Event Logistics

2
EVENT LOGISTICS
Kjetil K. Haugen
c Tapir Academic Press, Trondheim 2011
ISBN 978-82-519-2783-3
ISBN 978-82-519-2784-0 (epub)
This publication may not be reproduced, stored in a retrieval system or transmitted

in any form or by any means; electronic, electrostatic, magnetic tape, mechanical,
photocopying, recording or otherwise, without permission.
Layout: Kjetil K. Haugen

Cover Layout: Mari Røstvold, Tapir Academic Press
Paper: Munken lynx 90 g
Printed and binded by: AIT Oslo AS
Tapir Academic Press publishes textbooks and academic literature for universities
and university colleges, as well as for vocational and professional education. We
also publish high quality literature of a more general nature. Our main product
lines are:
• Textbooks for higher education
• Research and reference literature
• Non-fiction
We only use environmentally certified printing houses.
Tapir Academic Press

NO-7005 Trondheim, Norway
Tel.: + 47 73 59 32 10
Email: post@tapirforlag.no
www.tapirforlag.no
Publishing Editor: terje.togersen@tapirforlag.no

Preface
This book is written to be used as teaching material for a course in Event

Logistics. The course is planned to be given at Molde University College -
Specialized University in Logistics, the first time in fall 2010. This course is
a part of the Event Management programme launched in Molde, fall 2010.
To be able to understand this book, a basic knowledge in Logistics is
necessary. Some of the material is probably too advanced for readers with
only a basic knowledge of Logistics/Operations Management, and several
appendixes that signal this type of difficulty are used. So, readers with only a
minor level of knowledge in logistics should probably avoid these appendixes.
In order to meet these constraints, the planned course will be accompanied
by another (standard) text book in logistics at an intermediate level - for
instance “Production and Operations Analysis”, by S. Nahmias [21]. The
basic idea in teaching the course, is to capture essential Logistics modelling
through selected topics in (e.g.) [21] and then continue and finish up with
the contents of this book.
The structure of the book is consciously kept at a minimal academic level
- in the sense that literature references are kept at a minimum. The reason
for such a choice is of course partly laziness, but also the wish to produce
something that is more easily accessible than normal research literature.
Kjetil K. Haugen
Molde, Norway - Brno, Czech R. - Vienna, Austria - Budapest, Hungary

June - December 2010, January 2011
Acknowledgements
I wish to thank several of the local (and not so local) planners of the Event
Management Programme; Professors Harry A. Solberg (Sør Trøndelage Uni-
versity College), Holger Preuss (Johannes Gutenberg University, Mainz, GER)
and Hallgeir Gammelseter as well as Associate Professor Nigel Halpern (both
from Molde University College) who made invaluable contributions to the
programme.
A group of local Event Arrangers, Head of Molde International Jazz Fes-
tival; Jan Ole Otnes, Co-founder of Molde International Jazz Festival; Petter
Petterson JR., VP of International sales in Renkus-Heinz Professional Audio
Systems; Karl Brunvoll, Producer at the local theatre “Teatret vårt”; Hal-
vard Fiksdal, and local politician; Torgeir Dahl, formed an early planning
group which made considerable impact on the final master programme.
Finally, a large group of international professors in sports, tourism and
event management have committed themselves to aid the programme through
the supervision of masters’ theses.
I am very grateful to all of them!
Assoc. Prof. Asmund Olstad has read through the manuscript, thank’s a
lot Asmund!
Contents
1 Introduction 15
1.1 What is Logistics? . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.1 Looking for a definition . . . . . . . . . . . . . . . . . . 15
1.1.2 The importance of logistics, now and in the future . . . 17
1.1.3 Logistics theory, practice and research . . . . . . . . . 19
1.2 Services, Manufacturing and Events . . . . . . . . . . . . . . . 21
1.3 Event Logistics . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4 Events and Uncertainty . . . . . . . . . . . . . . . . . . . . . . 24
2 Event Forecasting 27
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 The fallacy of traditional time-series-based forecasting . . . . . 27
2.2.1 A simple example . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Utilizing explanatory information - the reason behind
the 2004 bump . . . . . . . . . . . . . . . . . . . . . . 30
2.3 The case of football demand forecasting . . . . . . . . . . . . . 32
2.3.1 An old MFK case . . . . . . . . . . . . . . . . . . . . . 32
2.3.2 The effect of Pre-sales . . . . . . . . . . . . . . . . . . 48
2.3.3 Capacity constraints . . . . . . . . . . . . . . . . . . . 52
2.4 Aggregated versus disaggregated forecasts . . . . . . . . . . . 54
3 Events and Inventory Management 57

3.1 Inventory Management - Introduction . . . . . . . . . . . . . . 57
3.1.1 The EOQ model . . . . . . . . . . . . . . . . . . . . . 57
3.1.2 Applying the EOQ approximately . . . . . . . . . . . . 58
3.1.3 The Lot-sizing approach . . . . . . . . . . . . . . . . . 59
3.2 Event Inventory Management - “News-boy” setting . . . . . . 61
3.2.1 “News-boy” basics . . . . . . . . . . . . . . . . . . . . 61
3.2.2 A simple T-shirt example . . . . . . . . . . . . . . . . 62
3.3 Extensions of the “News-boy” concept . . . . . . . . . . . . . 67
3.4 Final comments . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Event Production 69
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 The classical aggregate production planning model . . . . . . 69
4.3 Developing an event model alternative . . . . . . . . . . . . . 71
4.4 A simple example . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5 Final comments . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5 Event Supply Chains 75
6 Event Transportation 77
7 Events and Dynamic Pricing 79

7.1 Dynamic Pricing - Introduction . . . . . . . . . . . . . . . . . 79
7.2 Dynamic Pricing in Manufacturing . . . . . . . . . . . . . . . 81
7.3 Dynamic Pricing in services - Revenue Management . . . . . . 82
8 Events and “hype logistics” 85

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2 Just In Time - JIT . . . . . . . . . . . . . . . . . . . . . . . . 85
8.3 Postponement . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.4 Modularization . . . . . . . . . . . . . . . . . . . . . . . . . . 87
9 Event Facility Location 89
10 Event Sequencing 93
Appendices 95
A Calculations in figures 2.2 and 2.3 97
B Data for the case in Subsection 2.3.1 99
C LINGO model in Subsection 3.1.3 101
D Case solution in Subsection 3.1.3 103
E LINGO - the problem in Section 4.4 105
Bibliography 107
Index 111
List of Figures
1.1 The simplified production function of microeconomics . . . . . 20
2.1 Plot of yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009 29
2.2 Moving average (orders 2, 3 and 4) forecasting on the Teatret
Vårt data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Moving average (orders 2, 3 and 4) forecasting on the adjusted
Teatret Vårt data. . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4 MFK - Home match attendance, 1995 – 2001 . . . . . . . . . . 33
2.5 MFK - Home match attendance, 1995 – 2001 with average
values before and after observation 39 . . . . . . . . . . . . . . 34
2.6 SPSS output from the model . . . . . . . . . . . . . . . . . . . 38
2.7 SPSS output from the reduced model . . . . . . . . . . . . . . 39
2.8 Forecasts vs. observed attendance for the 2002 season . . . . . 44
2.9 Short and long term forecasts compared to actual attendance
- 2002 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1 An optimistic and a pessimistic density for t-shirt demand. . . 64

List of Tables
2.1 Yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009 . . . . 28

2.2 Yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009 . . . . 31
2.3 Explaining the variables of equation (2.5) . . . . . . . . . . . . 37
2.4 Forecast accuracy of explanatory variables . . . . . . . . . . . 41
2.5 MFK (point) performance . . . . . . . . . . . . . . . . . . . . 41
2.6 Home match schedule - 2002 season . . . . . . . . . . . . . . . 42
2.7 Forecasted and observed attendance - 2002 seasons . . . . . . 43
2.8 MFK - complete performance statistics for the 2002 season . . 45
2.9 Short term forecasts and observed attendance - 2002 seasons . 46
2.10 Short term forecasts and observed attendance - 2002 seasons . 47
3.1 Forecasted attendance demand for MFK home games - 2002

season . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 Forecasted coke demand for MFK home games - 2002 season . 58
3.3 Purchase plan for coca cola; MFK home games - 2002 season . 59
3.4 Tickets (and t-shirts) sold at MIJF: 2005 – 2009 . . . . . . . . 63
4.1 Data for the event aggregated production planning example . 73
A.1 Moving average calculations underlying figure 2.2 . . . . . . . 97

A.2 Moving average calculations underlying figure 2.3 . . . . . . . 98
B.1 The first twenty data points for the regression analysis . . . . 99
Chapter 1
Introduction
1.1 What is Logistics?

1.1.1 Looking for a definition
The term Logistics is old and was originally used in military operations. The
massive need for planning related to large transportation of soldiers, supplies
and technology in war time situations made military logistics important.
Today, this seems obvious from the failure of Napoleon (and Hitler). Still,
apart from military operations, logistics as a term is relatively new as a
scientific subject. For instance, Molde University College was among the first
institutions in Norway - in the mid eighties - to launch academic programmes
in logistics.
Logistics as an academic subject may largely be divided into two fairly
different sub categories.
• Quantitative Logistics
• Qualitative Logistics
Quantitative Logistics, as the term indicates, focuses on mathematical

modelling as the primal toolbox for handling logistics planning problems.
Outside of Scandinavia, the term Operations Management may be a fairly
good synonym for Quantitative Logistics and many scholars would bring
the term even further and define both Operations Management as well as
Quantitative Logistics as sub disciplines of Operations Research (OR). Op-
erations Research may be defined as a sub discipline of Mathematical Mod-
elling in general. Operations Research emphasizes discrete optimization, but
embraces other mathematical modelling disciplines such as Queuing Theory,
16 Introduction
Discrete Event Simulation and Forecasting. All techniques are very relevant
for logistics planning in general.
The above (perhaps) somewhat blurred discussion, brings us to a possible
definition of Quantitative Logistics:
Quantitative Logistics or Operations Management may be defined

as the application of OR techniques limited to the following areas:
• Forecasting
• Production Planning
• Inventory Management
• Transport Planning
The above definition might be viewed as very traditional and conserva-

tive by many logistics researchers as of today. Many would claim that the
above definition narrows down logistics (as well as Quantitative Logistics)
way too much. Some would say that far more of a company’s logistics prob-
lem is contained in the logistics concept than the above four topics - typical
examples may be pricing policies, technology choice, information strategies,
human resource management, contractual theory, supply chain management
and so on. Still, this text will (mostly) stick to the relatively narrow (but
precise) definition outlined above.
The topics listed above; forecasting, production planning, inventory man-
agement and transport planning) indicate that logistics and quantitative lo-
gistics is related to planning for certain parts of a company’s activities. One
way of looking at this could be the following:
All companies make decisions affecting their demand. Certain decisions
made “early” such as product design choices and technology choices define
physical aspects of products1 , while more direct market related decisions such
as pricing, marketing decisions as well as decisions related to the company’s
competitive situation typically are made somewhat later in the life-cycle of
a product. Given that all possible decisions in these two groups are made,
what remains is to produce (possible store) and transport the given prod-
uct to the market. This process contains the traditional logistics definition.
Consequently, logistics is a lot about streamlining this process of produce -
store - transport. So, concepts such as “the right amount at the right time
to the right place” hence makes sense.
1
At this point we implicitly assume a focus on manufacturing as opposed to service
production. We will return to these important concepts later on.
1.1 What is Logistics? 17
Given the above definition, many “ordinary people” (whoever they might
be) would perhaps argue that the real “sexy” company decisions are not lo-
gistics decisions. Defining the physical aspects of the product by creative
design and/or marketing strategies and complex pricing strategies are far
more challenging than the somewhat boring logistics decisions. To a certain
extent, the author may agree and if we take a slight look at the market evalua-
tion, we will probably find support for such a hypothesis - product designers
and marketing people normally make more money than those occupied in
the storage rooms. Still, a more modern view of logistics may change this
traditional thinking.
Before moving into the next subsection, here a few a few words on Qual-
itative Logistics. Qualitative Logistics approaches logistics problems from a
more philosophical perspective. The main difference compared to the quanti-
tative branch is perhaps related to the use of mathematical tools. Qualitative
Logistics research uses, as the name indicates, far less formal mathematical
tools, and degenerates to a more verbal “social science”-like angle of attack.
To some extent a few central topics, which may be found in this category, is
discussed in Chapter 8, still with a focus on events and with a quantitative
touch.
1.1.2 The importance of logistics, now and in the fu-

ture
Most people who deal with logistics - either practically or more theoretically
- would of course like to be an integral part of “sexy” company decision
making. It may be that certain aspects of modern reality may lend a helping
hand to frustrated “logisticians”. One phenomenon of the world that has
been quite obvious in recent years is by many authors termed globalization.
(For those seriously interested in the subject, the following literature pointers
may be relevant [9], [18], [16], [25].)
Now, many scholars and practitioners argue that in a globalized world
(a world with relatively small physical transportation costs and negligible
informational transaction costs), competition will increase. This seems like a
reasonable hypothesis, increased globalization leads to increased competition
in product (and service production) markets. After all, barriers that prevent
competition are all sorts of transaction costs. At the same time, certain
political processes, for instance, the development of EU has also had, and
probably will have, significant impact on reduction of barriers preventing
competition. As such, a hypothesis stating increased future competition
seems reasonable. Given such an assumption, it likewise seems reasonable to
18 Introduction
assume less possibilities for maintaining technological and or design advances

over significant time periods. A recent example, may be found in the Ipad,
which has become an enormous success perhaps due to a combination of
design and technology choices.
Still, in a globalized and competitive world, one would expect clones
and/or direct copies popping up very fast after such a success. Examples of
such has popped up, for instance, the Samsung Galaxy Tab, as one of several
responses to the Ipad.
As such, competition makes technological and/or design advantages far
more fragile. This argument has lead modern logisticians to argue that in
the future, competitive advantages in technology (or markets) will vanish due
to increased competition. Then, the only remaining dimension for creating
differences between companies is logistics. As such, one could say that in a
perfectly globalized and competitive world, the only possible way of creating
a difference is by choosing a different logistics system. Hence, it will not be
so much about choosing the right product as how to manufacture it.
Moving into events, things change somewhat2 . A central concept in most
events, whether they are from sports or other cultural scenes, is branding.
Branding may be defined (simply) as the non (or very hard) copyable part
of a product or service3 . Simply put, it is principally impossible to copy
Manchester United FC. One could of course try to copy the player qualities
one by one, the organization, manager skills, training skills, even localization,
but the final product will still not be MUFC. Copying the music of Beatles
or Rolling Stones is actually relatively easy to do, but the market potential
of such a strategy is still far from that of the real product.
Hence, we may conclude, and this is to a certain extent more relevant for
event production than other types of production, that potentially increased
globalization and increased global competitive environments do not necessar-
ily float as freely into event production as into traditional manufacturing or
service production4 . Actually, it seems reasonable to assume that for events,
the possibility of maintaining such brand advantages may be possible in spite
of a future “perfectly competitive world”.
2
Obviously, the Event concept has not been properly discussed or defined at this stage.
We will, however, return to a more thorough discussion in the next section, but for the
moment it will prove sufficient to think of an event as a gathering of people with a certain
objective, which could be entertainment (sports, music, theatre etc.) or knowledge (con-
ferences) and with some sort of organization - the event producer - supplying the event
content.
3
In terms of economic theory we could perhaps see it as parts of a product which has
the potential of a viable monopolistic advantage.
4
The terms manufacturing and service production will be discussed in detail in Sec-
tion 1.2
1.1 What is Logistics? 19
Even if the above arguments indicate that logistics as such may be viewed
as less important in a future competitive setting, it does of course not indicate
that logistics problems are irrelevant in event production. On the contrary,
organizing big “one-shot” events such as Olympic games possess huge and
very challenging logistics problems related to classical logistics topics such
as inventory management, production planning, transportation and infras-
tructural planning. Even relatively small (and size-wise insignificant) events
such as a small music festival in a small country like Norway holds numerous
logistic challenges.
1.1.3 Logistics theory, practice and research

An alternative view on Logistics
From an academic point of view, logistics theory may seem unclear and hard
to define. The fact that most textbooks in the topic presents their own
definition of logistics emphasizes such difficulties. However, logistics as a
topic is still very much oriented towards how to perform certain parts of
company tasks efficiently.
Personally, I prefer to view logistics as an integral and extended part of
microeconomic theory. A classic part of microeconomics deals with produc-
tion theory and the development of supply curves. The simple version of this
story typically starts with the concept of a production function; say:
f (N ) = X (1.1)
In equation (1.1), N denotes (a single) input (we can think of it as labour)

while X denotes a single output - produced amount of one product. This
equation defines what is referred to as production technology and a certain
cost of acquiring the input is of course also present. To make it as easy as
possible say;
C(N ) = CF + w · N (1.2)
C(N ) is hence the cost of hiring the labour N and it is assumed linear
with a fixed part CF and a proportional part where w is naturally interpreted
as the wage. Now, the “microeconomic story” continues, by an assumption
of profit-maximising behaviour by the firm and existence of a given market
price p, as follows: (Π() is company profits)
Π(X, N ) = p · X (CF + w · N ) (1.3)

20 Introduction
Then, by applying equation (1.1), (1.3) can be expressed (as a single

valued function in N ):
Π(N ) = p · f (N ) (CF + w · N ) (1.4)

First order conditions for maximization of Π(), Π0 (N ) = 0 then gives:
w
f 0 (N ) = (1.5)
p
Then, a supply curve can be derived through the following argument:
Inverting equation (1.5) (Solving it with respect to N );
w w
N = g( ) and utilising equation (1.1) once more X = h( ) (1.6)
p p
The point here is of course not the beautiful argument leading to supply
curves - a necessary building block in general equilibrium theory - but the
basic assumption of equation (1.1) or as shown in figure 1.1
Figure 1.1: The simplified production function of microeconomics
The point here is the extreme simplicity of the mechanism defined through
figure 1.1. In a microeconomic context, it is surely easy to extend to both
multiple inputs as well as outputs, but the more fine tuned (and in certain
instances) important company decisions such as number of set-ups, inventory
volumes, safety stock and so on will clearly not fit easily into the above theory.
As such, logistics may be viewed as a more fine tuned way of modelling
production functions. That is, the simple functional relationship defined
by figure 1.1 is simply too simplified to spawn essential decisions for most
real-world companies.
1.2 Services, Manufacturing and Events 21
Logistics research
The above logistics “defintion” makes understanding (quantitative) research

in the field comprehensible. If logistics is viewed as a simple extension of mi-
croeconomic production theory with the aim of a more fine-grained modelling
of production functions, it should not come as a surprise that optimization is
a key research concept. The fact that many obvious logistics problems (and
models) naturally contain integer variables (set-up and ordering for instance)
and uncertainty related to vital model input, leads to a need for optimiza-
tion techniques model as well as solution-wise. As modelling typically is
not viewed (by most researchers) as very difficult, it is again fairly obvious
to expect that Logistics research should emphasize solution techniques for
Logistics-oriented optimization problems. Indeed, this seems to be the situa-
tion. Many classical Logistics models such as LSP (Lot-Size Problem), CLSP
(Capacitated Lot-Size Problem), VRP (Vehicle routing Problem) and many
many others, have been extensively studied in research literature. The main
focus of these studies have, as indicated above, been on solution techniques.
As discrete optimization problems typically need computers to achieve so-
lutions in reasonable time, much of the focus has been on algorithmic tech-
niques.
In this text, such algorithmic techniques will play a minor role. On the
contrary, we focus on modelling issues. This seems natural for the intention
of the book, but also as Event Logistics raises some different and relevant
questions related to how classical Logistics models can and should be “trans-
lated” into the Event-setting. This does not indicate that solution related
research is less relevant within this topic, more an observed need to look
into the basic modelling first. The fact that most students may disagree on
the above mentioned easiness of mathematical modelling is of course also a
relevant textbook argument.
1.2 Services, Manufacturing and Events
At this point, it seems necessary to investigate the “Event”- concept some-

what closer. However, before we address this concept, it may be fruitful to
look into more classical logistics, and discuss the concepts of Manufacturing
and Services.
The Business Dictionary [2] defines these two terms as:
22 Introduction
Manufacturing:
“Includes all steps necessary to convert raw materials, compo-

nents, or parts into finished goods that meet a customer’s ex-
pectations or specifications. Manufacturing commonly employs a
man-machine set-up with division of labour in a large scale pro-
duction.”
and
Services:
“Intangible products that are not goods (tangible products), such

as accounting, banking, cleaning, consultancy, education, insur-
ance, know how, medical treatment, transportation. Sometimes
services are difficult to identify because they are closely associ-
ated with a good; such as the combination of a diagnosis with the
administration of a medicine. No transfer of possession or own-
ership takes place when services are sold, and they (1) cannot be
stored or transported, (2) are instantly perishable, and (3) come
into existence at the time they are bought and consumed.”
The above definitions tell us that most human economic activity that is
not defined as Manufacturing may be labelled Services. Obviously, manufac-
turing is related to physical goods and the ownership transferability as well
as storage possibilities. Typically, most services do not have such properties,
but very often parts of services; the report of a lawyer, the prescription or
an X-ray photo of the physician or the DVD of a concert are clearly man-
ufactured goods. Still, the main product without this added manufacturing
possess the main characteristics of a service.
Our main interest here is of course the “Event” concept. It seems fair
to categorize most events within the services category. After all, our com-
mon understanding of the concept implies individuals selling certain services
to spectators like music, sports or theatre. Still, it cannot be taken that
all events naturally fall within the services category. Think, for instance,
of a painting sales exhibition. The paintings are sold (hopefully), implying
ownership transferability. Furthermore, the paintings are physical objects in-
volving storage possibilities. Still, many painting exhibitions might easily fall
into our meaning of events. Consequently, the events concept may perhaps
not fall easily into the service category as a sub-group.
The Business Dictionary [2] also defines Events:
1.2 Services, Manufacturing and Events 23
Events:
“Occurrence happening at a determinable time and place, with or

without the participation of human agents. It may be a part of a
chain of occurrences as an effect of a preceding occurrence and as
the cause of a succeeding occurrence.”
This definition, still not very clear, adds an important point; time and
place. An event takes place in time and place not necessarily tomorrow or
next week, but at a predefined location at a specific point in time (or a set
of specific points in time). So, this sheds light on our art sales exhibition.
An art dealer shop is (typically) not an event as it is available all the time,
while our sales exhibition takes place only a limited time period and may as
such be categorized within the event category.
So, what else? Well, we may talk about “one-shot” events or repeated
(regular) events. The Molde International Jazz Festival will typically qualify
as a repeated event; it takes place more than once, but it is perhaps not
an ordinary service, as it is not present all the time. Olympic games or the
upcoming5 cross-country skiing world championship in Oslo are typical “one-
shot” events. Not necessarily meaning that they will not take place again.
After all, this skiing WC is the third one in Oslo since 19666 . The point is
simply that the market does not know if and when such an event will return.
This sub-categorization is important from a Logistics planning point of
view. It should be relatively obvious that “one-shot” events are significantly
more challenging than regular events when it comes to all phases of logistics
planning and operation.
Another sub-categorization that might be relevant for Event Logistics is
that of entertainment versus professional events. Clearly, when we think
about events, our first thought is perhaps related to the entertainment in-
dustry. However, many events such as research conferences or seminars have
all characteristics of events and may (and should) clearly be defined as such.
The main difference between the two is perhaps on the demand side, which
may behave significantly different. Still, this difference may not lead to too
much differences from an analytic (logistic) point of view.
Let me try to sum things up a bit. Clearly, most events are named events
because they are not available all the time. Certain events are predictable
in time and place. We know, for instance, that Molde football club will play
15 home matches in the Tippeligaen next season, but definitely not at the
5
At the time of writing.
6
Cross country World Championships were arranged in Oslo in 1966, 1982 as well as
now in 2010.
24 Introduction
location or choice of the market. Others are not very predictable in space;
think about the recent WC football at Russia and Qatar. As such, the fact
that an event is not necessarily present at any point in time may perhaps be
the most striking feature of the concept. The actual content of the event or
to what degree it can be placed within the service category is, as I see it, not
the main point. Consequently, the time and place dimension will be leading
our further analysis of Events and Logistics.
1.3 Event Logistics

Given the previous discussion, it seems straightforward to accept that Event
logistics should handle logistics planning and the special challenges that the
event-setting brings.
The simple fact that events takes place at (possibly unpredictable) points
in space and time obviously imposes special problems in relation to classical
logistics modelling.
For instance, the obvious lack of data related to historic demand should
force alternative forecasting methods. The fact that most events products are
difficult (if not impossibly) hard to store should change the inventory focus
on production planning models. The uncertainty related to demand forecasts
together with limited event horizons should make significant changes to the
focus of inventory management modelling. (The News-boy focus seems to
be an interesting candidate.) The fact that in most events, consumers are
brought to the product as opposed to the traditional manufacturing situa-
tion should indicate different transportation challenges, which also must be
reflected in the transportation modelling choices. The fact that many events
are effective monopolies must be reflected in a sensible handling of the sub-
ject. It opens up some interesting possibilities but at the same time creates
some additional problems.
So, this book will discuss, present and in some instances provide solutions
to classical logistics problems, tuned to fit the event-setting.
1.4 Events and Uncertainty

It should not be very hard to realize that the above discussion indicates
that uncertainty most certainly will play a significantly more important role
in event logistics planning than in traditional logistics planning. Arranging
Olympic Games is, as briefly discussed above, probably not carried out at the
given spot in the actual country for many years - if ever. As a consequence,
1.4 Events and Uncertainty 25
the ability to predict both customer demand as well as supply parameters

(costs for instance) forms critical and very challenging problems. This lack
of certainty both in demand as well as in supply makes the concept of un-
certainty far more relevant in events than traditional logistics planning.
The added fact that many of the actual events may take place in a rel-
atively distant future does of course only strengthen the above argument.
After all, the football world cup in Qatar in 2022 is defined and placed in
2010, 12 years ahead of the actual event. The simple fact that many of the
events that need planning (especially mega-events perhaps) lie in a distant
future makes planning not only possible, but also harder. Both future costs
as well as workforce availability may be very hard to predict many years in
advance
The fact that the above situation defines a long time period between the
knowledge of getting the event and the event time also opens up some added
possibilities, pre-sales for instance. This is something which is typical for
most events, and should in principle make certain parts of logistics planning
easier. On the other hand, it opens up for other (and potentially) more
complex opportunities; dynamic pricing for instance. The ability to pre-sell
tickets to an event surely also opens up the possibility of changing the price
virtually very close to the event. (This is often referred to as Dynamic Pric-
ing, Demand Based Management or Revenue Management. A topic which
will be closely examined in Chapter 7.)
Chapter 2
Event Forecasting
2.1 Introduction
All logistics planning needs demand forecast data. Even the most extreme
JIT-production environment, with a maximal flexible production system and
virtually no set-up times or costs, must at some level predict future demand.
This is kind of obvious for events, as most resources needed to produce the
event normally is proportional to the size of the audience. Typical examples
are food, drinks, housing, seating, transportation and so on.
2.2 The fallacy of traditional time-series-

based forecasting
Most logistics textbooks (see e.g. Nahmias [21]) recommend time-series based
forecasting methods. The reason for such a recommendation is quite obvious.
In a modern manufacturing setting, a huge number of products leads to
demand for fast, efficient and reasonably good forecasts. The huge product
count as well as the need for fast and efficient forecast updates makes (simple)
time-series methods appropriate. Even if such methods (perhaps) does not
produce forecasts with the highest accuracy nor provide any explanatory
power, the simplicity and speed of production makes these methods natural
candidates for modern corporate logistics modelling. The fact that historic
demand data are readily available (at almost zero cost) is surely also a good
argument for this choice.
However, moving into the event-situation, obvious differences exist.
Firstly, the number of products is limited. Additionally, the need for speedy
forecast updates is obviously not there, at least not to the extent of a running
28 Event Forecasting
business selling products in their market continuously. (There is normally

some real time between events even in the regular or repeated case.) The
fact (discussed above) that events takes place relatively seldom should also
indicate a certain lack of sufficient demand data. A typical yearly festival
(existing for 20 years) will at best have 20 observations of total demand,
which from a time-series modelling point of view might be sparse.
But maybe more important, time-series modelling does not (at all) ac-
count for a lot of information we normally have related to events. If we plan
an event say (some kind of music festival), we would at certain points in time
have registered pre-sales (normally a very good indicator for total sales), in
addition, we have means of affecting demand through advertising as well as
pricing, which definitely should be taken into account when demand fore-
casting is aimed at.
Such arguments should then lead to different model choices in events
as opposed to traditional logistics forecasting for manufacturing situations.
Causal (regression) type models seems very much more appropriate as they
open up for explanatory information.
2.2.1 A simple example

Let us look at a simple example.
Table 2.1 holds total (yearly) audience for the local theatre -“Teatret
Vårt” [4] in the time period 2000 – 2009.
YEAR AUDIENCE YEAR AUDIENCE

2000 28897 2005 31436
2001 38092 2006 37923
2002 39306 2007 39451
2003 34184 2008 31861
2004 50951 2009 29398
Table 2.1: Yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009
In figure 2.1, the data in table 2.1 is plotted. Examining the figure, we
observe some (expected) variations between years, but a peculiar bump in
2004. Actually, the number of sold tickets in 2004 was close to 30% larger
than the second largest year (2007), and more than 40% larger than the
period average.
Examining figure 2.1 further, it is fairly obvious that any seasonal varia-
tions are hard to justify. After all, why should some kind of cyclical pattern
2.2 The fallacy of traditional time-series-based forecasting 29
Historical demand for Teatret Vårt: 2000 - 2009

60000
50000
40000
30000
20000
10000
0
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Figure 2.1: Plot of yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009
be available over years7 .

The point here is simple. If no trend or seasonality is present when time-
series models are applied, unforeseen bumps are hard to account for. To
visualize this, some simple moving average forecasting models are incorpo-
rated and shown in figure 2.28 .
As can be seen from figure 2.2, the bump in 2004 is missed by all moving
average models, and the effect is being recreated in subsequent years, leading
to a very bad forecast fit. This is of course not very surprising knowing the
mechanism behind such simple time-series models, still, it points out the
typical problems involved in applying time-series modelling in event demand
forecasting.
7
Some kind of cyclical patterns over years may of course be present, for instance,
related to leadership changes. Changing top management will in this business very often
be triggered by low demand rates and top management changes may be observed as
relatively long term cycles. However, predicting such management changes in the future
(and the effect of them) may be more or less impossible.
8
Refer to Appendix A for the actual moving average calculations
Moving average forecasting

60000
50000
Original data
40000
2-order Moving
30000 Average
20000 3-order Moving

Average
Average
0
1998 2000 2002 2004 2006 2008 2010
Figure 2.2: Moving average (orders 2, 3 and 4) forecasting on the Teatret

Vårt data.
2.2.2 Utilizing explanatory information - the reason

behind the 2004 bump
As indicated above, in certain event situations, the explanatory simplicity is
so obvious (perhaps typically as opposed to traditional manufacturing situ-
ations) that such information both could and should be incorporated in the
forecast modelling. Applying some local knowledge 9 , it turns out that a very
special event took place in Molde in 2004. Then, a special play was staged -
“90-metersbakken” written by the local (and later very famous author, mu-
sician and former MFK player) Jo Nesbø10 , and even if one did not manage
to predict the success upfront, at least such information can and should be
applied in retrospect.
The point is of course that this type of play still have and did have a
very special (and positive) demand effect. But, and this is the point here,
unless the theatre plans to do something similar in the future, such an event
9
The brother of local Associate Professor Oskar Solenes is an actor, and did perform
in a certain play at Teatret Vårt in Molde in 2004 - thanks to Oskar for this great piece
of information.
10
He is actually visible on the front page of this book in the top left corner with a guitar.
2.2 The fallacy of traditional time-series-based forecasting 31
should, from a predictive point of view, perhaps be considered an outlier and

in some sense removed from the data.
A very simplistic strategy could be to remove the demand in 2004 and
replace it by the average demand (in red)11 as indicated by table 2.2. Now,
we apply additional information of explanatory character, which of course
is a sensible thing to do if the aim is to build a forecast model that should
provide forecasts with reasonable quality
YEAR AUDIENCE YEAR AUDIENCE

2000 28897 2005 31436
2001 38092 2006 37923
2002 39306 2007 39451
2003 34184 2008 31861
2004 34505 2009 29398
Table 2.2: Yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009
As can be observed from figure 2.3, things look a little bit better in the
sense that the simple time-series based forecasts hit better. However, this
approach to event forecasting is still not good. The main reason is obvious.
We do have much and very relevant information, which we, in many instances,
control ourselves, that are not (and should be) included in the analysis. In
this case, we choose which actors to employ, which plays to stage, which
days to play, how many shows, the prices and marketing and so on. All this
information is simply not used at all by using a time series approach. And,
as this type of information is typically available (more or less) at different
time-spots before the actual event takes place, the possibility of using it
(efficiently) is interesting.
The answer to this demand would be avoiding time-series based models
and instead focus more on causal models (regression models). Such an ap-
proach seems far more appropriate for event forecasting. In order to demon-
strate this approach, a case from the local football club - MFK will be pre-
sented in the next section.
11 28897+38092+39306+34184+31436+37923+39451+31861+29398
This number is calculated as 9 =
34505.333 ⇡ 34505
Moving averages for "adjusted" data

45000
40000
35000 Adjusted demand
30000
Average
20000
Average
10000
4-order Moving
5000
Average
0
1998 2000 2002 2004 2006 2008 2010
Figure 2.3: Moving average (orders 2, 3 and 4) forecasting on the adjusted

Teatret Vårt data.
2.3 The case of football demand forecasting

2.3.1 An old MFK case
Background
The local football team, Molde fotball klubb (MFK), has a glorious history
in Norwegian football. The team from the tiny city Molde (around 25000
inhabitants), holds the sixth place on the Norwegian marathon table (All
time premier division table) [6] only beaten by large city teams such as
VIF (Oslo) , RBK (Trondheim), Brann (Bergen) Viking (Stavanger) and
Lillestrøm (Oslo area).
MFK played in the Champions League (CL) in the 1999 season, meeting
Olympiakos (Greece), Porto (Portugal) and Real Madrid (Spain). The point
score for MFK returned only 3 points, (a home win against Olympiakos),
but it is still the only Norwegian team besides Rosenborg BK who has ever
participated in this tournament.
As such, this club is interesting to investigate by itself, in many ways being
a paradox through it’s continuous success over many years while located in a
very small city. It is probably safe to say that MFK comes from the smallest
2.3 The case of football demand forecasting 33
city ever to host a football team in CL.

Here, however, we will examine a case of demand forecasting through
formal regression modelling to demonstrate how it can be done. The case is
not recent, but stems from a presentation that the author gave at a football
seminar in Norway (NFF’s toppfotballseminar - Gardermoen) back in 2002.
Attendance - MFK home matches 1995 - 2001

14000,00
12000,00
10000,00
8000,00
6000,00
4000,00
2000,00
0,00
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91
Figure 2.4: MFK - Home match attendance, 1995 – 2001
Figure 2.4 holds home match attendance for MFK during the period
between 1995 and 2001. As can be readily observed from the figure, apart
from the relatively large variation between matches, something seemed to
happen around observation 40 (39 to be exact). This final observation is
perhaps easier to observe if average attendance before and after this time is
calculated and plotted alongside the original observations. This is done in
figure 2.5.
As figure 2.5 indicates, something must have happened around observa-
tion 39 or April 1998, which is the corresponding date. Actually, the exact
difference in the two averages amounts to 7274 4397 = 2877 or a percentual
(average) increase of around 65%. Obviously, something did happen. The
happening was a new stadium - today named “Aker Stadion”, nominated for
FIABCIs Prix d’Excellence in 1999 and winner of the Norwegian City prize
the same year.
Attendance and average attendance at MFK home

games in the period from 1995 - 2001
14000,00
12000,00
10000,00
8000,00
6000,00
4000,00
Original Match
2000,00 attendance
Average before and
0,00 after observation 39
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91
Figure 2.5: MFK - Home match attendance, 1995 – 2001 with average values
before and after observation 39
A regression model
A linear multiple regression model postulates a causal type of relationship
between an output variable (Yi ) and a set of N input variables Xi1 . . . , XiN .
On mathematical form: (it may prove handy also to define the number of
observations say M , hence i 2 1, . . . , M )
N
X
Yi = 0 + j Xij + ✏i (2.1)
j=1
The point here is not to dig deep into regression theory, but to look at it
as a case. As such, additional information related the theory may be found
in most logistics type of books (in a simple form) [21] or more advanced
specialized texts like e.g. [15].
Roughly, the application of regression modelling in forecasting can be
described through a 3 step process:
1) Establish your model (define Y and X1 , . . . , XN ).
2) Estimate regression parameters 0, . . . , N.

3) Use the finished model (after completing stages 1) and 2)) to find fore-
casts for a suitable future time period.
Perhaps the most challenging stage is the first one. Ideally, one would like
(from a scientific point of view) to have some underlying theory defining the
Y and the X’s. A classical example from physics may be helpful. Applying
Newton’s second law of mechanics, the following equation can be used to find
the distance s needed in vacuum for some object falling s length units:
1
s = gt2 (2.2)
2
Now, an experiment can be staged, where various values of s can be
defined, 1 m, 2 m and so on. And the time spent in falls t1 , t2 etc. can
be measured. Consequently, a set of s’s and t’s are the outcome of the
experiment. All the t’s can be squared and we can define Yi = si and Xi = t2i .
Then, the following version of (2.1) can be formulated:
Yi = 0 + 1 Xi + ✏i (2.3)
possibly with the added constraint 0 = 0. Then, after staging the regres-
sion process, the interesting unknown g (standard gravity) can be calculated
by the estimated regression parameter ˆ1 through ĝ = 2 · ˆ1 .
This way of doing regression analysis is per se “theoretically correct”.
However, most practical situations involves neither the ability to control
the output variable (e.g. MFK cannot (or will not) define the attendance
at a match) nor do we have a unified theory defining what variables (and
what relationships between them) explaining MFK attendance demand. Still,
various theories may provide sensible variables to be included in a model. As
such, most regression cases within economic or logistics theory may prove
fairly inadequate from a stringent theoretical point of view.
A regression model for MFK attendance

For the case at hand, it seems fair to divide the possible factors influencing
MFK attendance demand into (at least) three different groups:
1) Sports factors
2) Economical factors
3) External factors
The sports factors are related to the quality of the team and it’s oppo-
nents. The basic hypothesis is related to two different mechanisms, absolute
and relative quality. We would expect that the better the team is (abso-
lutely) the more people would like to watch it, but in addition, some kind of
sports oriented mechanism related to what is commonly defined as “uncer-
tainty of outcome” should perhaps also play some part. That is, if the team
is relatively too good compared to a given opponent one would expect that
public interest for the match might decrease.
The economic factors should largely be related to price and income.
Higher product prices ought to bring demand down while increased public
income should have an opposite effect. All kinds of marketing and sponsoring
choices by the club might of course also have effects.
Surely, there is a “field” of club decisions in between these two groups
such as coaching and player choices (made by the club) that in many cases
also may influence spectator interest12 .
The final category contains all other possible factors not naturally be-
longing to the two other groups such as weather, TV-match or not, match
day, a new stadium and so on. Back in 2002, I proposed the following list:
• Changed infrastructure (a new stadium for instance)
• Calendar effects (16th of May for instance)
• Derby effects (matches against special teams RBK for instance)
• Home and away teams form (performance in relatively recent matches)
• Home and away teams status (performance in previous seasons)
• Quality relative to each other (table position)
• Week-day and match-time (Sunday vs. Saturday for instance)
• Weather
• TV-match
• Importance of the given match (possibly late seasonal importance) for

good or bad tabular position
12
The recent coach change for Ole G. S. seems for instance to have had a significant
positive effect on demand in MFK at least.
The first thing to observe from the list above is the absolute absence of
economic variables. No prices, seasonal tickets, marketing, and so on. The
reason is simple: lack of this information or, perhaps more correct, a high
acquiring cost. Finding historic ticket prices turned out to be a difficult
task for the given period. At the same time, MFK has not had a very
“active” price policy in the years we discuss here, so it was expected that
these information bits should play a minor part in explaining attendance
demand anyway.
Furthermore, the list contains blue and red elements. The blue elements
were included while the red ones were not. The main reason for not including
the red elements was more out of convenience, I had relatively short time to
prepare the analysis and chose to do things relatively simple.
The actual model in mathematical form looks as:
atti = 0 + 1 · mf k posi + 2 · mot posi + 3 · mot 3sii + (2.4)

4 · mf k f ormi + 5 · mot f ormi + 6 · rbki +
7 · mai 16i + 8 · stadioni + 9 · branni + 10 · jazzi + ✏i
and with the explanation of the actual variables in (2.5) shown in ta-
ble 2.3. The first part of the table contains continuous variables, while the
final part contains binary variables (e.g. variables only taking values of 1 or
0).
Variable Explanation
atti Attendance in match i (i runs from the first match 1995 up to last in 2001)
mf k posi MFK’s position on the league table before match i
mot posi Opponents’ position on the league table before match i
mot 3si Opponent i’s average table position the three last years
mf k f orm MFK’s average points scored in the last 3 matches
mot f orm The opponent’s average points scored in the last 3 matches
rbk A binary variable; 1 if RBK is the opponent, 0 otherwise
mai 16 A binary variable; 1 if the match is played on May 16th., 0 otherwise
stadion A binary variable 0 up until data point 39, 1 otherwise
brann A binary variable; 1 if BRANN is the opponent, 0 otherwise
jazz A binary variable; 1 if match is played during the jazz festival, 0 otherwise
Table 2.3: Explaining the variables of equation (2.5)
To give the reader a more direct feeling for the model and its data, the
first 20 data points are shown in Appendix B. Now, the normal procedure is
to apply some kind of statistical tool (Excel or as in this case SPSS are typical
candidates) to estimate (find values for) the unknown regression parameters
ˆ0 , . . . ˆ10 13 .
The results (parts of the output from SPSS14 ) are shown in figure 2.6:
SPSS: Linear Regression
Unstandardized Standardized t Sig.

Coefficients Coefficients
Model B Std. Error Beta
1 (Constant) 4107,276 955,029 4,301 ,000
MFK_POS -46,677 72,333 -,045 -,645 ,521
MOT_POS -67,416 53,200 -,114 -1,267 ,209
MOT_3SI -20,251 45,626 -,035 -,444 ,658
MFK_FORM 408,916 209,162 ,137 1,955 ,054
MOT_FORM -99,246 214,428 -,037 -,463 ,645
RBK 4718,759 697,742 ,493 6,763 ,000
MAI_16 2036,426 632,063 ,198 3,222 ,002
STADION 2800,343 315,814 ,543 8,867 ,000
BRANN 607,613 607,339 ,063 1,000 ,320
JAZZ 1309,096 584,917 ,137 2,238 ,028
a Dependent Variable: ATT, R2=0.720
Figure 2.6: SPSS output from the model
We shall limit our interest in figure 2.6 to the third column: Unstan-
dardized Coefficients and the last one named Sig. The third one contains
the values of the parameters ˆ0 , . . . ˆ10 while the last column contains sig-
nificance probabilities. These significance probabilities tells us (popularly
described) whether it is probable that the actual estimates really are differ-
ent from zero. So a very small value indicates that the parameter itself with
relatively large probability may in fact be zero. As a consequence, those pa-
rameter estimates with (say values larger than 0.1 (90% significance level))
should be removed from the model. In figure 2.6, all these are marked in
grey.
The typical next step would then be to take out the insignificant variables
and rerun the regression model; now as:
13
The use of the notation β̂j is due to the fact that the numerical (estimated) values
principally are different from the model values in equation (2.5)
14
All necessary data for this analysis have been found on the RSSSF-website [6]
atti = 4 · mf k f ormi + 6 · rbki +

7 · mai 16i + 8 · stadioni + 10 · jazzi + ✏i
The results of this operation are shown in figure 2.7.
SPSS: Linear Regression
Unstandardized Standardized t Sig.

Coefficients Coefficients
Model B Std. Error Beta
1 (Constant) 2978,880 378,094 7,879 ,000
MFK_FORM 476,102 183,154 ,160 2,599 ,011
RBK 5156,216 579,584 ,539 8,896 ,000
MAI_16 1860,794 623,045 ,181 2,987 ,004
STADION 2815,065 309,117 ,546 9,107 ,000
JAZZ 1265,718 572,701 ,132 2,210 ,030
a Dependent Variable: ATT, R2=0.700
Figure 2.7: SPSS output from the reduced model
Let us take a slight look at the information on the bottom of figures 2.6
and 2.7. The R2 is an interesting statistic to judge. Roughly, it tells us how
much of the variation is explained through our model. Comparison of the two
figures, shows a very minor decrease in R2 indicating that the model (2.5) is
a better choice than the big original model. The fact that R2 is more or less
unchanged after the removal of a set of insignificant variables is normally
taken as a “proof” of reasonable modelling. The final model’s R2 of 0.7
means that 70% of the total variation is explained by the model. Comparing
such a figure to most real world regression models indicates a surprisingly
good fit, especially as many obvious relevant variables are excluded initially.
If we sum up the results so far, it seems as if MFK audience only cares
about the home team’s form, the opponent’s form is irrelevant. The same
holds for the status of the opponent (mot 3si), which also is insignificant
apart from the very significant opponent RBK. So, the only “sporting” or
event oriented variable that is included is the home team’s form. So, one
could say that MFK audience care more about special events (16th of may
and the jazz festival) and less about the game itself. The fact that table
position before the match both for MFK and the opponent plays no signif-
icant role is perhaps somewhat surprising, but does to some sense confirm
the fact that MFK audience is less interested in football than showing up at
the “right” matches.
The basic results of the model could perhaps be summed up as follows:
If MFK want to maximise its attendance, they should play all

matches at home on the 16th of May, the jazz festival should be
relocated from July to the May-week including the 16th of May,
MFK must have won the 3 previous games, and all games should
be played against RBK.
Forecasting the model

Now, let us move to the main point: applying the model to produce fore-
casts. Let us assume that we do not know the attendance values beginning
with the first MFK home game in 200215 . Then, the question to study in
this paragraph is that of finding forecasts for “future” home matches in 2002.
The typical problem with applying regression models in forecasting is that
they normally lead to a situation where certain (or in fact all) of the ex-
planatory variables also need to be forecasted. Then, if many explanatory
variables are in the model, and many of them must be forecasted (to pro-
duce the sought forecast), the aggregated uncertainty of forecasting several
variables (as opposed to one) might simply lead to bad forecasts. This is the
classical argument for using time-series models, where the need for forecast-
ing explanatory variables is simply not present. Still, previous arguments
related to the “nature of events” indicates that time-series approaches might
be unsatisfactory, and causal methods might simply be the only relevant
alternative.
Moving into the case at hand, we observe immediately that the explana-
tory variables must be forecasted, but luckily, most of them are relatively
easy to predict with almost 100% precision. Table 2.4 below sums up the
situation.
The first variable, mf k f orm is (given knowledge of the upcoming seasons
match schedule) hardest to predict. Surely, this forecast depends heavily on
what team we look at, and MFK might be one of the harder teams to forecast
15
Obviously we do know these values of today, but this thought experiment should still
be relevant.
Variable Forecast accuracy

mf k f orm Poor forecast accuracy
rbk 100% prediction accuracy (when the match schedule is given)
mai 16 100% prediction accuracy (when the match schedule is given)
stadion 100% prediction accuracy (without new stadium plans)
jazz 100% prediction accuracy (when the match schedule is given)
Table 2.4: Forecast accuracy of explanatory variables
this variable for in Norway. Note also, that the difficulty of using these types
of models for forecasting purposes depends a great deal on what variables we
turn up including in the model. Roughly in this case, only one variable seems
hard to predict. But remember that the original formulation also includes
other teams form-variables – a much harder forecasting problem.
Anyway, let us return to the present case. The table below (2.5) sums up
MFK’s form for the “observed” seasons:
Season (t) 95 96 97 98 99 00 01
Nt 26 26 26 26 26 26 26
Pt 47 33 45 54 50 40 44
Pt
Nt
1.8 1.3 1.7 2.1 1.9 1.5 1.7
Table 2.5: MFK (point) performance
The first row in table 2.5 contains the observed seasons. The second row
contains number of matches (Nt ) in each of the observed seasons. The third
row contains number of points (Pt ) obtained in each season, while the final
Pt
row contains computed average point score per season ( N t
). Now, our task
is to find an estimate (in principle dynamic) on the 3 last games average
for the next (upcoming) season. It seems very difficult (though perhaps not
impossible knowing the schedule) to produce individual match forecasts. So,
top make things simple we settle for a static (constant) value for our proposed
forecast for mf k f orm. Various options exists ranging from a total period
average up to using only the previous season. In this case, either of these
estimates turns out to be the same as can be seen from the simple average
calculation below:
2001
1 X Pt 1.8 + 1.3 + 1.7 + 2.1 + 1.9 + 1.5 + 1.7
= ⇡ 1.7 (2.5)
7 t=1995 Nt 7
Now, we can build our forecasts for the upcoming season. We assume that
we need to forecast all home matches and that next seasons match schedule
is available. This schedule is normally available well ahead of beginning of
the season, typically before Christmas. Table 2.6 shows this information (the
sequence of home matches for MFK) for the 2002 season:
Match Schedule Variables

1. BRANN -
2. VIKING -
3. LSK -
4. MOSS mai 16 = 1
5. VIF -
6. STABÆK -
7. ODD jazz = 1
8. LYN -
9. RBK rbk = 1
10. BRYNE -
11. START -
12. B/G -
13. SOGNDAL -
Table 2.6: Home match schedule - 2002 season
Now, the necessary information to build the forecast for all home games
in the upcoming season is available. Moving back to figure 2.7 we observe
that the constant has a value of 2978.85. This means that no matter what
(model-wise), at least this number will show up. The same holds for the
stadium variable which adds 2815.065 to the attendance forecast. Conse-
quently, under a reasonable assumption of no significant changes in the sta-
dium infrastructure, 2978.85+2815.06 = 5793.91 will show up independently
of any other variables. Now, the effect of the performance quality of MFK
can be added if we accept the argument above on an average point score
forecast of 1.7. We get:
BaseF orecast = 5793.91 + 1.7 · 476.102 = 6601.5834 ⇡ 6602 (2.6)

Now, apart from the match against RBK (adding 6156.218) to the value
of equation 2.6, the match at the 16th of May (adding 1860.794) and the
match under the jazz festival (adding 1265.718) the value from 2.6 defines
our forecast. Summing up, our forecast for the upcoming season is shown in
table 2.7 alongside the observed attendance figures.
Match Schedule Forecast Attendance

1. BRANN 6602 6204
2. VIKING 6602 5236
3. LSK 6602 5055
4. MOSS 8462 4817
5. VIF 6602 6868
6. STABÆK 6602 7810
7. ODD 7868 8137
8. LYN 6602 4902
9. RBK 11758 11167
10. BRYNE 6602 4303
11. START 6602 5057
12. B/G 6602 6105
13. SOGNDAL 6602 4850
Table 2.7: Forecasted and observed attendance - 2002 seasons
It is perhaps easier to analyse the quality of our forecast model by plotting

the forecasts and the observed attendances in the same figure, as shown in
figure 2.8:
Examining figure 2.8, we observe reasonably good fit apart from a singular
observation, the home match at the 16th. of May. This match (against MOSS
FK) produced a surprisingly low attendance. This could of course be due to
weather or maybe even an unattractive opponent, but still, apart from this
point, the model performs surprisingly well as I see it. Remember that all
information used to construct these forecasts are historic info and a relatively
naive (simple) way of forecasting the mf kf orm variable.
Short term forecasting

In this section (to demonstrate the technique), we will investigate short-term
forecasting on these type of models. In previous paragraphs, we constructed
forecasts for the whole upcoming season made at a single point in time. An
alternative forecasting technique (here named short term forecasting) could
be constructed by computing forecasts from match to match. Normally, such
an approach opens up for producing better forecasts, but we are of course
not guaranteed such an outcome in a given case. Still, the technique itself is
relevant and will be demonstrated shortly.
First, however, a few words on applicability of these two different ap-
proaches. A short term forecast is interesting to apply if the planning or
Comparing forecasts and attendance

figures for the 2002 season
14000
12000
10000
8000
Forecast
6000
Attendance
4000
2000
0
1 2 3 4 5 6 7 8 9 10 11 12 13
Figure 2.8: Forecasts vs. observed attendance for the 2002 season
decision we need the forecast for, can be executed before the next home
game. In a logistics setting, such decisions will typically relate to lead times.
Suppose we need to order soda for the next home match, and this soda can
be ordered now and be delivered within a time period sufficiently short (the
lead time) to reach the next home match. Obviously, we will not need fore-
casts for the rest of the upcoming home matches for the decision. However,
other decisions, stadium capacity, for instance, or other decisions of more
long term character (TV rights, sponsor money etc.) might need forecasts
for longer periods of time.
In this case, the only difference in the methods relates to the values used
for the mf k f orm variable. Given a situation, where we produce forecasts
from match to match, the forecast for the next match then leads to a different
information availability compared to the situation above. (Recall that we
used a very simple average over seasons for our value of 1.7 for this variable.)
Now, standing immediately before a match, we know the point score
obtained by the home team in the 3 previous matches. Consequently, we
can use this information and avoid using the forecasted constant 1.7-value,
and can instead compute “correct” values for this variable. Surely, we need
more information, but this is readily available at [6]. Table 2.8 holds the
necessary information. (In addition, we need the three last home matches of
the previous season, which turn out to produce mf k f orm = 1.3316 .)
Match H/A P mf k f orm Match H/A P mf k f orm

1 H 3 1.33 14 A 0 -
2 A 1 - 15 H 1 1.00
3 H 3 1.67 16 A 1 -
4 A 0 - 17 H 3 0.67
5 H 1 1.33 18 A 3 -
6 A 0 - 19 H 0 2.33
7 A 3 - 20 H 3 2.00
8 H 3 1.33 21 A 3 -
9 A 3 - 22 H 3 2.00
10 H 3 3.00 23 A 3 -
11 A 3 - 24 H 3 3.00
12 H 3 3.00 25 A 0 -
13 A 0 - 26 H 1 2.00
Table 2.8: MFK - complete performance statistics for the 2002 season
As can be observed from table 2.8, MFK had a very decent season in
2002, gaining a total of 50 points ending up being second. What is inter-
esting from our point of view, is the ability to produce dynamical forecasts
for the mf k f orm-variable. The calculations are straightforward. Recall the
definition - the average points scored in the 3 previous games. As a conse-
quence, the first home game (match 1 in the table) is calculated on previous
season data (see above.) The second home match (match 3 in the table),
is then computed by the two first games of the 2002 season (1+3=4 points)
and adding the draw (1 point) from the previous season, giving the value of
1+3+1
3
⇡ 1.67. The rest of the numbers are calculated similarly.
Now, the short term forecasts (ST Ft ) can easily be computed by the
following formula: (t runs over all home matches in table 2.8.)
8
>
> 5793.91 + 476.102 · mf k f ormt 8t\{8, 15, 19}
5793.91 + 476.102 · mf k f ormt + 1860.794 t 2 {8}
<
STFt =
> 5793.91 + 476.102 · mf k
> f ormt + 1265, 718 t 2 {15}
5793.91 + 476.102 · mf k f ormt + 5156.216 t 2 {19}
:
(2.7)
16
MFK gathered 4 points in matches 24 (3), 25 (0) and 26 (1) in the 2001 season. As a
consequence, the mf k f orm variable is computed as 34 ⇡ 1.33
Applying equation (2.7) with the data from table 2.8, results are obtained
as shown in table 2.9:
Match Schedule STF Attendance

1. BRANN 6427 6204
2. VIKING 6589 5236
3. LSK 6427 5055
4. MOSS 8288 4817
5. VIF 7222 6868
6. STABÆK 7222 7810
7. ODD 7536 8137
8. LYN 6113 4902
9. RBK 12059 11167
10. BRYNE 6746 4303
11. START 6746 5057
12. B/G 7222 6105
13. SOGNDAL 6746 4850
Table 2.9: Short term forecasts and observed attendance - 2002 seasons
Comparing these short term forecasts with the original ones as well as the
historical attendance data is easily done through a figure, as in figure 2.9.
It is easily observed from figure 2.9 that the short term forecasts are
initially better. The green curve is closer to the red curve (though not very
much). In the mid parts we observe some variations, but from around home
match 5 up to 7, the new short term forecasts perform better. However,
by the end of the season, we see the opposite situation, where the original
forecasts perform better than the short term ones. So, it is not obvious which
of these two forecast methods that will turn out to be the best one. In order
to compare with more exactness, some kind of error measures needs to be
calculated. In logistics, MAD17 (Mean Absolute deviation) is normally used,
and in order to shed some more light on this comparison, MAD for the two
models are computed in table 2.10.
As table 2.10 indicates, both methods perform relatively equally, but the
short term (presumably best) forecasts perform worst. So our initial hypoth-
esis of utilizing more and better data leading to general forecast improvement
in the one-step (short term) final method did not work out. In fact this is
a good example, because it shows that there are no guarantees when fore-
PN
17
M AD = N1 t=1 |Ft Dt | with Dt being actual demand, N number of observations
and Ft forecasted demand.
Comparing Short and Long term forecasts

14000
12000
10000
8000
Forecast LT
Attendance
6000
Forecast ST
4000
2000
0
1 2 3 4 5 6 7 8 9 10 11 12 13
Figure 2.9: Short and long term forecasts compared to actual attendance -
2002
Match Schedule MAD LTF MAD STF

1. BRANN 398 233
2. VIKING 1366 1353
3. LSK 1547 1372
4. MOSS 3645 3471
5. VIF 266 354
6. STABÆK 1208 588
7. ODD 269 601
8. LYN 1700 1211
9. RBK 591 892
10. BRYNE 2299 2443
11. START 1545 1689
12. B/G 497 1117
13. SOGNDAL 1752 1896
MAD 1314 1324
Table 2.10: Short term forecasts and observed attendance - 2002 seasons
casting is the business. However, the reason for these somewhat unexpected
results is actually very easy to find (in retrospect). This season, the 2002
season was a season where the big favourite (RBK) started out very poorly
and the dark horse LYN lead the league most of the season. MFK, however,
started out reasonably and stayed second most of the season. It might be that
this season, turned on fans expectations for a league win (the first one ever),
and the form variable turns out to improve the forecasts in the early parts
of the season. However, during the season, LYN deteriorated (slowly), while
the big favourite RBK started to improve. It might well be that the MFK
audience lost their hopes (especially I believe) after the home loss against
RBK in home match 9. As such, one might have expected an overestimation
of attendance figures by the end of the season - especially by the short term
model. The facts are that MFK actually played very well by the end of the
season. But it might very well be that the audience had kind of lost hope
for the first ever victory. This can kind of be observed from table 2.9 for
matches, 10, 11, 12 and 13 where the short term forecasts systematically
overestimates attendance compared to the long term forecasts.
This (retrospective) hypothesis can of course be relatively easily tested
by restricting the MAD calculations up to and including home match 9, the
match against RBK. Doing so, we find a MAD of 1221 in the long term case,
and 1118 in the short term case, a MAD (significant) decrease around 10%,
which of course corresponds better with our initial hypothesis.
2.3.2 The effect of Pre-sales

Pre-ordering or pre-sales may be defined as deciding to buy or buy a product
before it is available for sale. As events tend to take place at discrete time
spaces with “some air in between”, pre-sales is possible and interesting for
event producers. Actually, this is more of a norm than an exception in this
area. In traditional manufacturing, this is perhaps more of an exception, but
we observe certain products utilizing these mechanism nowadays. Typically,
it may be hyped products such as announced computer games, or certain new
technology like IPADs or hyped cell phones. But, we also see this occurring in
more traditional markets; cars, for instance, may be pre-ordered these days.
From a logistic and forecast point of view, this is obviously a benefit. If you
pre sell seats for a concert, then you know (with certainty) that at least this
number will turn up or at least you know with certainty that the given ticket
money will be certain. This is a typical situation in football matches where
seasonal tickets guarantee the income, but may not guarantee that persons
show up.
Pre-sales products may take different forms. They may be restricted to a
single event or more than one event. Seasonal tickets pre-sells a whole home
season of matches, and certain special combined tickets (day-passes) are sold
in festivals covering up more than one event. This latter products are often
referred to as bundles (bundling) in economic theory.
Some (relatively limited) research has been done mainly on time-series
modelling - see for instance [10] and [17]. We will not pursue this models
further, but limit our discussion to causal (regression-type) models. We
choose to do this, not only due to the complexity of the time-series cases,
but also due to the “event-arguments” presented previously.
Obviously, we see different types of pre-sale situations, not only various
bundling options as discussed above, but also different choices when it comes
to numbers and time periods. Certain event-producers may choose to limit
the possible number of pre-sold tickets to less than ticket capacity others may
not. Certain event producers may choose to define a pre-sales period ending
in reasonable time before the normal sales takes place. A football club such
as MFK is a good example of both of these types of situations.
MFK’s home ground - Aker Stadion - has a capacity of 11167[1] today.
Still, the stadium record is 1330818 from a game against RBK (who else)
back in 1998 [1].
MFK chose (before this season) to limit the number of seasonal tickets to
less than the capacity. Surely, this is a kind of luxury problem (e.g. having
the option of selling out capacity on pre-sales), but certain football clubs
have this option (typically the biggest and most popular, like MUFC and
Barcelona. But due to some special circumstances this season, (hiring a
new coach named Ole G. Solskjær) this option seemed to pop up for MFK.
Anyway, they chose to limit the number of pre-sold tickets. Such a strategy,
given that demand is bigger than supply, will of course also lead to the second
type of situation; that is, pre-sales ends before regular sales starts.
So, as of today, (December 2010), MFK has sold out their defined avail-
able seasonal tickets and as a consequence, they know (with certainty) that
the total demand for all next year’s matches will be larger than or equal to
this number. What they do not know with certainty is how many of these
seasonal ticket customers will show up on the stadium in each of the matches.
From a logistical point of view, these numbers are perhaps just as interesting
as the financial consequences of having certain income. After all, most lo-
gistics problems is about handling actual demand at the event when it takes
place. However, to keep complexity at a reasonable level, we assume that all
seasonal ticket customers show up at the stadium in our simplified treatment
18
The reason for this difference is due to varying FIFA/UEFA regulations on the number
of seats.
of the situation.
Let us investigate the situation from a mathematical point of view. Using
the notation from above, total demand for a certain match t, attt , can be
expressed as:
attt = attST NT
t + attt (2.8)
where attSTt and sttN
t
T
hold customers buying seasonal and normal tickets
respectively. Alternatively and slightly more complex, a new decomposition
could be added;
attST
t = attST
t
su
+ attST
t
nsu
(2.9)
where attSTt
su
holds seasonal ticket customers showing up at match t,
STnsu
while attt are not showing up. Now, suppose MFK (or any general event
producer) are able to perform the decomposition (2.8) on historical data.
The meaning of this statement is simply that they have registered not only
the total demand, but one of the other components of equation (2.8) as well
- typically this would be attSTt . Then, the following strategy for adjusting a
regression model to adopt pre-sales situations is feasible:
1) Compute attN
t
T
by 2.8
2) Establish a regression model with attN

t
T
as the dependent (Yi ) variable
3) Predict attN
t
T
by the model from 2) and add the known (or forecasted)
ST
attt to produce a final forecast for attt .
A similar procedure can of course be defined if information of the decom-

position (2.9) is available.
There are, however, certain methodological problems related to the above
outlined procedure. Firstly, information may be lacking and there may be
certain differences between various situations. For instance, seasonal tickets
may show very different price profiles over time. MFK is a very good example
on this. As such, the informational content about the share of seasonal tickets
between seasons may be highly questionable. Secondly, and probably more
important, there is an obvious possibility of dependence (correlation) between
seasonal and non-seasonal tickets. Suppose MFK have chosen a relatively
low price strategy for seasonal tickets in a given season. Then (clearly)
one would expect less non-seasonal tickets sold in most matches as opposed
to the opposite situation. Such logic will typically produce methodological
problems in applying simple regression analysis. Without going into details,
the answer is then to do things differently, and model a system of equations
instead of a single equation. Such an approach is (surely) feasible, but far

ahead into relatively advanced econometric theory, clearly outside the scope
of this text.
Consequently, a different approach may be more suitable. Suppose (just
as an example) that MFK this season has sold 7000 seasonal tickets. Also
suppose that a regression model has produced the forecasts of table 2.7. Even
though the numbers of table 2.7 are invalid for the upcoming season, they
prove a relevant point.
Given the initial assumption of 7000 sold seasonal tickets, a forecast of
6602 is of course silly. If we know that at least 7000 tickets are sold for
all matches, surely we should adjust our forecast to take this information
into consideration. A very naive way of adopting this might be simply to
adjust all forecasts below 7000 up to 7000, but this is neither correct nor
sensible. Remember the basic point in regression, minimizing the squared
errors. A strategy of simply adjusting some forecasts will most probably not
correspond with a model that minimizes the sum of the squared errors.
Hence, we need to adopt a strategy which is bit more logical and slightly
more complex. The general multiple linear regression model is defined in
equation (2.1). Solving with respect to the error terms ✏i yields:
( N
)
X
✏ i = Yi 0+ j Xij (2.10)
j=1
Squaring the error terms yields:

" ( N
)#2
X
✏2i = Yi 0 + j Xij (2.11)
j=1
and defining;
M
X
2
S ( 0, . . . , N) = ✏2i (2.12)
i=1
The classical regression problem can then be formulated as
min S 2 ( 0 , . . . , N) (2.13)
β0 ...,βN
The meaning of equation (2.13) is straightforward. We want to find values

for unknowns 0 , . . . , N which provide the minimal value for the function
S 2 ().
The situation we have formulated above is, however, somewhat different.
We know that the forecasted value of our model should at least be larger
than a certain value (7000 in the example). Hence, this information should
be added to equation (2.13) as a constraint. In general terms, say that the
amount of seasonal cards already sold is named Y. Then, the minimization
problem of equation (2.13) can be reformulated to:19
minβ0 ...,βN S 2 ( 0 , . . . , N ) (2.14)

PN
s.t 0+ j=1 j Xij Y
The seemingly slight difference between equations (2.13) and (2.14) has
significant consequences for solution strategies. The optimization problem
of equation (2.13) is unconstrained on a “well behaved” function with a
well known analytical solution. However, the problem of equation (2.14)
is constrained (a quadratic objective with a linear constraint) and is hence
normally classified as a quadratic programming problem (QP). A QP is
in general relatively easy to solve (normally marginally harder than an LP).
However, neat analytical solutions do in general not exist, and in most cases
we need to apply specialized software, for instance, LINGO, CPLEX or the
Excel solver.
Let me try to sum up a little bit. The fact that events in most situa-
tions (naturally) open up the option of pre-sales, should and will in most
cases lead to better (more accurate) forecasts. However, to achieve these
improved forecasts, the methodology may change and may lead to increased
methodological complexity.
2.3.3 Capacity constraints

Capacity constraints on product demand is another concept typical for
events. As briefly discussed above, (see Subsection 2.3.2), most events will
have certain constraints on maximal attendance. This may be due to physical
constraints, size of the concert hall or stadium or related to various security
means. Even though most manufacturing markets will have upper bounds
on some kind of maximal demand (it may, for instance, be very unlikely to
sell an Ipad to each world citizen), in practice such upper bounds for man-
ufacturers are almost never relevant, interesting or binding as we say in OR
terms20 . As a consequence, this is a typical topic for events. Obviously, it is
19
Recall from regression analysis that the error terms are assumed N o(0, σ✏2 ); hence
when predicting through a regression equation, the expected value of the error terms are
always zero.
20
Actual competitive markets will almost always disqualify a given producer from cap-
turing the whole market
a “luxury problem”. If an event producer is in a market situation where cus-

tomers routinely buy the whole capacity, he should be very happy. Still, in a
forecasting perspective, this information must be considered if high quality
forecasts are to be produced.
As with the lower bound in the previous section, an upper bound must be
added to the basic forecasting model. This will work almost exactly like the
pre-sales case in the previous subsection - the only difference is the choice of
inequality sign in the constraint. Now, our forecast should be forced to be
under the capacity limit. Hence, in the MFK case, a upper bound of 11167
must be added to the simple regression model.
It may be interesting to look back on our previous unconstrained fore-
casting examples. If we examine tables 2.7 and 2.9, we observe two forecasts
actually violating the capacity constraint. Our model produced a long term
forecast of 11758 for the RBK-match in the 2002 season. Clearly, the ca-
pacity constraint of 11167 is violated. The same is observed in the case of
the short term forecast (table 2.9 where the forecast for the same match was
calculated to 12059 - an even greater violation. As this upper limit of 11167
was defined before the 2000 season, surely this information was available and
should be incorporated in the forecasting model.
Again, it may be tempting just to scale down these two forecasts to the
capacity upper bound, but the argument on minimizing sum of squares still
should hold. Hence, an adjusted regression version would look like (Ȳ is the
upper capacity limit)
minβ0 ...,βN S 2 ( 0 , . . . , N ) (2.15)

PN
s.t 0+ j=1 j Xij  Ȳ
Surely, in most situations, both pre-sales and capacity constraints may

be present, and a constrained regression model such as
minβ0 ...,βN S 2 ( 0 , . . . , N ) (2.16)

PN
s.t 0+ j=1 j Xij  Ȳ
PN
0+ j=1 j Xij Y
should be used. Both models (2.15) and (2.16) are QP’s (like the model of
equation (2.14)) and must be solved by mathematical programming software
as opposed to standard regression analysis tools like SPSS, which was applied
to produce the forecasts in tables 2.7 and 2.9.
2.4 Aggregated versus disaggregated fore-

casts
The title aggregated versus disaggregated forecasts relates to different lev-
els of forecasts. In events (as argued previously), the normal situation is a
one-time situation or a situation with a lot of time in between events. As a
consequence, the main problem for a given event demand forecast is perhaps
more the ability to break down an aggregated forecast into a set of disaggre-
gated forecasts. Think about Olympic Games. Surely, it is hard enough to
forecast total demand (number of visitors coming to see the games), but it
is perhaps even harder and also more relevant from a logistics point of view,
to be able to break down the total demand into single event demands. Given
a reasonable forecast for total demand, say FT OT , how does this decompose
into various events as indicated by equation 2.17?
FT OT = F100 meters running + FHandball + . . . + F50 M back swimming

(2.17)
In equation (2.17) an underlying assumption of parallel events is (of
course) inherent.
In general, it is harder to forecast at an disaggregated level than an ag-
gregate level. Think about a car producer. It may be fairly easy to get a
reasonable forecast on the total number of cars sold next year. However,
the various brands, engine sizes, extra equipped and coloured cars may be
much harder to guess. The same applies in disaggregated event forecasting.
Even if you are able to foresee the total number of tickets sold at the Molde
International Jazz Festival, the disaggregated level, that is, who will attend
the Charlie Parker, Dizzie Gillespie, Dexter Gordon21 etc. concerts might be
way tougher.
In a logistics setting, it is these disaggregated forecasts which are the
most relevant. The reason should be obvious. Resource consumption by
event audiences are of course mostly related to singular events; proportional
to audience numbers, and hence in a logistics setting, we are normally more
interested in disaggregated than aggregated forecasts.
A typical (practical) approach could be to use some kind of formal method
to arrive at a forecast for total demand F̂T OT and then either formal or
less formal try to decompose the total demand into singular event forecasts.
Producing some shares, say ↵i , as the share of total demand for singular event
i could be a reasonable approach. Then, given the existence of a forecast-
21
RIP.
2.4 Aggregated versus disaggregated forecasts 55
share ↵
ˆ i , one could produce the wanted disaggregated forecast for singular
event i by:
F̂i = ↵
ˆ i · F̂T OT (2.18)
There is, however, some obvious practical problems involved in arriving at
reasonable ↵ˆ i ’s. At least, we can differentiate between two types of events. 1)
Certain events will have the same content in each event from one arrangement
to another. Sports events will typically be like this. The 100 meters run is
(more or less) the same from one Olympics to another. 2) Other events, like
musical festivals will, however, have different artists on the opening and final
days from one year to another. As a consequence, one must be careful by
copying consumer preferences in between events.
Another problem, typically arriving in sports mega events, is related to
difference in location. Almost always, sports mega events, (Olympic games,
football World Championships etc.) are relocated form one instance to the
other. And very often very far. For football WC’s for instance, it is Europe
one year, Latin America 4 years later, and then perhaps Africa or Asia after
4 new years. Consequently, possible ↵ ˆ i ’s fitting good one year might fit very
bad 4 years after, as the average spectator may change substantially moving
from one continent to another. In such a situation, a combination of infor-
mation related to the actual mega-sports event but paired with attendance
(spectator) behaviour related to localization should perhaps be a guidance
for establishing “good” ↵ ˆ i ’s.
In addition to the above mentioned problems, time keeps moving fast.
Typically, at least a year (musical festivals) and up to 2 or 4 years (sports
mega events) passes between events, and consumer behaviour observed in
one event might change simply due to time.
As a consequence, arriving at good forecasts for arrangements within
events may be hard. Still, the need is obvious to be able to perform logistics
planning.
Chapter 3
Events and Inventory

Management
3.1 Inventory Management - Introduction

3.1.1 The EOQ model
Inventory management deals, in its simplest form, with balancing the trade
off between inventory and order costs. The basic modelling hence assumes
two cost elements; a binary order cost occurring when each order is placed,
and an inventory cost, proportional to inventory levels, (actual or average).
The basic classic model, found by minimizing total (average) inventory
and order costs, often referred to as Wilson’s formula, the square root formula
or the EOQ formula, then states that the optimal order quantity Q⇤ (using
the notation of Nahmias [21]) can be found by:
r
2K
Q⇤ = (3.1)
h
where K is the cost placing an order, h is the cost per unit stored (on
average) and is demand for the actual planning time period.
The modelling leading to the EOQ formula (3.1) makes some serious
assumptions. Among these, the two most relevant are:
1) Constant demand over time
2) Deterministic (perfectly predictable) demand
Both these assumptions are of course highly unrealistic in most practical

situations. Typically, we will not know demand with certainty, and the time
58 Events and Inventory Management
profile of the demand forecasts will not be constant. Our example from
Chapter 2 shows this very clear. Our forecasts did not hit perfectly, and
demand throughout the 2002 season did not turn out to be constant.
Still, the simple EOQ formula has proven itself to be practically valuable
due to its simplicity and as a tool for finding approximate solutions.
3.1.2 Applying the EOQ approximately

Let us return to our long term forecast from table 2.7, as shown in table 3.1
Home game 1 2 3 4 5 6 7
Forecast 6602 6602 6602 8462 6602 6602 7865
Home game 8 9 10 11 12 13 -
Forecast 6602 11758 66022 6602 6602 6602 -
Table 3.1: Forecasted attendance demand for MFK home games - 2002 season
Let us further assume that MFK’s experience is that (on average) 1% of

their spectators buys a bottle of coke during a match, and MFK needs to
determine how much coke to buy and when. The assumption of 1% coke
buyers leads to a simple calculation of future coke demand based on atten-
dance demand by simply dividing all forecasts in table 3.1 by 100 leading to
(rounding the numbers) the coke demand forecasts as shown in table 3.2
Home game 1 2 3 4 5 6 7
Forecast 66 66 66 85 66 66 79
Home game 8 9 10 11 12 13 -
Forecast 66 118 66 66 66 66 -
Table 3.2: Forecasted coke demand for MFK home games - 2002 season
By further assuming that the inventory costs (h) between home games
are $0.2 and order costs (K) each time coke is ordered, amounts to $100,
the EOQ formula can be applied as an approximate solver for finding the
optimal purchase strategy.
In the formula, the only missing information is , average periodic de-
mand, which is easily found by adding all numbers in table 3.2 and dividing
by 13 to obtain the average. We find:
10 · 66 + 85 + 79 + 118
⇡ 72.5 (3.2)
13
3.1 Inventory Management - Introduction 59
And the EOQ formula (3.1) then becomes:

r
2 · 100 · 72.5
Q⇤ = ⇡ 269 (3.3)
0.2
As a consequence, each time we order, we should order 269 bottles of coke.
In practice, when approximating through the use of this formula, we need to
make some adjustments if actual forecasts are not constant. Table 3.3 below
illustrates this here.
Home game 1 2 3 4 5 6 7
Forecast 66 66 66 85 66 66 79
Ordered 269 - - 269 - - -
Inventory 203 137 71 255 189 123 46
Home game 8 9 10 11 12 13 -
Forecast 66 118 66 66 66 66 -
Ordered 269 - - 269 - - -
Inventory 247 129 63 266 200 184 -
Table 3.3: Purchase plan for coca cola; MFK home games - 2002 season
Note that we have chosen to buy 269 also before home game 11. As our
planning horizon, stops at home game 13, we will end up with “too much”
inventory in the final period. Obviously, we could have adjusted to end up
with a smaller amount of rest inventory after the last home game, but as this
is coke, we might perhaps just as well keep some for the next season.
The total costs of this purchase plan are easy to calculate. We simply
add all inventory numbers together, multiply by the inventory cost of 0.2 and
add 4 times the order cost of 100. This gives a total cost for the approximate
plan of 2113 · 0.1 + 4 · 100 = 611.3
3.1.3 The Lot-sizing approach

Classical Lot-sizing relates to production planning, not Inventory Manage-
ment. Let us, however, examine the classical production lot-size model in a
Mathematical Programming setting. As discussed in the course previously, a
mixed integer linear programming formulation of the simple lot-size problem
is shown below in equations (3.4) – (3.8).
T
X
Min Z = Kt t + h t I t + c t x t (3.4)
t=1
s.t.
xt + I t 1 I t = d t 8t (3.5)
0  x t  Mt t 8t (3.6)
It 0, 8t (3.7)
t 2 {0, 1} 8t (3.8)
In the production planning setting, xt is produced amount, It is stored

amount, t is a binary variable defining if production takes place in time
period t. These variables define the decision variables in the optimization
model. The parameters Kt , ht , ct , T and Mt holds per period Set-up costs,
inventory costs, production costs, number of time periods and “Big M’s”
respectively.
The simple point is that the inventory management or purchasing situa-
tion exactly resembles the situation described by the above model with some
simple reinterpretation of variables and parameters. If set-up costs Kt are
substituted with ordering costs, we achieve the required purchasing logic.
That is, if an order is placed, (and only then) a order cost of Kt occurs.
Furthermore, if production amounts are substituted by ordered amounts and
production costs ct are interpreted as purchase costs, the model above works
perfectly in the inventory management setting.
In order to derive an exact solution, the problem with ht = 0.28t, Kt =
1008t, ct = 08t, T = 13 and dt picked from table 3.2 must be fed into
some kind of solver software. Here, LINGO is used, and the actual LINGO-
formulation as well as the solution file is presented in appendices C and D.
As can be observed (Appendix D), the optimal solution costs 607. A
relatively modest change from the EOQ-approximate solution of 611.3 (see
above). The major structural change between the two solutions is that the
optimal solution involves three purchase points (before home matches 1,5
and 9) as opposed to the 4 of the approximation (before home matches 1, 4,
8 and 11). Still, the EOQ approximation turned out quite nice here, as the
objective improvement was only around 7%.
Also note that by utilizing a similar reinterpretation as above, an addi-
tional intermediate approximative solution could be obtained by the Silver-
Meal heuristic - see Nahmias [21].
3.2 Event Inventory Management - “News-boy” setting 61
3.2 Event Inventory Management - “News-

boy” setting
3.2.1 “News-boy” basics
The above examples, will not be very representative for what we previously
have labelled “one-shot”-events. Both model types above assume the pos-
sibility of storing the actual resource in between events. This assumption
might work good for running event producers such as a football team or a
jazz-club, but not in the case of festivals. Most festivals happen once a year,
and the possibility of storing resources between events may be unavailable,
either due to durability issues or simply due to festival uniqueness. A typical
example on the latter type could be some kind of gadget sold as a proof of
entering the given event. The Molde International Jazz Festival (MIJF) sells,
for instance, t-shirts.
Another argument (also discussed previously) against the above modelling
types, is uncertainty. When the focus is on festivals or mega-sport events with
long time periods between event arranging, one should expect forecasting to
become significantly harder. Then, it might seem far more sensible to try to
take such uncertainty into account - model-wise.
The “News-boy” model seems to be a very adequate candidate for cov-
ering such types of situations. In a classical “News-boy” model, a certain
demand (assumed, uncertain and described by a probabilistic mechanism)
for some product is present. The product is bought at a certain price, and
then sold typically at a higher price in the normal demand period. Fur-
thermore, after the normal (in-event) demand period, the product can still
be sold, but for a less favourable price (typically assumed smaller than the
buying price). Hence, this situation resembles our inherent “one-shot” event
definition. During the event, a certain monopolistic situation occurs, creat-
ing an in-event demand. This demand changes after the event, typically in a
negative direction. The decision to make in the model is the ordering quan-
tity (Q⇤ ). That is, how much to buy of a given resource before the event.
Even though we have used the gadget as a symbol of this type of demand, it
should be fairly easy to see that this situation actually fits most resource pur-
chases for an event, at least those resources which maybe sold in an in-event
market. Food and drinks will of course also have similar characteristics. The
beer is typically more expensive within premises of the event than outside.
As a consequence, this modelling concept ought to be very interesting for
event inventory planning. To some extent, the term inventory management
may be somewhat misleading here - as the model is a single period model,
but the purchasing dimension is present. Hence, even if normal inventory

decisions (how much to store between each model period) is not within the
basic “News-boy” modelling concept, this family of models still tend to be
placed within the inventory management category (see e.g. Nahmias [21]).
Let us examine the results of the basic “News-boy” model. (We skip the
mathematical derivation, and simply refer to Nahmias [21] or Ravindran et.
al. [24] for complete background information.)
Input for the model is given by: (Parameters ↵ and are given lower and
upper bounds for demand outcome respectively.)
f (Q) : Probability density for demand Q 2 [↵, ]

co : overage cost, cost per unit of positive remaining inventory
cu : underage cost, cost per unit of unsatisfied demand
The optimal order quantity Q⇤ is then (in the continuous case) found by
solving;
cu
F (Q⇤ ) = (3.9)
cu + co
R Q∗
where F (Q⇤ ) = a f (Q)dQ. If we choose a discrete probabilistic formu-
lation (e.g. change from the continuous f (Q) to a discrete Px (Q) = P (x 
Q)), the solution changes slightly to:
cu
Px (Q⇤ ) (3.10)
cu + co
3.2.2 A simple T-shirt example

Now, let us illustrate an application of the “News-boy” model from Subsec-
tion 3.2.1 through a simple event-oriented example. MIJF sells t-shirts, one
or several special made for each festival.
One key question for the festival (apart from the pricing policies for the
t-shirts, briefly discussed in Section 3.3) is the amount to order or buy. If
MIJF orders too few, they lose potential sales profits, but if they order too
many, they will have to sell the remaining t-shirts at a lower price. Let us
first assume that the sales of these t-shirts are proportional to attendance
numbers. That is, the more tickets to concerts sold, the more t-shirts are
sold. This seems to be a reasonable assumption, but it may well be that it
is too simplified. Furthermore, let us assume the sales of a single t-shirt to
simplify even more. Let us continue by examining some available statistics.
Table 3.4 contains total ticket sales from 2005 up to 2009 [26].
Year 2005 2006 2007 2008 2009

#Tickets 28074 27081 29889 30173 34704
#T-shirts 9358 9027 9963 10058 11568
Table 3.4: Tickets (and t-shirts) sold at MIJF: 2005 – 2009
Now, in this example, we are not interested in the number of tickets

sold, but the number of t-shirts sold. And our simplifying assumption of a
proportionality between tickets and t-shirts comes handy to construct t-shirt
demand (the final row in table 3.4). Obviously, to arrive at this row, we must
specify this proportionality, and based on some loose information from the
MIJF management, we choose the following: for each third ticket sold, one
t-shirt is sold. As a consequence, the final row is found simply by dividing
the second row by 3.
Looking at table 3.4, we observe a positive trend, but not a clear pattern.
Hence, some uncertainty regarding the possible ticket sales for the upcoming
2010 festival22 seems reasonable.
In order to complete the example, finding the number of t-shirts (Q⇤ ) that
MIJF should order before the (no gone) 50 years anniversary 2010 festival,
we need two bits of additional information:
1) f (Q)
2) cu and co
The only available information for addressing the content of the proba-
bility density function f (Q) is the content of table 3.4. How to do this is
clearly not obvious. Nahmias [21] provides an example where a histogram
is constructed and then a discrete probability density. This might be ac-
ceptable in a situation where you have a reasonable amount of observations.
Unfortunately, our data sources for the ticket sales (and hence also the t-shirt
demand) is far from adequate. We are stuck with only 5 observations, far to
restricted for such an approach. So, what could we do? One approach could
simply be to try to use some continuous density functions roughly spanning
the observed area. Figure 3.1 indicates a possible approach.
If we take a closer look at figure 3.1, we observe that two different den-
sities have been suggested. To the left, a somewhat pessimistic approach -
a uniform distribution meaning that anything in between 9000 and 1300 t-
shirts could be sold with equal probability. To the right, a far more optimistic
22
At the time of writing, the 2010 festival is long gone, and it is possible to obtain the
ticket sales for this year. However, we assume lack of knowledge of this number here.
f(Q) f(Q)
bQ
a+
)=
f(Q
Uniform h
Triangular
Q Q
9000 13000 9000 13000
Figure 3.1: An optimistic and a pessimistic density for t-shirt demand.
(triangular) version is suggested where more probability mass is located to

the right (more sales). In retrospect, perhaps the left uniform alternative
would have been the best, recalling the extremely bad weather during the
first part of the festival. However, thinking as if we are standing before the
festival, a more optimistic approach may have seen to be better - after all,
a 50 years anniversary should indicate some special event with really great
artists and much audience.
The numbers 9000 and 13000 are of course important. They are chosen
on the basis of the minimal and maximal historic observations from table 3.4
roughly by simply rounding down 9358 to 9000 and rounding up 11568 to
13000. The reason for rounding up more on the positive (right) side is again
due to the previous argument on a big anniversary festival. Recall, that the
approach of defining f (Q) on a closed interval [↵, ] effectively means that
the probability of selling t-shirt amounts below ↵ and above is zero.
Now, to be able to apply equation (3.9), the distribution functions FU (Q)
and FT (Q)23 must be found. The first step in a two stage procedure involves
finding mathematical expressions for the densities - fU (Q) and fT (Q). The
structure of the two density functions may be outlined directly as:
⇢
h if Q 2 [9000, 13000]
fU (Q) = (3.11)
0 elsewhere
⇢
a + b · Q if Q 2 [9000, 13000]
fT (Q) = (3.12)
0 elsewhere
23
The subscripts U and T refers to the Uniform and Triangular alternatives respectively.
Then, our task is to identify the three unknown constants h, a and b in

equations (3.11) and (3.12).
h is easy to find, as total probability mass (the area of the rectangle of
to the left in figure 3.1) must equal 1. We find:
h(13000 9000) = 1 ) h = 0.00025 (3.13)

To establish a and b, we must utilize the same (argument as above), as
well as the fact that fT (Q = 9000) = 0. We get:
(13000 9000) · (a + b · 13000)

= 1
2
a + b · 9000 = 0
or slightly rewritten as
a + b · 13000 = 0.0005 (3.14)

a + b · 9000 = 0 (3.15)
The above linear system of equations with two unknowns are easily solved
by subtracting 3.15 from 3.14 giving:
0.0005
(13000 9000) · b = 0.0005 ) b = ) b = 0.000000125 (3.16)
4000
and
a + 0.000000125 · 9000 = 0 ) a = 0.001125 (3.17)

The second step would then involve transforming the density functions
fU (Q), fT (Q) into distribution functions FU (Q), FT (Q) by: (it is easier done
by keeping the original letters (h, a, b) than substituting in numbers)
Z Q
FU (Q) = hdQ (3.18)
α
and
Z Q
FT (Q) = (a + b · Q)dQ (3.19)
α
giving;
Z Q
FU (Q) = hdQ = [hQ]Q
α = h(Q ↵) (3.20)
α
and
Z Q  Q
b b
FT (Q) = (a+b·Q)dQ = aQ + Q2 = a(Q ↵)+ Q2 ↵2 (3.21)
α 2 α 2
Now, the left part of equation (3.9) is taken care of through the two alter-
native expressions for F (Q) in equations (3.20) and (3.21). The remaining
necessary information is then related to obtaining values for cu and co . Luck-
ily24 , this information (at least partially) is available. It turns out that the
t-shirt’s ordinary (in festival) price last year (2010) WAS 180 NOK, while
the post festival price is 30 NOK (see [8]). Unfortunately, we also need in-
formation on the buying price (or production cost) for the t-shirts. This
information is typically harder to obtain. After all, if the festival buys cheap
t-shirts from China and sell them expensively, market knowledge about it
may affect demand negatively. Hence, most event producers in this situation
will be reluctant in providing this information publicly. As a consequence,
we are left by guessing, and a reasonable guess could be 50 NOK as the
total buying/producing costs for each shirt. Now, cu and co can be easily
calculated. cu is calculated as the profit loss of ordering under the actual
demand. If too little is available, a potential profit of 180 50 is lost; hence,
cu = 130. co is the profit loss occurring if one orders too much. In that case
the t-shirts are bought at 50 and sold at 30 leading to aco = 50 30 = 20.
Now, the often referred to critical ratio (right hand side of equation (3.9))
can be computed as:
cu 130
critical ratio = = ⇡ 0.867 (3.22)
cu + co 130 + 20
Hence, our two different proposed models for demand uncertainty lead to
the following equations to be solved for Q⇤U and Q⇤T :
h (Q⇤U ↵) = 0.867 (3.23)

and
b
a (Q⇤T ↵) + Q⇤2
T ↵2 = 0.867 (3.24)
2
24
Thanks to Assistant Professor Olav Hauge for this information
3.3 Extensions of the “News-boy” concept 67
Equation (3.23) is easy to solve. We find:
0.867
Q⇤U = + ↵ = 3466.67 + 9000 ⇡ 12467 (3.25)
h
Equation (3.24) is quadratic in the variable and will hence involve slightly
more complex algebraic manipulation. The solution is left for the reader as
an exercise.
The main point to observe her is how “close” the solution (12467) is to the
upper demand limit of 13000. The reason is obvious, as we loose a lot (130)
on under-ordering, but very little on the opposite (20), it is a good strategy
to order many. Actually, the managing director of the jazz festival (Jan Ole
Otnes) stated in various TV interviews before the festival that they would
order 10000 t-shirts and expecting to get rid of all. Our analysis indicates
a slightly different strategy, but again, we lack a lot of information here as
opposed to being insiders.
3.3 Extensions of the “News-boy” concept

In previous sections, a simple “News-boy” type event-oriented example was
presented. Even though some readers might find this presentation complex
enough, it is important to stress that our modelling in most instances are
ridiculously simplified. Think about the t-shirt example. We assume only
one t-shirt for sale. Reality is different, at the 2010 festival, 2 different t-
shirts (different motives, but the same basic shirt) was sold. This seemingly
simple change may actually have major consequences for our model concept.
It seems fairly obvious, in such a situation, to assume that one t-shirt might
be more popular than the other, and a certain demand correlation mecha-
nism must be added to the model - not a straightforward model change to
implement.
Furthermore, choice of production process (buying finished shorts or split-
ting the production process in parts to utilize postponement) is something to
consider and which surely makes the analysis even far more involved. By pre-
processing t-shirts and trying to postpone finalizing them, one can utilize the
possibility of observing demand during the festival and hence try to increase
finalizing the popular t-shirts as opposed to the less (observed) popular.
The two simple extensions above introduce both multi-period as well as
multi-product “News-boy” problems. Luckily, researchers have spent much
time and effort on introducing different extensions to this problem, so from
an operative point of view, it may be a lot of help to get from research
literature. A nice survey by Khouja [19] distinguishes 11 different extension
categories, involving both multi- period and product situations, as well as

many other.
In addition to the above mentioned possible extensions, another brand
of modern research literature is interesting to consider. Recall from our
example, that the parameters cu and co were assumed parameters, that is,
given exogenous to the model. In reality, certain parts of these parameters
are under control of the decision maker. For instance, the two prices 180
(in-event selling price) and 30 (after-event selling price) can and must be
determined by MIJF. Choosing different values for these will surely change
the problem. Presumably, by increasing 130, less t-shirts at higher prices
are sold during the festival while doing the opposite, more t-shirts are sold
at lower prices. Obviously, the pricing policy must affect total expected
profits. However, to analyse these types of problems within the framework
of “News-boy” models, some demand relations (of stochastic nature) must
be added to the model. Simply spoken. We need to know how (uncertain)
demand changes when prices are changed. This adds even more complexity
to the problem, but has gained some research interest the latter years. The
excellent review by Petruzzi and Dada [23] sums up most existing research
in this area up to date.
3.4 Final comments

The special nature of events indicates that a single period stochastic inven-
tory management model concept (“News-Boy” or Newsvendor) seems as a
good starting point. Still, reality demands (perhaps) far more advanced ap-
proaches than the simple original “News-Boy” model. Doing this is surely
challenging. However, this topic is very much at the research frontier, and
one should expect further enhancements and improvements possibly fitting
events and service production even better. As such, for those interested in
doing research in the field, this “niche” might be an interesting option.
Inventory management in the traditional sense focus a lot on inventory
decision in a periodic setting. This problem area is perhaps less relevant for
events, still, purchasing in general is very relevant for most event producers,
so using the term purchase management is perhaps better from a scientific
point of view. In this text, we have still chosen to use the vocabulary of
traditional logistics.
Chapter 4
Event Production
4.1 Introduction
Classical production planning of the lot-sizing type is not very relevant in

event logistics. The reason ought to be obvious. Deciding how much to
produce of each product, and when to change from one product to another,
is surely not the most relevant event logistics decisions. Artist booking or
artist sequencing may certainly be relevant, but we will treat this subject
separately in Chapter 10.
Aggregate planning or workforce planning modelling may, however, be
relevant in the event situation as well. The ability to plan correct usage of
either an existing workforce, using hired labour or volunteers, may definitely
be a problem that most event producers are facing. Especially, the choice of
volunteers is very relevant for small and medium sized event producers.
Let us start by reinvestigating the classical aggregate production planning
model as outlined by Nahmias [21]:
4.2 The classical aggregate production plan-

ning model
T
X
Min Z = [cH Ht + cF Ft + cI It + cR Pt + cO Ot + cU Ut + cS St ] (4.1)
t=1
70 Event Production
s.t.
W t = W t 1 + Ht F t 8t (4.2)
Pt = Knt Wt + Ot Ut 8t (4.3)
I t = I t 1 + P t + St d t 8t (4.4)
Ht , Ft , It , Ot , Ut , St , Wt , Pt 0 8t (4.5)
The variables and parameters of the model can be described as follows:

Variables:
Z = Objective function value - total cost

Wt = Number of employees (workforce) in time period t
Pt = Amount produced (units) in time period t
Ot = Number of units produced through the use of overtime in time period t
Ut = Number of units produced through the use of undertime in time period t
It = Amount in inventory by the end of time period t
Ht = Number of people hired in time period t
Ft = Number of people fired in time period t
St = Number of units subcontracted in time period t
Parameters:
dt = Demand forecast for time period t

cH = Cost related to hiring a worker
cF = Cost related to firing a worker
cI = Inventory cost per unit per time period of storage
cR = Production cost per unit within regular time
cO = Extra cost per unit related to the usage of overtime
cU = Per unit undertime cost
cS = Per unit subcontracting cost
nt = Number of production days in time period t
K = Number of units produced by a single worker in one day (workforce productivity)
I0 = Initial inventory volume
W0 = Initial workforce volume - number of employees before the planning horizon
T = Planning horizon
4.3 Developing an event model alternative 71
4.3 Developing an event model alternative

The above model is meant for a traditional manufacturing situation where
manufacturing takes place all the time. In an event setting, surely our plan-
ning horizon is limited, but it surely also spans some time. For the MIJF
(and similar events mega-sports events may last from 5 days up to 3 weeks)
the obvious planning horizon is 6 days. But even for one-day events, we can
organize our work-force planning in several time periods, for instance, by
splitting a day in to several time periods of one or several hours.
Now, this model’s25 basic trade-off is between using inventory to keep a
stable work force (if hiring and firing costs are high compared to inventory
costs) or hiring and firing very dynamically in the opposite (cost) situation.
It ought to be fairly obvious that inventory has no place in the event setting.
Remember that this is finished goods inventory, and we cannot choose to
produce today and sell tomorrow (at least unless we open up for TV and
taping). In any case, the basic event situation must be consumed at the
production time, so we need to remodel the inventory parts. This is in
fact very simple, as we simply can remove it. However, doing this means
(principally) to introduce infinite inventory costs in the model above. If we
do so, the solution to that model is obvious. In that case, we will hire and fire
dynamically to meet demand as we move along. As a consequence, keeping
the above model concept unchanged and simply removing inventory variables
makes sense, but produce an obvious model solution.
So, we need to make some relevant changes to the above model to make
it suit the event situation better. We have already discussed that in events,
different groups of workers contribute to a single final product. The interest-
ing practical situation emerges as the cost structures of these groups vary a
lot. Typically, we have professionals (earning a salary) and we have various
groups of volunteers with low salaries - some even costless. In addition (to
make things reasonable), it seems sensible to assume that hiring and firing
costs as well as productivity also may differ between different types of work-
ers. For instance, I would believe that hiring volunteers might be a much
tougher job (at least for a new event) than hiring through payment and the
labour market. The practical meaning of this statement is of course that
hiring costs for volunteers might be higher than for professionals. Another
feature may be on the firing side. If one recruits a set of volunteers (say for
MIJF), then it might be infeasible (high costs) to fire them before the end of
the planning horizon.
In addition to the above arguments, there are issues related to facts such
25
Equations (4.1) – (4.5)
72 Event Production
as this; all worker groups can not perform all tasks involved in the production
process. Such issues may be taken care of model-wise, but will introduce a
kind of multi-product setting as we have to define different tasks within a
single or multi-product setting. So for the time being, we try to keep things
simple and overlook such complicating matters.
Following up the above arguments, we define groups of workers. Let us use
the subscript j for groups, j 2 {1, . . . , J}. Given this assumption, we need
to redefine some cost elements. Suppose we make the following transition:
New Parameters:
cR ! cR
j : Production cost (salary) per unit in workgroup j
cH ! cH
j : Hiring cost per worker in workgroup j
cF ! cFj : Firing cost per worker in workgroup j
K ! Kj : Productivity of worker in workgroup j
Furthermore, we have to redefine some of the variables and get rid of

some of the others. It seems reasonable to take out IT , Ot , Ut as well as St .
The use of over/undertime could be kept, but only for certain professional
groups. However, to keep things simple, we take out the above mentioned
variables. The new set of variables then becomes:
New Variables:
Ht ! Hjt : Number of people hired in workgroup j and time period t

Ft ! Fjt : Number of people fired in workgroup j and time period t
Wt ! Wjt : Number of employees (workforce) in time period t in workgroup j
Pt ! PjT : Amount produced in time period t by worker from workgroup j
Given the above redefinitions, the revised model is easily set up as:
J X
X T
cH F R
⇥ ⇤
Min Z = J Hjt + cj Fjt + cj Pjt (4.6)
j=1 t=1
s.t.
Wjt = Wj,t 1 + Hjt Fjt 8jt (4.7)

Pjt = Kj nt Wjt 8jt (4.8)
XJ
Pjt = dt 8t (4.9)
j=1
Hjt , Fjt , Wjt , Pjt 0 8jt (4.10)
4.4 A simple example 73
The model above (equations (4.6) – (4.10)) is easily explained. The ob-
jective (4.6) adds up all costs through workgroups j and time periods t.
Each work group j in time period t contains workers based on the number of
workers in the previous period, Wj,t 1 , plus the hired amount minus the fired
amount - equation (4.7). The production amount generated by workgroup j
is computed by multiplying the productivity (Kj ) for the group by possible
working days (or hours or whatever might be suitable) and by the number
of workers in the group - equation (4.8). Finally in equation (4.9), the com-
mon produced amount (for the single product) is computed by adding up all
production contribution over all groups to meet demand.
4.4 A simple example

In order to investigate the model a little bit further, we define a very simple
example. Let us assume two workgroups, a professional (expensive) group
named P and a cheap (volunteer) group named V . Table 4.1 defines various
parameters
CjR CjH CjF K

P 10 100 50 2
V 1 500 300 1
Table 4.1: Data for the event aggregated production planning example
We observe from the example of table 4.1 that the P group costs more
salary-wise CPR = 10 as opposed to CVR = 1. On the other hand, both hiring
and firing is significantly more expensive for the V -group. The P -group is
also assumed twice as productive as the V -group.
Furthermore, to keep things simple, we look at a 2-period model where
we assume (perhaps reasonably) that demand increases from the first to the
second period. (Most events tends to put the most significant artists at the
end.) We use d1 = 20 and d2 = 50 to simulate this. Finally, we assume
the organization has 5 professional employees before the event, and zero
volunteers, that is, WP 0 = 5, WV 0 = 0. The linear programming model is
implemented in LINGO, and model formulation and solution is presented in
Appendix E.
Looking at the solution in Appendix E, we observe that in spite of the
cheapness of the volunteers, the optimal solution involves only the usage of
professionals. To test whether the model seems reasonable, we could try
to increase the cost of professionals to see if we get a change in workforce
74 Event Production
assignment. We change CPR from 10 to 1000 to be on the safe side. This

change produces the following solution;
Variable Value Reduced Cost

HP1 0.000000 101.0000
HP2 0.000000 150.0000
FP1 5.000000 0.000000
FP2 0.000000 0.000000
PP1 0.000000 0.000000
PP2 0.000000 0.000000
HV1 20.00000 0.000000
HV2 30.00000 0.000000
FV1 0.000000 300.0000
FV2 0.000000 1300.000
PV1 20.00000 0.000000
PV2 50.00000 0.000000
WP1 0.000000 2047.000
WP2 0.000000 948.0000
WV1 20.00000 0.000000
WV2 50.00000 0.000000
where we observe that the early (now very expensive) professionals are
fired in the first period and the now (relatively) very cheap volunteers are
hired up at the necessary level in both periods 1 and 2. As a consequence,
the basic logic of our model seems to work.
An interesting question to raise is whether this model only produces either
or solutions related to usage of different workforce groups. A proof is left for
the reader as an exercise.
4.5 Final comments

The very simplified model presented in previous sections must be judged
as a starting point for this type of analysis. It seems obvious that more
complex product formulations should be introduced to make such models
closer to realism. The reason why certain event producers mix different work
groups is of course also related to the fact that there are certain competence
demands on certain operations. Sound engineers can and should not be
picked at random for instance. As such, our one task type of model is highly
unrealistic.
Chapter 5
Event Supply Chains
Lately, Supply Chain Management has grown to be a “hot topic” in tradi-

tional logistics. Wikipedia [5] quotes Harlan [11] and defines the term as:
Supply chain management (SCM) is the management of a net-

work of interconnected businesses involved in the ultimate provi-
sion of product and service packages required by end customers.
Supply chain management spans all movement and storage of
raw materials, work-in-process inventory, and finished goods from
point of origin to point of consumption.
As globalization and increased global competition has evolved, special-

ized agents had emerged on the scene, forcing a producer to relate to a vast
number of agents on the supply side. The growth of suppliers and the compe-
tition between them is good for production efficiency, but creates some “new
problems” on the logistics side. The choice of suppliers and the management
of supply chains are suddenly important company logistics decisions. As a
consequence, the need for supply chain management has emerged.
In events, similar structures have evolved. Think about a modern interna-
tional football team. In the old days, players, managers and club staff were
recruited among locals. Today, the picture has changed dramatically, and
even small Norwegian football clubs have their own scouts travelling around
the world looking for international players. A modern football club’s ability
to handle it’s contact net of scouts, agents and clubs far away, might very
well be the factor that defines success or failure.
The same pattern is evident in other events. A modern international
music festival will have to use international performers. These performers
are travelling the world all the time, set out by their managers and book-
ing agencies. The ability to recruit such artists for a given event producer
76 Event Supply Chains
leads to a necessity for having vast and efficient information and networking
structures available.
Within this framework, competition and cooperation becomes relevant
subjects to handle. If the MIJF wants artist A, who is also wanted by
a competing festival in Norway (for instance the Kongsberg International
Jazz Festival), the ability to handle such competitive/cooperative situations
becomes evident. As a consequence, more elaborate tools than ordinary
optimization, is needed. Game theory is such a tool that deals with strategic
interaction between agents.
Applying game theory as a tool for better understanding and handling
the challenges related to competition and collaboration seems to be growing
in modern supply chain management analysis. Certain relevant textbooks
are arriving like, for instance [7], as well as relevant research material. Some
relatively new (personal) research, illustrating how simple game theory may
be applied in a kind of event supply chain management setting might even
be relevant to investigate - [27], [14].
We will not pursue these topics in more detail here as the lack of relevant
event oriented research material is so obvious. Still, it seems fairly easy to
predict that future will hold a lot of interesting research material covering
event supply management.
Chapter 6
Event Transportation
Transportation is a classical problem area within manufacturing. If you

produce cars in Japan, and want to sell them in Europe, you can not expect
your customers to travel down to Japan, and organize transportation of their
new vehicle back home. As a consequence, most manufacturers will have to
organize their own transportation scheme. Such organizing involves many
important decisions, such as choice of volume (how many items and what type
of items to transport at each shipment), mode (which type of transportation,
car, boat, plane etc.), frequency (how often), use of professional transporters
and so on. As a consequence, a lot of effort has been put into formulating and
solving various transportation problems such as the classical transportation
problem, transshipment problems, vehicle routing problems etc. Many of
such problems pose great challenges in solution. Especially, various versions
of the vehicle routing (how to find cost efficient routes given a set of customers
to visit) problem are hard to solve efficiently. As a consequence, a vast
literature in logistics transportation research exists.
In the event setting, things are to some extent turned around. As most
event producers are located somewhere, and their product is immaterial, it
can not be transported to the customers, at least not in the original form.
(Broadcast versions can of course be fed to the customers, but this line
of production falls into traditional manufacturing, and will hence not be
further discussed here.) As a consequence, most small and middle sized event
producers face the opposite situation. Instead of transporting their product
to the customers. They face the problem of transporting their customers to
the product.
This may of course propose similar challenges as for the manufacturer.
However, in most situations, event producers rely on existing public or pri-
vate transportation means to fulfil such transportation demand. Obviously,
certain situations, special locations, or very large (mega) events may lead
78 Event Transportation
to transportation problems that need to be solved, still such problems are

mostly solved by agents other than the event producer. An example may
clarify. Before the 1994 Lillehammer Olympics, analysis of road quality and
forecast demand pointed out a need for better quality and capacity of roads
between Oslo and Lillehammer. As a consequence, a part of the contract
between IOC and the organizers resulted in a major upgrade of this road -
finished immediately before the games. But, and this is the point here. This
road upgrade was made by those who do it all the time, “Vegevesenet” not
a part of the event production team.
The above argument points out that for small and medium sized events,
transportation problems - in the normal sense - do perhaps not classify as
necessary problems to care about for event producers. Surely, if the event is
big, or the location is vast or far away from everything, some resources must
normally be spent on attacking transportation needs, but still in most case,
more by convincing politicians to make the “right” decisions than actually
solving the transportation problem.
Chapter 7
Events and Dynamic Pricing
7.1 Dynamic Pricing - Introduction

In many event markets, black market activity is present. Black market ac-
tivity can be defined as a situation where a certain good which is sold can be
sold again relatively fast to a higher (or lower) price than the buying price.
Black market activity is normally regulated by law to be illegal.
The first situation (selling to a higher price) is the most common one in
event markets. Most of us have been offered tickets to events (immediately)
before the event at prices significantly higher than the price tag on the ticket.
How can this occur? As we have discussed earlier, the option of pre-
selling tickets is good for the event producers because it makes certain logistic
decisions easier. On the other hand, pre-selling tickets opens up a problem for
the event producer, namely that of controlling, understanding and forecasting
future demand. If an event producer guesses wrong on the dynamic demand
situation after pre-selling a ticket, it may well be that demand increases and
opens up for possible profits for a ticket owner. Additionally, certain event
producers are not allowed (or choose not) to change the price of tickets over
time. If this is the situation and a ticket owner observes increased demand,
the option of reselling the ticket may become tempting.
What has this to do with logistics? The question is obviously relevant,
as we already have claimed that pricing decisions typically are not defined
into the logistics toolbox. So, let us imagine another example. Think about
a grocery store, selling food. If we follow the daily demand patters of most
grocery stores, we observe that demand has two peaks. One in the morning,
and one (typically very big) in the afternoon. So, the customers tend to
use our store in a certain predictive but time-varying pattern. Our costs
are (surly) dependent on the number of customers we serve. So, if we could
80 Events and Dynamic Pricing
choose we would probably prefer a constant daily demand pattern (shifts,

and overtime would be minimized then). Now suppose furthermore, that we
are monopolists. If that is the case, we can assume the existence of dynamic
demand curves defining the relation between the price and the quantity for
our customers. Suppose we name such demand curves ft (pt ). That is, if we
choose the price pt in time t, we receive the sales quantity (demand) ft (pt ).
Normal goods would lead to a situation where a high price will return less
sales and vice versa. Then, given these assumptions, it is easy to realize that
a constant demand pattern ought to be obtainable for the grocery store by
simply increasing prices somewhat in the morning, somewhat more in the
evening and perhaps increasing them somewhat in between. The point is
simple. If we can control the price (and quantity; as monopolists we can) we
can exercise this control to affect our demand. And as demand affect logistics
costs, we must coordinate such decisions if, for instance profit maximization
is our objective.
Why do we need an assumption of monopoly? Let us again rethink our
grocery store example. If we exercise our price changing strategy, prices on
our products will be higher in the morning and quite high in the afternoon.
If the price on milk is 100 NOK per litre around 4 o’clock in the afternoon,
surely our customers will go to the next store if it exists (a non-monopolistic
situation). So, the consequence will (in a competitive situation) not be a
stable demand pattern after all.
In the event case, monopoly is more of the standard than the exception.
As such, these ideas should fit events far better than most manufacturing or
service situations. That is of course also the reason why black markets emerge
far more often in event settings than in normal manufacturing situations.
So far, we have argued as if all dynamic pricing situations involve an
increasing demand as time goes by. This is of course not necessarily the
case. Personally, I have experienced (on several occasions) the possibility
(and even the reality) of buying tickets for football games, just before the
game starts, at lower prices than the official ticket price. Surely, if demand
may increase over time, it may also decrease.
Let us try to sum up: given that the event producer has a profit-
maximizing objective, and is able to predict demand curves in a set of future
periods, and is (at least approximately) a monopolist, and if his future period
demand varies, then he should consider dynamic prices on his product. If
he chooses not to, black market activity is a certain consequence, and profit
that could have entered the pockets of the event producer is instead entering
other’s (black market seller’s) pockets.
Surely, most of these assumptions may not fit reality perfectly. For in-
stance, whether event producers are profit maximizers could be strongly ques-
7.2 Dynamic Pricing in Manufacturing 81
tioned. Would IOC naturally be considered profit maximizers, What about

MUFC or MIJF? The ability to forecast future period demand as demand
curves may of course be hard in practice, to judge a monopoly assumption
the same.
Finally, a few words on dynamic pricing in reality. Right before Christ-
mas, Norwegian retailer Elkjøp got heavy media on their Christmas pricing
policy. (See for instance [3].)
The prices (on various electronic equipment) turned out to differ (in one
direction) immediately before and after the initiation of Christmas shopping.
Normally, it is almost impossible to identify a conscious dynamic pricing
policy. However, in this case a secret note revealing the strategy was found.
Most of us know and accept that airline tickets are dynamically priced.
This seeming paradox is interesting if one would try a dynamic pricing pol-
icy in practice. It seems as if customers accept dynamic pricing in certain
markets, not in others. The reason might be habit or the fact that airline
tickets are also dynamically low priced. Anyway, modern technology (mainly
accessible and cheap Internet) opens up the possibility of using dynamic pric-
ing. If prices are to change rapidly over time, it is of course necessary to be
able to communicate such prices to the consumers. (This is not necessarily
related to the fact that a conscious underlying dynamic pricing strategy, but
that the prices must be communicated to be able to get trade.) Personally,
I would not be surprised if more service, manufacturing and especially event
producers start applying dynamic pricing in the near future.
7.2 Dynamic Pricing in Manufacturing

Let us start by looking at a classic manufacturing situation and look back on
the simple lot-size problem we formulated in Section 3.1.3 in equations (3.4)
– (3.8). Now, we adopt the assumptions above, and reformulate the prob-
lem to take pricing into account. Recall that demand is no longer given as
parameters, but endogenously taken care of in the model through a new set
of decision variables - pt price on the product in time period t. In order to
move forward, we must specify the structure of the demand functions. Let
us make it as easy as possible and assume linear demand functions. That is;
dt = ↵t t · pt (7.1)
where dt is the calculated demand if price pt is chosen in time period t
and ↵t and t define the time varying linear demand curves.
Now, to make things reasonable, we need to change the objective to in-
clude revenue. When we allow ourselves to change prices (include prices as
82 Events and Dynamic Pricing
decision variables), it dos not make sense to minimize costs as the original
lot-size model did. Hence, we need to move to profit maximization and as
profits are calculated by subtracting costs from revenue we need to calculate
the per period revenues of this problem. This is straightforward: (quantity
times price)
Rt = (↵t t · pt )pt (7.2)

Now, the pricing version of the lot-size problem of equations (3.4) – (3.8)
can be formulated as:
T
X
Max Z = Rt Kt t ht It ct xt (7.3)
t=1
s.t.
xt + I t 1 I t = d t 8t (7.4)
0  x t  Mt t 8t (7.5)
It 0, 8t (7.6)
t 2 {0, 1} 8t (7.7)
↵t
pt 0 8t (7.8)
t
with dt and Rt defined in equations (7.1) and (7.2).

The above problem (equations (7.3) – (7.8)) was originally formulated
and solved by J. Thomas in 1970 [28]. Thomas proposed a Dynamic
Programming-based algorithm (inspired by Wagner and Whitin’s famous pa-
per [29]) which solves large problem instances very fast.
If we study this problem closer, we observe that the objective is non-
linear; quadratic to be specific, and it contains binary variables. In general,
this means that standard (MI)LP-solvers such as LINGO and CPLEX not
necessarily are able to handle it. Later relevant work includes this author
among others in multi-product and capacitated extensions [12], [13]. Inter-
estingly enough, such extended problems turn out be easier to solve than
their cost minimizing (CLSP) counterparts.
7.3 Dynamic Pricing in services - Revenue

Management
In the previous section - Section 7.2 - we discussed briefly how dynamic
pricing might be adapted to a manufacturing situation including storage
7.3 Dynamic Pricing in services - Revenue Management 83
possibility and set-up. If we move our vision to services, which by the way
resembles events much more, we observe (naturally) that set-up as well as
storage are of less importance. This is of course in accordance with previous
discussion on the topic (see Chapter 4). Hence, a strategy in to adopt this
situation to services or events for that matter could simply be to remove
storage possibilities (take out the It variables) as well as set-up (remove
the t variables). This is of course possible, and if this is done, we are left
with a problem (basically) only containing the objective. This problem is
simply a multi-period (decoupled) classical monopoly, and the solution is
straightforward: (M R = M C)
↵ t ct
p⇤t = (7.9)
2 t
This is of course a very simple solution, and given that reality fits, it could
definitely be applied. However, as usual, reality may not fit. For instance,
in both service and event production, certain dependencies may exist. If one
applies a dynamic pricing strategy say for pre-sales of tickets to a concert
known to the audience, the cheap tickets (if unlimited) will sell fast and
hence affect future demand after this point in time. Furthermore, the ability
to actually predict independent demand curves over time might be difficult
and in some sense fairly infeasible. Given these two arguments, it is perhaps
obvious to think about uncertainty of demand and the concept of Revenue
Management is focusing on this as well as the above modelling features. The
point is basically very simple. If one really wants to apply dynamic pricing
in events, reality may be too complex to just use the simple approach of
equation (7.9). We will not pursue these topics further here, just point at
some survey literature that may be both relevant and helpful in investigating
further options in event dynamic pricing and/or revenue management. The
paper by McGill and Van Ryzin [20] may be a good starting point
Chapter 8
Events and “hype logistics”
8.1 Introduction
Logistics is a somewhat immature discipline and (perhaps especially) the less
quantitative agents on the research and application side tends from time to
time to spend a lot of time on buzz-words. Logistics vocabulary is full of
such buzz-words, as JIT (Just in time), Modularization and Postponement
to name the most important ones. Let me try to be slightly clearer: the inten-
tion here is NOT to say that these words and their meaning is unimportant,
neither for manufacturing, service nor event production agents. Still, one
might get the feeling from time to time that these words introduce ground
breaking news. The main point of this short chapter is to explain the most
relevant concepts and outline their importance for event logistics.
8.2 Just In Time - JIT

Let us start by looking at the model underlying the simple EOQ model and
formulate total costs:
Q
T C(Q) = K +h (8.1)
Q 2
Now let us do exactly the opposite of what we did in Subsection 3.1.3 of
Chapter 3, interpreting K as a set-up cost instead of the normal (Inventory
management) order cost, and hence Qλ as number of set-up’s. In addition, we
need to assume that production runs infinitely fast, but these assumptions
play little or no role in the argument.
Now, the standard argument goes as follows. We minimize total costs by
solving T C 0 (Q) = 0, which produces the EOQ formula (3.1).
86 Events and “hype logistics”
Now, this argument is based on several assumptions, one of them (not

very often discussed by the way) is that K is given and unchangeable. This
K, the set-up cost in our production, is a technology parameter underlying
our technology choices, so in principle we can change it. Normally, changing
it significantly costs a fair sum of money. But still, the option is there. Let
us assume we choose to change it from K to zero. What effect does it have
on our argument? In that case, the total cost function simplifies to:
Q
T C(Q) = h (8.2)
2
and we minimize it by minimizing Q. Minimizing Q (it can of course
not be zero) means a very small Q and negligible average inventory. This is
the basic point in JIT. Keep almost no inventory and produce small series
or demand as it emerges, and this is important. Either, you will have to
invest large sums of money in both technology and the right manpower to be
able to force both set-up costs and times to zero. (Whether this turns out
to profitable is of course another story.) Alternatively, you could be lucky,
living in a part of the world where labour is cheap and flexible, yielding
similar consequences.
Looking at event production, inventory is not a topic for the finished
goods. As such, the main point in JIT of minimizing inventory is simply
not relevant; however, this is important. The above argument also has con-
sequences related to uncertainty in demand. If demand is highly uncertain,
the value of a flexible production system is higher than in the deterministic
case. So, by increasing organizational flexibility the producer also manages to
handle uncertainty much smoother. The necessity of forecasting is of course
(in principle) far less if the producer can meet any kind of demand at any
point in time with the given technology and manpower. An event producer
will in many cases face considerable demand uncertainty. As such, a JIT-type
of production system is good. The reason for choosing volunteers is of course
a part of this, as a volunteer plays the role of a cheap and willing worker,
able to show up at any point in time and do whatever is needed. Hence, JIT
is nothing new in event production, actually it is the oldest concept of all in
the event setting.
8.3 Postponement
Postponement means, as the word indicates, to postpone or delay the produc-
tion process. In a situation with demand uncertainty, it may be a good idea
to try to postpone major parts of production as late as possible or at least
8.4 Modularization 87
as late as necessary until uncertainty is revealed. Think about our example

of selling t-shirts at MIJF from Chapter 3, Subsection 3.2.2. If MIJF are
very much uncertain on t-shirt buyers taste, they could postpone the whole
t-shirt colouring and printing process until they observe what the festival
customers prefer. Or even better, they could stage a poll before the festival,
letting potential t-shirt buyers vote on their preferred t-shirts, letting them
buy it before the festival. This would be considered extreme postponement.
We see a lot of postponement in modern manufactured products. Modern
cell phones have a lot of visual tuning possibilities for backgrounds, menus,
transparency, animation and so on. Why? This is again postponement. In-
stead of guessing how customers would like their cell phones, let them decide
themselves.
So, if postponement is “manna from heaven” why not always do it. The
reason might be lack of uncertainty or costs. Postponing will typically involve
increased costs. T-shirts finished with print will normally be cheaper and
perhaps of better quality than those “home-made” at the festival. Finally,
postponement may not be possible. Think about postponement in a concert
setting. The idea is simple, if you do not know what the audience wants, let
them choose. So in a concert setting, if some of the audience wants Frank
Zappa and others the Beatles at the same time, it becomes kind of complex
to fix. Especially as both Zappa and the Beatles no longer are here. Letting
the home audience in a football match decide that now is the time for a home
goal is a pleasing idea, but still obviously infeasible.
To sum up, in situations with high demand uncertainty, postponement
may be good, but not at all cost. In the event setting, it may be high demand
uncertainty, but pre-sales soften the effect.
8.4 Modularization
Modularization means being able to produce a maximal number of finished
products with a minimal number of components (modules). Apart from the
fact that being able to modularize a production process may be cost efficient
(cheaper than the alternative), it is a strategy to achieve postponement.
If you are able to produce everything based on a small number of basic
components, you can postpone until customer preferences are revealed. The
classical example is a car engine, with which minimal tuning can change
from 100 HP until 350 HP. The single module engine can hence spawn a lot
of different products meeting customer preferences fast.
In events, we actually see modularization resemblance. Again, the local
jazz festival is a brilliant example. In latter years, the festival has chosen
88 Events and “hype logistics”
something referred to as artist in residence. This single artist will have

several concerts of different types (a multitude of products) from a single
module. Presumably, the main point of doing this is cost efficiency rather
than postponement, but the postponement strategy is definitely there. A
good advice for the festival could be to postpone the programme for the
artist in residence shows, letting customer demand decide. So, if Nils P.
Molvær stays until Saturday in the jazz week and the audience at that day
would prefer country music, let him perform it (given that he is able to).
Again, this kind of strategy is harder to foresee in sports events. But in
principle, keeping a minimal number in a football squad with flexible players,
that is, having 15 players, who could play well in all pitch positions could be
a possible modularization strategy in football.
Chapter 9
Event Facility Location
Facility location in classical logistics is related to the problem of choosing

(and designing) locations for production. If we again consider MIJF (or
any other festival for that matter), they have the choice of picking among
a set of possible predefined production locations. In Molde, these locations
may be Teatret Vårt, Alexandria Hotel, Bjørnsonhuset, Idrettens Hus, Molde
Cinema, The new “Teater and Jazz” house”, Aker Stadium and so on.
So, let us assume the existence of a set of locations Li , i 2 1, . . . , I. Let

us furthermore assume that each of these locations has some set of associ-
ated attributes. Such attributes could be costs, capacity, quality, transport
distance etc. Suppose in the absolute simplest possible situation, we con-
sider only cost attributes and define Ci as the cost attribute of location i.
Suppose furthermore that a given event should be assigned to one (and only
one) location at a minimal cost. In a mathematical programming setting,
this task can easily be formulated by introducing a set of binary variables:
say i defined as:
⇢
1 if location i is picked
i = (9.1)
0 elsewhere
90 Event Facility Location
and formulating the following MILP (PIP26 )

I
X
Min Z = i Ci (9.2)
i=1
s.t.
I
X
i =1 (9.3)
i=1
i 2 {0, 1}8i (9.4)

The above problem (equations (9.2) – (9.3)) is of course ridiculously sim-
ple in a Mathematical Programming context. It merely picks out the smallest
of a set of given costs. Hence, it is no point feeding it into LINGO or CPLEX.
Still, as a building block for more advanced situations it may be handy.
Now, let us furthermore assume that it is not only costs but also the
quality of the location that are interesting to judge. Let us assume that
each location has a certain quality attribute, say Qi , where a high value on
Q means better quality and that a given lower bound of quality Q is given
(either by law or by our own choice.). The above problem can then easily be
extended to handle the new situation simply by adding
I
X
i Qi Q (9.5)
i=1
to the problem defined by equations (9.2) – (9.3).

Again, it is easy to derive at a specialized algorithm. Pick the smallest
Ci where (9.4) is satisfied.
A far more challenging situation (solution-wise) emerges if we change
our focus slightly. Let us now assume that we have booked a set of artists
Ai and that a set of possible locations Li are picked. Obviously, the set
of possible locations must be larger than the set of artists if we consider a
parallel set of events, which we indeed intend to do. However, to simplify, let
us furthermore assume that the number of booked artists equals the number
of possible locations I. In order to progress, we also need to make a slight
change in the cost structure. It seems reasonable to assume that artist costs
may be dependent on location. If a certain artist is booked at a low-quality
location, he or she may accept smaller fees. If this is the case, Cij must exist.
That is the cost of assigning artist i to location j 27 . In order to formulate
26
Pure Integer Program
27
It is perhaps easier to see this structure if we look at revenues instead of costs, but
the main logic is unchanged from a model point of view.
91
a model for this situation, we need a slightly more complex binary variable.
We define:
⇢
1 if artist i is located at location j
ij = (9.6)
0 elsewhere
Then, a minimal total cost artist-location-assignment can be found by
solving the following problem:
I X
X I
Min Z = ij Cij (9.7)
i=1 j=1
s.t.
I
X
ij = 18j (9.8)
i=1
I
X
ij = 18i (9.9)
j=1
ij 2 {0, 1}8i, j (9.10)
Above, (9.8) picks one (and only one) artist to each location j, while (9.9)
secures that each artist i gets a location.
The problem defined by equations 9.7 – 9.10 is commonly referred to as
an Assignment Problem in logistics or OR research literature. This problem,
as most name-tagged mathematical programming problems in logistics, has
undergone a vast research effort related to extensions and specialized algo-
rithmic development. Because of space concerns, the chapter concludes here
and the interested reader is directed to the excellent review by Pentico [22].
Chapter 10
Event Sequencing
Event Sequencing may be defined as how to sequence (or schedule) sub events
within a bigger event optimally.
Sequencing in classical manufacturing logistics is related to relatively
short-time or operational decisions and is normally defined in a machine
setting. How should jobs be fed into a single or a set of machines (serial,
parallel or networking structure) to achieve certain targets. In most cases,
such targets are related to efficiency in the form of various measures of speed
or capacity utilization. So a classical machine scheduling problem focuses on
job sequence decisions (sorting jobs) to achieve certain efficiency goals.
In events, typically neither speed, nor capacity utilization is a big issue.
On the other hand, demand is a big issue, and if demand may be affected
by certain event sequencing decisions, then event sequencing might be both
commercially important and challenging.
Identifying sequence demand links on the event scene is straightforward.
Think about a football league, the sequence of games may obviously affect
demand both positively and negatively. A simple example may clarify: sup-
pose the local team should meet the main opponent RBK. If this match (say
a match in Molde) takes place at a point in time (late in the season) where
much of table placement uncertainty is revealed, local attendance demand
may be low. On the other hand, if the table placement (and season timing)
is such that this given match is very important for final placement, demand
might be very high. Strictly speaking, this problem is not solved by the
event producer, but by the league owner or the regulator. It is of course still
relevant, but perhaps of less importance for most event producers.
The same situation is also present for festivals. Normally, demand (in
general) is higher towards the end of the festival28 . Then the question of how
28
Most festivals start before a week-end and ends during a week-end. The same type of
94 Event Sequencing
to sequence the artists may be relevant. For instance, a certain very high
quality artist may draw audience more or less independent of the timing of
his performance, while other artists may need some “calendar help”.
Another point, is the booking situation. Most artists have schedules
already, and the possibility of fitting certain artists into certain specific time-
slots may affect not only local demand, but also costs related to hiring the
artists.
So, we can conclude that event scheduling is relevant. Various mathe-
matical models can now of course be discussed and adapted to the situations
described above. This is an important point, and in order to actually do
this, some critical input information is needed, namely the “demand link”.
In order to have any hope of solving such problems formally, information
on how the schedule (or sequence) affects demand must be quantitatively
present. Such information is hard to establish and in most cases simply un-
available. Obviously, scheduling decisions must be made, and they are made
all the time by all event producers; still a formal analysis of such problems
may demand data that is not present, and (perhaps) very costly obtained29 .
Due to the above arguments, we refrain from more formal modelling ap-
proaches in this chapter. Still, it seems fairly obvious to this author that
most event producers could get far better products if some more systematic
approaches are applied in their event scheduling.
pattern may also be present, even at one-day events, where attendance demand may grow
through the event independently of artist scheduling.
29
If the cost of obtaining necessary input information is very high, the question of
performing formal analysis becomes a trade-off, which does not necessarily point at doing
the analysis as the optimal solution.
Appendices
Appendix A
Calculations in figures 2.2

and 2.3
The actual moving average calculations underlying figures 2.2 and 2.3 are
given in tables A.1 and A.2:
YEAR DEMAND F̂t2 F̂t3 F̂t4

2000 28897
2001 38092
2002 39306 33494,5
2003 34184 38699 35431,67
2004 50951 36745 37194 35119,75
2005 31436 42567,5 41480,33 40633,25
2006 37923 41193,5 38857 38969,25
2007 39451 34679,5 40103,33 38623,5
2008 31861 38687 36270 39940,25
2009 29398 35656 36411,67 35167,75
Table A.1: Moving average calculations underlying figure 2.2
In the above tables, the forecasts F̂tj are computed by the standard for-
mula:
j
1X
F̂tj = Dt i (A.1)
j i=1
where Fˆtj denotes forecasted demand for time period t given moving av-
erage of order j and Dt is observed historic (actual) demand data.
98 Calculations in figures 2.2 and 2.3
YEAR DEMAND F̂t2 F̂t3 F̂t4

2000 28897
2001 38092
2002 39306 33494,5
2003 34184 38699 35431,67
2004 50951 36745 37194 35119,75
2005 31436 34344,5 35998,33 36521,75
2006 37923 32970,5 33375 34857,75
2007 39451 34679,5 34621,33 34512
2008 31861 38687 36270 35828,75
2009 29398 35656 36411,67 35167,75
Table A.2: Moving average calculations underlying figure 2.3

Appendix B
Data for the case in

Subsection 2.3.1
att mfk pos mot pos mot 3si mfk form mot form rbk mai 16 stadion brann jazz
4430,00 1,00 3,00 5,30 3,00 3,00 0,00 0,00 0,00 0,00 0,00
5434,00 1,00 11,00 15,00 3,00 1,00 0,00 0,00 0,00 0,00 0,00
7300,00 1,00 3,00 15,00 3,00 4,30 0,00 1,00 0,00 0,00 0,00
5830,00 2,00 8,00 13,00 2,30 2,00 0,00 0,00 0,00 0,00 0,00
5046,00 2,00 10,00 9,00 1,30 1,30 0,00 0,00 0,00 0,00 0,00
12990,00 2,00 1,00 1,00 2,00 2,30 1,00 0,00 0,00 0,00 0,00
4857,00 2,00 5,00 5,00 1,30 2,00 0,00 0,00 0,00 0,00 1,00
2966,00 2,00 14,00 15,00 1,30 0,00 0,00 0,00 0,00 0,00 0,00
4379,00 2,00 5,00 5,30 2,00 2,00 0,00 0,00 0,00 0,00 0,00
3606,00 2,00 7,00 7,00 1,30 1,30 0,00 0,00 0,00 0,00 0,00
2228,00 2,00 13,00 7,70 0,70 1,00 0,00 0,00 0,00 0,00 0,00
2335,00 2,00 4,00 3,00 1,30 1,00 0,00 0,00 0,00 0,00 0,00
3435,00 2,00 9,00 6,70 0,30 0,70 0,00 0,00 0,00 1,00 0,00
4918,00 13,00 12,00 15,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00
5051,00 9,00 10,00 15,00 1,00 1,00 0,00 0,00 0,00 0,00 0,00
3083,00 7,00 12,00 8,00 1,00 1,30 0,00 0,00 0,00 0,00 0,00
8365,00 4,00 1,00 3,00 3,00 2,00 0,00 1,00 0,00 0,00 0,00
4486,00 4,00 6,00 6,30 2,30 1,30 0,00 0,00 0,00 0,00 0,00
3500,00 3,00 12,00 10,30 1,70 1,30 0,00 0,00 0,00 0,00 0,00
3592,00 5,00 6,00 4,00 1,30 1,00 0,00 0,00 0,00 0,00 0,00
Table B.1: The first twenty data points for the regression analysis
Appendix C
LINGO model in
Subsection 3.1.3
The LINGO version of the MIP-problem of equations (3.4) – (3.8) is presented

below:
Min = 100*d1 + 100*d2 + 100*d3 + 100*d4 + 100*d5 + 100*d6

+ 100*d7 + 100*d8 + 100*d9 +100*d10 + 100*d11 + 100*d12 + 100*d13
+ 0.2*I1 + 0.2*I2 + 0.2*I3 + 0.2*I4 + 0.2*I5 + 0.2*I6 +0.2*I7
+ 0.2*I8 + 0.2*I9 +0.2*I10 +0.2*I11+ 0.2*I12 + 0.2*I13;
X1-I1 = 66;
X2 +I1 - I2 = 66;
X3 +I2 - I3 = 66;
X4 +I3 - I4 = 85;
X5 +I4 - I5 = 66;
X6 +I5 - I6 = 66;
X7 +I6 - I7 = 79;
X8 +I7 - I8 = 66;
X9 +I8 - I9 = 118;
X10 +I9 - I10 = 66;
X11 +I10 - I11 = 66;
X12 +I11 - I12 = 66;
X13 +I12 - I13 = 66;
X1 <= 942*d1;
X2 <= 942*d2;
X3 <= 942*d3;
X4 <= 942*d4;
102 LINGO model in Subsection 3.1.3
X5 <= 942*d5;
X6 <= 942*d6;
X7 <= 942*d7;
X8 <= 942*d8;
X9 <= 942*d9;
X10 <= 942*d10;
X11 <= 942*d11;
X12 <= 942*d12;
X13 <= 942*d13;
@BIN(d1);
@BIN(d2);
@BIN(d3);
@BIN(d4);
@BIN(d5);
@BIN(d6);
@BIN(d7);
@BIN(d8);
@BIN(d9);
@BIN(d10);
@BIN(d11);
@BIN(d12);
@BIN(d13);
Appendix D
Case solution in
Subsection 3.1.3
The LINGO solution for the MIP problem of equations (3.4) – (3.8) with
data for the MFK 2002 season is presented below:
Global optimal solution found.

Objective value: 607.0000
Objective bound: 607.0000
Infeasibilities: 0.000000
Extended solver steps: 23
Total solver iterations: 765
Model Class: MILP
Total variables: 39
Nonlinear variables: 0
Integer variables: 13
Total constraints: 27
Nonlinear constraints: 0
Total nonzeros: 90
Nonlinear nonzeros: 0

D1 1.000000 100.0000
D2 0.000000 -88.40000
D3 0.000000 -276.8000
104 Case solution in Subsection 3.1.3
D4 0.000000 -465.2000
D5 1.000000 100.0000
D6 0.000000 -88.40000
D7 0.000000 -276.8000
D8 0.000000 -465.2000
D9 1.000000 100.0000
D10 0.000000 -88.40000
D11 0.000000 -276.8000
D12 0.000000 -465.2000
D13 0.000000 -653.6000
I1 217.0000 0.000000
I2 151.0000 0.000000
I3 85.00000 0.000000
I4 0.000000 0.8000000
I5 211.0000 0.000000
I6 145.0000 0.000000
I7 66.00000 0.000000
I8 0.000000 0.8000000
I9 264.0000 0.000000
I10 198.0000 0.000000
I11 132.0000 0.000000
I12 66.00000 0.000000
I13 0.000000 1.000000
X1 283.0000 0.000000
X2 0.000000 0.000000
X3 0.000000 0.000000
X4 0.000000 0.000000
X5 277.0000 0.000000
X6 0.000000 0.000000
X7 0.000000 0.000000
X8 0.000000 0.000000
X9 382.0000 0.000000
X10 0.000000 0.000000
X11 0.000000 0.000000
X12 0.000000 0.000000
X13 0.000000 0.000000
Appendix E
LINGO - the problem in

Section 4.4
Formulation:
min = 100*HP1 + 100*HP2 + 1*FP1 + 50*FP2 + 10*PP1 + 10*PP2 +

500*HV1 + 500*HV2 + 300*FV1 + 300*FV2 + 1*PV1 + 1*PV2;
WP1 = 5 + HP1 - FP1;

WP2 = WP1 + HP2 - FP2;
WV1 = 0 + HV1 - FV2;
Wv2 = WV1 + HV2 - FV2;
PP1 = 2*WP1;
PV1 = 1*WV1;
PP2 = 2*WP2;
PV2 = 1*WV2;
PV1 + PP1 = 20;

PV2 + PP2 = 50;
106 LINGO - the problem in Section 4.4
Main parts of solution:
Global optimal solution found.

Objective value: 2700.000

HP1 5.000000 0.000000
HP2 15.00000 0.000000
FP1 0.000000 101.0000
FP2 0.000000 150.0000
PP1 20.00000 0.000000
PP2 50.00000 0.000000
HV1 0.000000 0.000000
HV2 0.000000 9.000000
FV1 0.000000 300.0000
FV2 0.000000 1291.000
PV1 0.000000 0.000000
PV2 0.000000 0.000000
WP1 10.00000 0.000000
WP2 25.00000 0.000000
WV1 0.000000 0.000000
WV2 0.000000 432.0000
Bibliography
[1] Aker stadion – from wikipedia, the free encyclopedia. Retrieved from
the internet: http://en.wikipedia.org/wiki/Aker\_Stadion, 2010.
[2] The business dictionary. Retrieved from the internet: http://www.

businessdictionary.com, 2010.
[3] Forbrukerrådet raser mot elkjøp: - nei, det går ikke an! Retrieved
from the internet: http://www.vg.no/nyheter/innenriks/artikkel.
php?artid=10012513, 2010.
[4] Norsk teater og orkesterforening. Retrieved from the internet: http:

//www.nto.no, 2010.
[5] Supply chain management – from wikipedia, the free encyclopedia.

Retrieved from the internet: http://en.wikipedia.org/wiki/Supply\
_chain\_management, 2010.
[6] L. Aarhus. RSSSF – norwegian football archive. Retrieved from the

internet: http://www.rsssf.no/stats/AlltimePremier.html, 2010.
[7] G. P. Cachon and S. Netessine. Game theory in supply chain analysis. In

P. Gray, editor, Tutorials in Operations Research. INFORMS, Hannover,
MD, 21076 USA, 2006.
[8] Molde International Jazz festival. Jazzshop. Retrieved from the in-
ternet: http://www.moldejazz.no/2003/site/main/produkt.php?page\
_id=2706, 2010.
[9] A. Giddens. The consequences of Modernity. Polity Press, Cambridge,

UK, 1990.
[10] V. M. Guerro and J. A. Elizondo. Forecasting a cumulative variable

using its partially accumulated data. Management Science, 43(6):879–
889, 1997.
108 BIBLIOGRAPHY
[11] C. M. Harlan. Supply Chain Management, Purchasing and Supply Man-

agement, Logistics, Vertical Integration, Materials Management and
Supply Chain Dynamics. In: Slack, N (ed.) Blackwell Encyclopedic
Dictionary of Operations Management, Cambridge, UK, 1996.
[12] K. K. Haugen, A. Olstad, and B. I. Pettersen. The profit maximising

capacitated lot-size (pclsp) problem. European Journal of Operational
Research, (176):165–176, 2007.
[13] K. K. Haugen, A. Olstad, and B. I. Pettersen. Solving large-scale profit

maximization capacitated lot-size problems by heuristic methods. Jour-
nal of Mathematical Modelling and Algorithms, 6(1):135–149, 2007.
[14] K. K. Haugen and H. A. Solberg. The soccer globalization game. Euro-

pean Sport Management Quarterly, 10:307–320, 2010.
[15] L. Kahane. Regression Basics. Sage Publications, Inc., Thousand Oaks,

California, USA, 2001.
[16] R. M. Kanter. Thriving locally in the global economy. Harvard Business

Review, 5(75):151–160, 1995.
[17] S. Kekre and T. E. Morton. Forecasting using partially known demands.

International Journal of Forecasting, 6(1):115–125, 2002.
[18] M. Khor. Globalization of world politics. In J. Baylis and S. Smith,

editors, The Globalization of World Politics, An Introduction to Inter-
national Relations, page 15. Oxford University Press, 1999.
[19] M. Khouja. The single-period (news vendor) problem: literature re-

view and suggestions for further research. OMEGA, The International
Journal of Management Science, 27:537–553, 1999.
[20] J. I. McGill and G. J. Van Ryzin. Revenue management: Research

overview and prospects. Transportation Science, 33(2):233–256, 1999.
[21] S. Nahmias. Production and Operations Analysis. McGraw–Hill, Singa-

pore, 4 edition, 2001.
[22] D. W. Pentico. Assignment problems: A golden anniversary survey.

European Journal of Operational Research, 176(2):774–793, 2007.
[23] N. C. Petruzzi and M. Dada. Pricing and the newsvendor problem: A

review with extensions. Operations Research, 47:537–553, 1999.
BIBLIOGRAPHY 109
[24] A. Ravindran, D. T. Phillips, and J. J. Solberg. Operations Research –

Principles and practice. John Wiley & Sons, New York, USA, 1987.
[25] R. Robertson. Globalization - Social Theory and Global Culture. Sage

Publications, London, UK, 1992.
[26] Statistisk sentralbyrå. Kulturstatistikk. ISBN 978-82-537-8000-9, Oslo-

Kongsvinger, 2010.
[27] H. A. Solberg and K. K. Haugen. The international trade of players in

european club football – the consequences for national teams. Interna-
tional Journal of Sports Marketing & Sponsorship, 10:74–93, 2010.
[28] L. J. Thomas. Price-production decisions with deterministic demand.

Management Science, 18(11):747–750, 1970.
[29] H. M. Wagner and T. M. Whitin. Dynamic version of the economic lot

size model. Management Science, 5(3):89–96, 1958.
Index
“90-metersbakken”, 28 Calendar effects, 34

“News-boy”, 59, 60 Causal, 26
“one-shot”, 21 cell phones, 46
“sexy”, 15 Charlie Parker, 52
China, 64
accounting, 20 classical transportation problem, 75
agents, 73 cleaning, 20
aggregate production planning, 67 CLSP, 19
aggregated forecasts, 52 coca cola, 57
airline tickets, 79 concert hall, 50
Aker Stadion, 47 constants, 63
Aker Stadium, 87 constraint, 51
Alexandria Hotel, 87 consultancy, 20
algorithmic techniques, 19 continuous case, 60
art dealer, 21 country music, 86
Artist booking, 67 CPLEX, 50, 80, 88
artist in residence, 86 critical ratio, 64
artist-location-assignment, 89 Cross country, 21
artists, 53
Asmund Olstad, 5 day-passes, 47
Assignment Problem, 89 demand, 56
attendance demand, 35 demand uncertainty, 64, 84
average periodic demand, 56 Demand-Based Management, 23
density functions, 62, 63
banking, 20 Derby effects, 34
Barcelona, 47 Dexter Gordon, 52
Beatles, 16, 85 disaggregated forecast, 53
Bjørnsonhuset, 87 Discrete Event Simulation, 14
Black market, 77 discrete probabilistic formulation, 60
booking, 92 distribution functions, 62, 63
Branding, 16 Dizzie Gillespie, 52
Brann, 30 DVD, 20
BRYNE, 41 dynamic demand, 77
bundling, 47 dynamic demand curves, 78
112 INDEX
Dynamic Pricing, 23 inventory costs, 56, 69

dynamic pricing, 23, 78, 79 Inventory Management, 14
dynamic pricing strategy, 79 Inventory management, 55
IOC, 76, 79
education, 20 Ipad, 16
electronic equipment, 79
Elkjøp, 79 J. Thomas, 80
EOQ, 55 Jan Ole Otnes, 5
EOQ approximation, 58 Japan, 75
EOQ model, 55, 83 JIT, 25, 83, 84
equilibrium, 18 Jo Nesbø, 28
Event Management, 3
Event Sequencing, 91 Karl Brunvoll, 5
Excel solver, 50 Kongsberg International Jazz Festi-
explanatory variables, 38 val, 74
Facility location, 87 lead time, 42

FIFA, 47 league owner, 91
First order conditions, 18 Lillehammer Olympics, 76
flexible production system, 25, 84 linear demand functions, 79
Forecast accuracy, 39 linear system of equations, 63
Forecasting, 14 LINGO, 50, 71, 80, 88
forecasting, 14 Logistics, 3, 13
Frank Zappa, 85 logistics, 3, 13–15
frequency, 75 logistics decisions, 67
logistics planning, 23
Game theory, 74 long term, 56
game theory, 74 long term forecasts, 46
globalization, 15, 73 lot-size model, 80
globalized, 16 lot-size problem, 57
LSP, 19
Hallgeir Gammelseter, 5 LYN, 46
Halvard Fiksdal, 5
Harry A. Solberg, 5 MAD, 44
Hitler, 13 man-machine, 20
Holger Preuss, 5 Manchester United FC, 16
manufacturing, 14
Idrettens Hus, 87 market price, 17
infrastructural planning, 17 match day, 34
insurance, 20 Match Schedule, 44
Intangible, 20 Mathematical Modelling, 13
integer variables, 19 Mathematical Programming, 57
INDEX 113
mathematical programming software, operational decisions, 91

51 Operations Management, 3, 13
McGill, 81 Operations Research, 13
medical treatment, 20 optimization, 19
MFK, 30, 34, 38, 40, 43, 47 optimization problem, 50
MFK attendance, 33 order costs, 55
microeconomic theory, 17 Oskar Solenes, 28
MIJF, 59, 66, 79, 85 Oslo, 21
MILP, 88 overage cost, 60
minimization problem, 50 overtime, 68
mixed integer linear programming, 57
parallel events, 52
mode, 75
Petter Petterson JR, 5
Modularization, 85
philosophical perspective, 15
Molde, 3
PIP, 88
Molde football club, 21
planning horizon, 69
Molde International Jazz Festival, 5,
Porto, 30
59
Postponement, 84
Molde University College, 3, 13
pre-sales, 51
monopolistic advantage, 16
pre-selling, 77
monopoly, 78
pricing policies, 14
MOSS, 41
Prix d’Excellence, 31
moving average, 27
Production cost, 68
MUFC, 16, 47, 79
production function, 17
Napoleon, 13 Production Planning, 14
networking, 74 production planning, 14, 17
Newsvendor, 66 profit maximization, 80
Newton, 33 profit-maximising behaviour, 17
NFF, 31 purchasing, 66
Nigel Halpern, 5 Qatar, 23
Nils P. Molvær, 86 quadratic, 65
Norway, 17 Qualitative Logistics , 13
Quantitative Logistics, 13, 14
objective, 79 Queuing Theory, 13
ODD, 45
Olav Hauge, 64 RBK, 30, 40, 41, 46, 47, 91
Ole G. Solskjær, 47 Real Madrid, 30
Olympiakos, 30 regression model, 32
Olympic Games, 22 regression modelling, 32
Olympic games, 17 regression parameters, 32
one-day events, 69 regression theory, 32
114 INDEX
regular time, 68 total demand, 52

regulator, 91 trade off, 55
research concept, 19 Transport Planning, 14
revenue, 79 transport planning, 14
Revenue Management, 23, 80 transportation problems, 76
Rolling Stones, 16 transportation research, 75
RSSSF, 36 transshipment problems, 75
Russia, 22 Triangular, 62
TV rights, 42
S. Nahmias, 3 TV-match, 34
sales exhibition, 20
scouts, 73 UEFA, 47
seasonal tickets, 35 uncertainty, 22
service production, 14 underage cost, 60
services, 20 undertime, 68
set-up, 20 Uniform, 62
set-up cost, 84 uniform distribution, 61
SOGNDAL, 45 upper bound, 51
spectators, 20 upper bounds, 50
sponsor, 42 Van Ryzin, 81
SPSS, 51 vehicle routing problems, 75
STABÆK, 41 VIF, 30
stadium, 50 Viking, 30
START, 45 volunteers, 67, 69
sum of squares, 51 VRP, 19
supply chain management, 74
supply chains, 73 work-force planning, 69
supply curve, 18 workforce assignment, 71
supply curves, 18
X-ray, 20
table placement uncertainty, 91
tangible, 20
Teatret Vårt, 87
Teatret Vårt, 28, 30
technology choice, 14
The Business Dictionary, 19
theatre, 16, 20
time-series, 25
time-series modelling, 26
Tippeligaen, 21
Torgeir Dahl, 5

Event Logistics

Uploaded by

Copyright:

Available Formats

Event Logistics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Event Logistics

Uploaded by

Copyright:

Available Formats

2

This publication may not be reproduced, stored in a retrieval system or transmitted

Layout: Kjetil K. Haugen

• Textbooks for higher education

• Research and reference literature

We only use environmentally certified printing houses.

Tapir Academic Press

Publishing Editor: terje.togersen@tapirforlag.no

This book is written to be used as teaching material for a course in Event

Molde, Norway - Brno, Czech R. - Vienna, Austria - Budapest, Hungary

3 Events and Inventory Management 57

5 Event Supply Chains 75

7 Events and Dynamic Pricing 79

8 Events and “hype logistics” 85

9 Event Facility Location 89

A Calculations in figures 2.2 and 2.3 97

B Data for the case in Subsection 2.3.1 99

C LINGO model in Subsection 3.1.3 101

D Case solution in Subsection 3.1.3 103

E LINGO - the problem in Section 4.4 105

1.1 The simplified production function of microeconomics . . . . . 20

3.1 An optimistic and a pessimistic density for t-shirt demand. . . 64

2.1 Yearly ticket sales at Teatret Vårt - Molde; 2000 – 2009 . . . . 28

3.1 Forecasted attendance demand for MFK home games - 2002

4.1 Data for the event aggregated production planning example . 73

A.1 Moving average calculations underlying figure 2.2 . . . . . . . 97

1.1 What is Logistics?

Quantitative Logistics, as the term indicates, focuses on mathematical

Quantitative Logistics or Operations Management may be defined

The above definition might be viewed as very traditional and conserva-

1.1.2 The importance of logistics, now and in the fu-

assume less possibilities for maintaining technological and or design advances

1.1.3 Logistics theory, practice and research

In equation (1.1), N denotes (a single) input (we can think of it as labour)

Π(X, N ) = p · X (CF + w · N ) (1.3)

Then, by applying equation (1.1), (1.3) can be expressed (as a single

Π(N ) = p · f (N ) (CF + w · N ) (1.4)

Figure 1.1: The simplified production function of microeconomics

The above logistics “defintion” makes understanding (quantitative) research

1.2 Services, Manufacturing and Events

At this point, it seems necessary to investigate the “Event”- concept some-

“Includes all steps necessary to convert raw materials, compo-

“Intangible products that are not goods (tangible products), such

“Occurrence happening at a determinable time and place, with or

1.3 Event Logistics

1.4 Events and Uncertainty

the ability to predict both customer demand as well as supply parameters

2.2 The fallacy of traditional time-series-

business selling products in their market continuously. (There is normally

2.2.1 A simple example

YEAR AUDIENCE YEAR AUDIENCE

Historical demand for Teatret Vårt: 2000 - 2009

be available over years7 .

Moving average forecasting

20000 3-order Moving

Figure 2.2: Moving average (orders 2, 3 and 4) forecasting on the Teatret

2.2.2 Utilizing explanatory information - the reason

should, from a predictive point of view, perhaps be considered an outlier and

YEAR AUDIENCE YEAR AUDIENCE

Moving averages for "adjusted" data