing of many retail operations dealing with perishable
products, like milk, fruit and meat. The manager of a local
supermarket has a single, deep shelf for cottage cheese. A customer
who comes to buy a container of cottage cheese will always grab
the one at the front of this deep shelf. In order to avoid spoilage, and
hence monetary loss or loss of reputation to the supermarket, how
should the stock boy be instructed to place newly arriving
containers of cottage cheese? Should he place them at the front of
the shelf so that customers will grab them first, or should he place
them at the back of the shelf, behind containers that have not yet
been sold?

The shelf can hold 20 containers of cottage cheese. Every 3.5 days
(Monday morning and Thursday afternoon) new cottage cheese
containers arrive from the distributor. The distributor leaves just
enough containers to fill the shelf (e.g., if there were five containers
on the shelf when the distributor arrived, she would leave 15 new
containers). Customer arrivals are exponentially distributed with an
average of one day apart. A container is considered spoiled if it has
been on the shelf for more than 10 days when it is sold.

We simulate one year (365 days) and provide for statistics on the
number of containers that were sold in a spoiled state for the two
different methods for stocking the shelf.

Hints: Use a SPLIT block to create enough containers to fill the

shelf with cottage cheese containers, when both the copies and the
original are considered. Use an ARRIVE block to start measuring
the time a container spends on the shelf before being purchased by a
customer and a DEPART block to conclude the measurement of this
shelf time. Use SEIZE and RELEASE blocks to assure that only
61
one container at most can be at a position in the front of the shelf at
any given time.

Use a WAITIF block, so that the container in the front of the shelf
must first wait if and as long as a customer is not in the store and
another WATIIF block to have it wait, while a customer is actually
picking the container from the front of the shelf. Use a QTABLE
statement with only 2 classes and a class width of 10, so that any
times greater than 10 minutes will be tallied as an overflow,
representing cottage cheese containers sold in a spoiled state.

89. Customers arrive by car to shop at a supermarket. The parking

lot has space for 650 parked cars. If a customer fails to find a
parking space, the customer leaves immediately without shopping.
A customer can walk to the supermarket from the parking lot in 60
seconds. Shoppers purchase between 5 and 100 items, uniformly
distributed. A customer buying 10 items or less, will generally use a
basket (70 provided). A customer buying more than 10 items will
generally use a cart (650 provided).
Shopping time per customer depends on the number of items
purchased (8 seconds per item). Customers select items and then
join the shortest queue at one of 20 checkouts. Customers
purchasing less than 10 items may choose the express checkout.
Checkout time takes 2 seconds per item purchased, plus a time of
25, 30, or 35 seconds. This time depends on the method of payment
(cash, check or credit card which are assumed equally likely or
probable). After checking out, a customer walks to the car (60
seconds), loads goods and leaves the parking lot. If there are no
carts or baskets available on arrival, customers wait until one is
returned.

62
The arrival rate of customers is exponentially distributed, starting at
600 per hour for half an hour, then 900 per hour for one hour, then
450 per hour for another hour and finally 300 per hour thereafter.
Run the simulation for 3 hours and determine the transit time of
customers, as well as the utilization of the parking lot, carts baskets
and checkouts.

90. In a duopoly situation, of the so called Bertrand type, the two

firms produce identical products. If one firm has a higher price than
the other, it will not meet any primary demand, i.e., it can only sell
if the other firm has not enough stocks. If it has a lower price p, it
faces a demand corresponding to potential annual sales q =ap-b,
where a =30000 and b =1.5.
To model this, we first input initial values of some constants, like
initial prices and costs, price1, price2, cost1 and cost2. We also
define the demand that a firm faces, if it charges a lower price,
lowrpr, than the competitor. This is equivalent to q =ap-b. We also
input the two constants prilow and priadd, used for changing the
prices based on stocks. We finally input protim, production time.
Since we name the firms 1 and 2 so that firm 1 is the firm with the
initially lower price, we also set the temporary lower price, lowrpr,
initially equal to price1.

We also define the capacities of the inventories of the firms,

allowing them a very high capacity. At the start of the first week
there are no inventories. As starting values we also set, for each
firm, the sales of the preceding week (before the start), salLa1 and
salLa2, as the highest possible weekly sales.

The model should have two main segments. In the first one we
generate a report and decision event each week of 7 days, starting at
63
time 0. We here first print the number of the week and then the
profits made this far, as well as the number of units in stock of each
firm.
Next we check if firm 1 has no stocks. If so, the price has been too
low and price1 is increased by priAdd. If not, i.e. stocks are not
empty, price has been too high and we decrease price1 by priLow
times the number of units in stock, so that price is lowered more if
the stocks are large. The price may, however, not be lower than unit
cost. In that case price1 is set to cost1. The changes in price done by
firm 2 are similar.

In this first segment we also deal with production. Each firm sets its
production quantity equal to the amount it sold in the preceding
period. After a potential production period of proTim days, the
products are ready to be put into inventory to be available for sales.
However, if a firm’s stock is already larger than its planned
production, then no production takes place this week. We also
determine what is now the lower price of price1 and price2, and set
this as lowrPr.

In the other main segment we generate every single order from

customers wanting to buy at the lower price. The average time
between two such orders, measured in days, is 52*7 divided by the
number units demanded annually. The actual time between two
orders varies, however, stochastically according to the negative
exponential distribution.

If firm 1 has the lower price, we go to the address sltes1, where we

test if firm 1 has any stocks. If so, firm 1 can sell and we go to sell1.
If firm 1 does not have any stocks, demand goes to firm 2, i. e. we
go to sell2. If firm 2 does not have any stocks either, demand goes
back to firm 1, where one waits until it has got stocks again.
64
If firm 2 has the lower price, we go to the address sltes2, where
similar conditions apply. If the firms have equal prices, we proceed
by random with 50 percent chance to sltes2 and with 50 percent
chance to the next block sltes1

At the address sell1 we might first have to wait until there is some
unit in firm 1’s stocks. We then take one unit out of firm 1’s stocks
and increase the number of sold units this week by 1 as well as firm
1’s revenues by the price of the product. We also increase firm 1’s
costs of goods sold by the unit cost of the product. At the address
sell2 the corresponding happens to firm 2.

Build a program that simulates this situation and produces a graph

for the development of the prices of the two firms for the first year
as well as a weekly report on the cumulative profits and on
inventories.

We can exemplify with the following input values: price1=24,

price2=26, cost1=12, cost2=13, a=30000, b=1.5, priLow=0.1,
priAdd=0.5, and proTim=5.

91. Karl is writing a business plan for a company he would like to

start to produce private hobby airplanes. It would import the parts
from the USA, assemble the planes and then sell the planes to
customers in Estonia. Karl would like to simulate the operations of
the company for the first two years, each of 240 working days.

Karl will use his business plan to obtain € 200,000 initial capital (in
cash) from an investor. In case more cash is necessary at some point
for operations, the company can take a loan of up to € 0.5 million
from the bank. At the end of each month interest must be paid, 1%
of outstanding loan (12% annually). The principal is not repaid
during the first two years; only the interest is paid.
65
In case the loan has reached the maximum amount and more cash is
still necessary, bankruptcy is declared and the simulation stops.
Bankruptcy is also declared if the equity (assets minus liabilities) on
the quarterly balance sheet is negative.

The parts are shipped from overseas in ship containers, each

containing up to a maximum of 5 sets. The price for delivering a
single container is € 20 000 and each set of parts costs € 40 000.

Karl will consider making an order every three weeks (15

workdays), the first time on the first day. If at that moment there is
no order pending and the stock of plane parts (including WIP) is
less than or equal to some fixed reordering level, he will make an
order.

Each order is made for a single container and Karl can choose any
quantity of sets between 1 and 5. (Both the reordering level and the
quantity of sets will be kept fixed throughout the simulation.) It
takes between 24 and 30 working days for a container to arrive,
each real value in the interval being equally likely. The invoice
must be paid within 15 workdays after an order is received.

The planes will be assembled by mechanics working for the

company. The plan is to rent a hangar (costs € 4000 per month),
where the assembly of a single plane is possible at any given
moment. It requires 2 to 5 mechanics for the assembly (monthly
wage for each is € 3 000). Wages and rents are paid at the end of the
month.

The average time (in working days) taken to assemble a single

plane is given below. The assembly time is random, but in 95% of
cases it is within ± 20% of the average.

66
 Number Average
of workers assembly time
2 15
3 11
4 10
5 9
Once a plane is ready, it is sent to the airport and parked, ready for
customers. The parking space can be considered unlimited, but the
company has to pay a parking fee. The fee, to be paid at the end of
each month (of 20 working days), is € 1000 for each parked plane at
that moment.

The customer arrivals are modelled as a Poisson process, i.e., the

inter-arrival times follow the exponential distribution. The expected
number of customers during one year depends on the price of the
planes according to the data below. (For values between the listed
prices, use continuous interpolation.)
Price in Expected nr.
€ 1000 of customers
60 25
65 18
70 15
75 10
If there is a finished plane parked at the airport when a customer
arrives, it is sold immediately; otherwise customers join a waiting
list. Once a plane is sold, an invoice is written. Half of the
customers transfer the money within 27 workdays, but 5% take 63
days or more. The average time taken is 30.

Use the Erlang distribution with shape factor 3 and corresponding

mean to model this delay.
67
Sales are the only source of revenue. Variable costs are the ordering
and shipping costs of parts, and fixed costs are the monthly
payments for rent, wages and interest. Profit = Revenue – Costs. On
the balance sheet, all stock of parts, WIP and ready planes are
valued at shipping + purchasing costs per item. Stock, Accounts
receivable and Cash constitute Assets, while Loan and Accounts
payable constitute Liabilities. Equity = Assets – Liabilities.

Model the physical processes described above and also keep track
of the underlying accounting and cashflows. The produced output
should contain a plot of cash & loan, quarterly printouts of the
profit & loss statement and the balance sheet. All accounts should
be in thousands of €, without decimals.

You should experiment and decide on suitable values for:

 Reordering level for parts (0 - 3)
 Quantity of sets per order (1 - 5)
 Number of mechanics to hire (2 - 5)
 Price for planes (€ 60 000 - 75 000)
The chosen values shall remain constant throughout the simulation.
The objective is to have a high expected profit, while keeping the
probability of bankruptcy during the first two years low. At least
one run should be without bankruptcy. Customer waiting times
should also be reasonably short.

92. We can by simulation estimate what in Management Science is

called the value of perfect information. This refers to a situation
with stochastic variations. One can for such a situation first
determine what decision would be made, if one, due to perfect
information, in advance knew what event would take place.

68
One can next determine what decision would be made, if one is not
able to forecast what event will take place, but only which decision
would maximize expected profits. The value of perfect information
is then the difference in profit between the case with perfect
information and the case with uncertainty.

We illustrate this with a situation of a newspaper seller, who has to

order a certain number of a special foreign newspaper for sales each
day. The number of papers that can be sold varies between 25 and
28, with 10 percent probability of 25, 30 percent probability of 26,
50 percent probability of 27 and 10 percent probability of 28. At the
end of the day the newspapers are worthless.

Each unit ordered leads to a cost of $ 8, while each unit sold leads
to an income of $ 10. If one has ordered fewer units than those
demanded, sales only amount to the number ordered.

We test for the ordering of 25, 26, 27 or 28 units. This can for all
aGPSS versions be done by first running 400 times, the first 100
runs with the order quantity of 25, the next 100 runs with 26, the
next 100 runs with 27 and finally runs 301 – 400 with 28. For each
of these four batches of 100 runs, we add up the profits and
calculate the average profit for each such batch.

At the end of the 400th run, we determine which value of the order,
25 - 28, has given the highest average profit. This value then
becomes the value of “non-perfect information”.

For runs 401 – 500 we set the order quantity equal to the sampled
demand value and calculate the average profit for the case when we
in advance know what demand value will be sampled.

69
At the end of the 500th run we calculate the difference between this
average value and the value of “non-perfect information” and print
this value as the Value of Perfect information.

93. We study the decision on the manning of a team that shall

develop a new software product. The product is expected to be of
interest only during the next four years. The company believes that
the product cannot be sold after this. We are hence only interested in
the profits during the next four years.

The product can only be put on the market once the development is
finished. The expected value of the finishing time varies with the
number of developers according to the following estimate:

No. of Average
developers development
time
1 1000
2 500
3 280
4 240
5 200
6 180
7 165
8 155
9 150

There is a daily cost per developer of $ 1600. If the software has not
been developed within three years, development is stopped. The
development cost is to be paid at the end of each month. Once the
development is ready, sales start in the immediately following
month. Average sales are expected to be 2000 units per month.
70
Actual sales vary according to a normal distribution with a standard
deviation of 400 units. The price of the software system is $ 299.
The unit cost of production, incl. packaging with manuals, is $ 50.

All monthly profits are discounted back to the present. The

discounting rate is 18 percent, corresponding to 0.05 percent per
day. We want to estimate the total expected discounted profit for
different numbers of developers, what variations are likely and
which number of developers seems most suitable.

94. In a restaurant there are different synchronization problems.

First, the waiter/waitress waits until each guest in a group has
chosen a meal. In order to serve the meals for the group together,
the meals wait until the last one for the group is prepared. Then they
wait until one of two waiters/waitresses are free. The table is
normally cleaned when all guests have finished their meal.

A groups of customers arrive on average every 10 minutes,

following a Poisson pattern. The number of persons in each group
and their relative frequency is as follows:

1 0.2
2 0.24
3 0.1
4 0.2
5 0.1
6 0.16

There are six tables in the restaurant; four tables for four people and
two for six people. Each group waits until there is a table free to
accommodate all in the group. If there are already three groups
waiting, the whole arriving group goes to another restaurant.
71
For the number of drinks in each group, the time to consume these
drinks, the time of each group to choose and next to order the food,
the time to eat the food, and finally the time to pay and the time
spent after paying, you are free to decide on what the values
(possibly following distributions or functions of other factors) that
you find most suitable. The same refers to the time for the barman
to prepare the drinks, the cook to prepare the food, the time of one
of the two waiters to serve and finally to clean the table.

95. We revise Model100 by assuming that the city is large enough

to have five service bays and five mechanics in its vehicle-
maintenance garage. We let the number of vehicles coming in for
scheduled maintenance each day be uniformly distributed between
12 and 18, instead of between 2 and 4.

We continue to let police cars arrive for unscheduled repair in a

Poisson stream, but now at a rate of five cars every two days,
instead of one every two days. We use the same preemptive queue
discipline that was used in Model 100 and use the other
assumptions made there, i.e. that the mechanics work their 8-hour
day without taking breaks and that the garage is open 7 days a
week. We simulate for 25 days.

96. This simulation deals with beer drinking. The program should
start by asking the user: “Want to drink Swedish beer in 40 cl
glasses? Y/N”, with hopefully Y as the answer.

It then asks for the average time between sips in minutes and next
for the average size of a sip in cl. Finally it should ask for how
much beer in cl. you have left in your glass, when you order a new
glass.

72
Then the simulation starts. The real time between sips follows the
negative exponential distribution of the average input time. We
assume that the true sip varies according to the normal distribution
with the input average and with a standard deviation of 10 percent.
The new glass of beer comes 10 minutes after the order, but varying
with the normal distribution with a standard deviation of 2 minutes.

The drinking activity is recorded with a log with the following types
of output lines:

At time 0.00 I order one beer

At time 10.91 I drink 5.26 cl beer
..
At time 46.02 I order one beer
..
At time 59.06 my glass is empty

The simulation is ended after five beers with the output:

Five beers are enough. Carry me home!

97. This example deals with the evaluation of European stock

options. It is very simple and meant to give a taste of the role of
simulation in determining the value of these options, i.e. to show
both the potentials and problems involved.
ytu
We deal with a call option where the buyer (owner) of the option can
buy the stock at a pre-set price, the so-called exercise price, after one
year, if she wants to. We have for the model chosen the following
formula for estimating the value of the stock at the end of a period:
y*exp((r-s2/2)*t+s*sqrt(t)*fn$snorm)
y is the value of the stock at the start of the period, r is the annual
interest rate as a fraction, s is the annual volatility of the stock, also
73
given as a fraction and t the time after which the new value is
reached.

We shall in the formula above have t as a month and we can then in

12 steps calculate the value of the stock after one year. Next we
calculate the difference between this value and the exercise price.
The value of the option at the end of the year is then the maximum of
this difference and 0, since the option owner will only use her option,
if the difference is positive. The value at the start of the year is then
the present value of this.

For running the model, we first study the case when the starting price
is 100, the exercise price is also 100, the annual volatility is 25 and
the interest rate is 5 percent. For the case that we have constant
volatility, we can compare the value obtained from the simulation
with that obtained from Black&Scholes formula calculators.

The interesting thing with simulation in this case is to obtain option

values also for a case when the volatility changes during the year, a
case not covered by the Black&Scholes formula. We shall here only
study the simple case when volatility changes from an initial
volatility by a monthly increase. We can allow this change in
volatility to be stochastic, so that this change follows the normal
distribution with the change being the average change in volatility
times (1+fn$snorm*stddev), where stddev has been read-in. We can
e.g. study the case when the average monthly increase in volatility is
1 and its standard deviation is 0.2.

We run the simulation a great many times. Since, for example, 400
times is the maximum allowed for experiments for the extended
student version of aGPSS, we write the model so that we in the same
run can obtain results for more than 400 runs.

74
We hence need one segment that goes from month 1 to month 12 for
each run and one segment that starts every 12 months and then sets
the start values for each year, i.e. for the number of the month and
the initial volatility. Finally we need a segment that stops the
simulation after m months, where m is the number of runs *12.

At the end of each run, we add the present value of the option value
in this run to the sum of such present values in the earlier runs. In the
last run we divide this sum with the number of runs to obtain the
expected value of the option.

The desired output is this expected value of the option. It is also of

interest to have the possibility to show a graph of the values of the
stock at the end of the year.

The model should allow for input of the starting price, the exercise
price, the interest rate, the initial volatility, the monthly change in
volatility, the standard deviation of this change and the number of
runs.

98. Our next model is a simple version of the models that our
students have built for a Swedish company selling corporate
switchboards. The company originally used an analytic queue
theory model, but early simulation models showed that the
analytical models lead to errors, e.g. regarding waiting times and
lost customers, of up to 20 percent.

The model deals with an insurance company buying a new switch

board. This company shall in co-operation with the seller of the
switch board, having the simulation model on his laptop, determine
the number of in-going trunk lines for the switch board and the
number of agents that shall handle customer calls.
75
The arrival of customer calls follows the Erlang distribution with a
parameter n of 3 with an average IAT that depends on the time of
the day. The average IATs in seconds during the day are estimated
to be as follows:

9 – 11 30
11 – 13 50
13 – 14 20
14 – 16 25
16 – 18 30
After 18 30

The switch board opens at 9 a.m.. After 6 p.m., no new calls are
admitted into the system, but the calls having arrived before that are
kept in the system until 7 p.m., when the system is closed down.

The calls first come to the switch board. If all trunk lines are busy,
the caller has the choice of either calling back later or calling
another insurance company, implying a lost customer. 70 percent
prefer to call back later. They will then wait 10 minutes on average,
with a standard deviation of 2 minutes. They then call back, but if
there is still no free agent, they call another insurance company,
implying another lost customer.

Those calls that are admitted into the system might still have to wait
for a free agent. The time needed by the system for finding a free
agent is 20 seconds. If all agents are busy, the caller is put on hold
until an agent is free.

The caller then spends a certain time with the agent. To model this
time, we assume an average of 4 minutes.

76
In order to have this reflect the risk of some lengthy calls, we
assume a basic time with an average of 3 minutes, following the
normal distribution with a standard deviation of 30 seconds. To this
basic time we add an average of 1 minute, following the negative
exponential distribution. After this total time, the call is completed
and the agent and the trunk line are free for other calls.

At the end of the simulation of one day, we want to calculate the

costs of these operations for a certain number of agents and trunk
lines. The total costs, which are to be printed out, contain three
components:

1. The salaries to the agents, which is the number of agents

times the daily cost of an agent.
2. The cost of the switch board, which is the number of trunk
lines times the daily cost of each trunk line.
3. The cost of losing a customer, which is the number of lost
customers times the cost of losing a customer, i.e. foregone
profits.

As start values we input the daily cost of an agent (e.g. 100), the
daily cost of a trunk line (e.g. 50) and the cost of a lost customer
(e.g. 200).

99. We shall here make a simulation of the value of a new high tech
company. We shall do this by making a forecast of various factors
such as sales, profits, dividends, year by year, for a great many
years up to a horizon year T.

In the first step, step a, we build a completely deterministic model,

with no random effects. We set the value of the horizon year to 40

77
We shall start with the forecast of the development of the market for
the main product of the company. As a first step, we forecast the
total market for the product that the corporation is about to sell. We
shall first forecast the total market in terms of a constant economy.
We shall for the forecast of this total market assume that the
development of the sales of the firm's product follows the familiar
pattern of the logistic curve, used in Exercise 73.

As the first parameter we set the market saturation level ymax, to

100. The next parameter is a growth rate gy, set to 0.003. The
market growth yt from year t to t+1 is gy(ymax-yt) yt, where yt
represents the total (relative) market in period t. We can in this way,
by specifying an initial value, y1, of y in year 1, 10 (e.g. K units),
calculate the total market for year after year.

We also allow the total demand for the firm's product to grow in
proportion to the expansion of the total economy, e.g. with
population growth, i.e. a growth of a fixed percent each year in an
exponential fashion. We shall here forecast an index-like value for
the whole economy, the GNP, with a value of 100 for the first year,
and then increasing by a constant growth rate of 3 percent. The total
market for the product is then be defined as the earlier forecasted
”economy-independent” market multiplied by value for the total
economy.

Having defined the total market, our next step is to determine the
sales of our specific corporation. This will depend on the
development of the market share of the corporation, starting at 10
percent. We assume that the market share s will develop over time,
eventually gradually reaching a maximum market share of 25
percent. Thus the change in the market share st from period t to year
t+1 is gs(smax - st), where gs is set to 0.15. Multiplying the total
78
market in each year with the market share, we get the total sales of
the studied corporation.

The next step is to determine how the annual profit develops over
time. We assume that profit is dependent on the profit margin. This
percentage will also develop over time. We assume that the profit
margin m develops gradually over time from its starting value m1,
set to 10, to gradually approach a final value, mmax, of 20. The
change in the profit margin from year t to year t+1 is gm(mmax - mt),
where gm is a growth rate for the profit margin, of 2 percent.

We must also include a fixed cost, of 100, leading possibly to initial

losses. For the sake of simplicity we assume in this simple model
that this fixed cost is constant over time. Multiplying sales with the
profit margin and deducting the fixed cost, we obtain the
development of profits over time.

Having established the development of the annual profit of the

corporation, we next forecast the development of dividends. We
assume a fixed dividend ratio, of 50 percent, such that the annual
dividend is this ratio times the profit, provided the annual profit is
>0 . In the case of a loss, the dividend is 0. We shall next in our
model calculate the sum of the present values of all dividends up to
the horizon T. We assume a discount rate of 10 percent.

In order to take also the dividends after the horizon year T into
account, we first assume that all dividends after T are the same as
the dividend at year T. We can next calculate the lump sum value at
year T of all dividends after the year T as the present value of all
dividends from the horizon T to infinite time, i.e. equal to the
dividend in year T divided by the discount rate. We add the present
value of this lump sum to the total value of the corporation.
79
We also have to include the effect of the profits not distributed as
dividends, i.e. for each year the profit minus the dividend. We
assume that these amounts are invested into a special reinvestment
account carrying an annual interest, different from the discount rate,
of only 7 percent. We allow this reinvestment account to grow each
year until the horizon year T. In this year we calculate the present
value of this reinvestment account and add it to the total value of
the corporation.

The value of the company is thus calculated as the sum of the

present value of all dividends from year 1 to the horizon year T plus
the present value of the lump sum in year T plus the present value of
the investment account.

In step 1 we want to have nine graphs over the development of 1.

the market as such, 2. GNP, 3. total sales on market, 4. the market
share, 5. sales, 6. margin, 7. profits, 8. the three components
building up the total value and finally 9. the company’s value.

In step b we want to investigate the effect of some random

variations. We shall assume three variations, following the normal
distribution. The GNP growth rate can vary between 0.01 and 0.05,
the market maximum value between 80 and 120 and the margin
maximum value between 10 and 20, all with a 95 percent
probability that the sampled value will lie in between the lower and
upper values given above.

We shall in this case abstain from graphs but just investigate to

what extent the size of the horizon year influences the total value of
the corporation. Will there be a significant difference in this value,
if the horizon lies 30 or 40 years away?
80
100. The car producer Ovlov has close cooperation with the French
Napoleon company. The two companies are now in the process of
planning their purchase of a new metallic paint. This new paint will
protect better against corrosion.

It is purchased from three different suppliers, one in Düsseldorf, one

in Esbjerg and one in Torino. This paint is used in the three car
plants, in Antwerp, on Hisingen and in Paris.

The demand for this paint is dependent on when and where the cars
are produced. One opens a barrel with paint and fills the painting
machine when one has a whole batch of 100 cars ready for painting.
Before this, one has to clean the spraying nozzles. On average one
has to open a new barrel with paint every 7.5 hours on Hisingen,
every 8 hours in Antwerp and every 9 hours in Paris, but the times
between these barrel openings vary and seem to follow the negative
exponential distribution.

In case that one does not have any barrels with this paint, one has to
wait with painting the 100 cars until paint barrels arrive. The delay
of a series of 100 cars costs 500 € (i.e. 5 € per car) for each hour of
delay for all three production units.

When simulation starts, there are stocks of 20 barrels in Antwerp,

20 in Paris and 15 on Hisingen. One has a capacity of stocking 1000
barrels in Antwerp, 500 on Hisingen and 400 in Paris.

The production of paint is done in batches of 10. In Düsseldorf it

takes on average 20 hours to produce a batch of 10 barrels, but time
varies so to that it in 95 percent of all cases takes between 12 and 28
hours, seemingly according to the normal distribution. In Esbjerg
and Torino conditions are similar, with variations between 24 and
81
40 hours, and 24 and 36 hours, respectively. The barrels are stored
while waiting to be fetched. There is a capacity to store 1000 barrels
(i.e. 100 batches of 10 barrels) in Düsseldorf, 500 in Torino and 400
in Esbjerg.

This type of metallic paint hat to be transported in special trucks. At

present there are three such trucks; at simulation start one in
Düsseldorf, one in Esbjerg and one in Torino. They leave (in this
order) for that one of the three plants that has the longest waiting
line of unpainted cars. The average travel time in hours for the
trucks between the places is given in the following table:

Antwerp Hisingen Paris

Düsseldorf 8 22 12
Esbjerg 17 14 20

Torino 18 28 16

The actual time needed for each truck will, however, vary,
seemingly according to the normal distribution, so that 95 percent
of the times will lie between 60 and 140 percent of the average
time. There is, however, also a lower limit on the time needed,
given in the following table:
Antwerp Hisingen Paris
Düsseldorf 6 16 9
Esbjerg 13 9 14
Torino 12 20 11

82
All times shown above include both loading and unloading, which
is not very time consuming. When a truck has arrived at the car
production plant, one unloads the 10 barrels. The truck is then ready
to go back to one of the three paint producers. It will go to the color
producer that has the largest number of barrels in stock. If no
producer has any stocks, it will go to Düsseldorf. The times for the
return travels are the same as the times going to the car producers.

Since the trucks are expensive and the trucks are manned by
specially trained persons, one calculates with a cost of € 1000 per
hour of truck travelling time. When the trucks are only waiting,
there is an hourly cost of € 700 per hour.

After the trucks have returned to the color plants, they load 10
barrels, provided 10 are in stock; else they wait until 10 have come
into stock. They next again determine which car factory has the
longest line of waiting unpainted cars and go there. In this way, the
trucks go back and forth between the color and car plants.

We simulate for 100 days of 24 hours. Both the car plants and the
color plants work in many shifts around the clock. We want to
calculate and print the total costs for the trucks plus the cost of
production having to wait.

We want to simulate both for the present case of 3 trucks and the
case of one additional truck, starting in Düsseldorf. Make enough
runs to determine if it is worthwhile getting an extra truck.

83