Network Analysis

Syllabus
Thomas de Graaff

Version 2.2 – October 17, 2017

Vrije
c Universiteit Amsterdam, Faculty of Economics and Business Administration
Department of Spatial Economics
Preface

This syllabus deals seemingly with a rather eclectic set of topics, with one common denominator;
the presence of networks—being tangible or non-tangible. I decided to write this syllabus because
appropriate literature covering the total course is not yet available in one volume. Moreover, most
of the literature dealing with the presence of especially tangible networks is rather technical and
not deemed suitable material for business or economics students at the master level. Fortunately,
quite recently,—especially in the field of non-tangible consumer networks—several excellent
textbooks have emerged for business students. Unfortunately, they all cover only a minor part of
the whole field. Nevertheless, this course will draw upon the use of one (and the most famous)
of these textbooks: namely Shapiro and Varian (1999) and will draw substantially upon a
few others. This syllabus aims to cover those other interesting topics that are associated with
networks: namely, network sectors, transport sectors and the macroeconomic impacts of the
surge in ICT in especially the late 1990s and 2000s. Moreover, the syllabus complements Shapiro
and Varian as well in those rare cases where we do belief that Shapiro and Varian miss something.
Proper disclaimers are in order here. Chapters 3 and 5 are basically written by Prof. dr.
Erik T. Verhoef, and only slightly moderated by me. Moreover, Chapter 6 draws heavily upon
Gelauff and de Bijl (2000). Finally, Chapter 4 is a rather severe modification of Aalders et al.
(2002). I am the sole responsible for Chapter 2 and of course for all remaining mistakes—being
textual or intrinsical—that can and will still be found.

Thomas de Graaff

Amsterdam, July 8, 2004 (last adapted October 17, 2017)

iii
iv
Contents

Preface iii

Contents iv

1 Introduction 1
1.1 Networks in the economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The impact of networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 A typology of networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Networks and Graphs 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Graph theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Symmetric networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 Asymmetric networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.3 Properties of networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.4 Intermezzo: the symmetric connections model . . . . . . . . . . . . . . . . 15
2.3 Real-world networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Random networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Small-world networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.3 Scale-free networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Network optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Minimum spanning tree: Prim’s algorithm . . . . . . . . . . . . . . . . . . 20
2.4.2 Fully connected networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.3 Hierarchical connected networks . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.4 Routing problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.5 Network evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Economic implications of networks . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.A Matrix multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

v
vi CONTENTS

3 Market Power and Networks 33

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Market power: an introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 A spectrum of market structures . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.2 Price discrimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.3 Monopoly rents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Sources of market power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Market power arising through economies of scale and scope . . . . . . . . 43
3.3.2 Market power arising through government policies . . . . . . . . . . . . . 49
3.3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4 Network Sectors: Privatisation and (de)regulation 53

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 Network sectors in the Netherlands . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Government regulation instruments . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 Social welfare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4.1 Social welfare and regulation . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.5 Market failure and government failure . . . . . . . . . . . . . . . . . . . . . . . . 66
4.5.1 Market failure in network sectors . . . . . . . . . . . . . . . . . . . . . . . 67
4.5.2 Government failure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 An example: the energy sector in the Netherlands . . . . . . . . . . . . . . . . . 72
4.A Integration versus division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.A.1 Reasons for vertical integration . . . . . . . . . . . . . . . . . . . . . . . . 74
4.A.2 Vertical division of network sectors . . . . . . . . . . . . . . . . . . . . . . 76
4.A.3 Effects on social welfare . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.B Natural monopolies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.B.1 The ‘traditional’ view on natural monopolies . . . . . . . . . . . . . . . . 79
4.B.2 A ‘contemporary’ view on natural monopolies . . . . . . . . . . . . . . . . 80
4.B.3 Natural monopolies versus normal monopolies . . . . . . . . . . . . . . . . 82

5 The Transport Sector: Roads 83

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Road traffic congestion in The Netherlands: some facts and figures . . . . . . . . 84
5.2.1 Development over time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2.2 Concentration in space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2.3 Concentration in time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 An economic market approach to the analysis of traffic congestion . . . . . . . . 87
CONTENTS vii

5.3.1 Demand side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

5.3.2 Supply side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.3.3 Equilibrium and optimum . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.4 Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4.1 The optimal congestion toll . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4.2 The efficiency of non-price policies: an example of ‘government failures’ . 101
5.4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.5 The social acceptability of congestion pricing . . . . . . . . . . . . . . . . . . . . 104
5.5.1 Some theoretical considerations . . . . . . . . . . . . . . . . . . . . . . . . 105
5.5.2 Some empirical evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6 Macroeconomic implications of ICT 109

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2 ICT as a general purpose technology . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.3 Productivity and economic growth . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.3.1 The empirics: The United States . . . . . . . . . . . . . . . . . . . . . . . 113
6.3.2 The empirics: The Netherlands . . . . . . . . . . . . . . . . . . . . . . . . 115
6.3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.4 Market behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.4.1 Markets for ICT goods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4.2 Markets for information goods . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4.3 Markets in general . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.5 The end of the business cycle? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.6 Lower structural inflation? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.7 Policy implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Bibliography 122
Chapter 1

Introduction

1.1 Networks in the economy

This syllabus deals with networks, the various forms of networks, the cause of network formation
and the consequences. It does so from an economic perspective, not so much from a technical,
social or optimization perspective. For those perspectives I refer to other courses.
If you think about it for a few seconds, networks are ubiquitous in everyday life. You
tap your water from a network of pipelines. You travel by train over a network of railways
and telephone calls are—or better, were—made through a network of telephone cables. And
these are just examples where the network is quite visible in the forms of pipelines, rails, and
cables. There are other numerous examples where the network is only visible in the form of
points (the so-called nodes). Withdrawing money from an ATM, sending somebody a letter,
and using a mobile phone does not directly involve a physical network. However, most people
will immediately acknowledge that these examples still involve distribution and transmission
networks from banks, post offices, and aerials, respectively, despite the fact that physical lines,
cables and infrastructure are missing (at first sight).
Besides these more obvious forms of networks, where the networks themselves are crucial for
the goods or services delivered (without rails no trains), there are other types of networks, which
play a more indirect but an equally, and perhaps even more, important role. Such networks may
be formed by a variety of entities, from which the following two have been extensively discussed
in economic literature.
Firstly, firms form networks with suppliers, clients, employees, and even with other firms. If
many firms cluster together this may create spillover effects. Namely, firms may benefit from
specialized suppliers, the pool of specialized employees is larger, and firms may learn from each
other, because of the frequent contacts between their employees (the ‘Silicon Valley’ effect).
Secondly, consumers may also form networks by simply buying the same product. Thus,
there is a network of Microsoft Windows 10 users and a network of Linux users. Both groups do
not only differ in their computer operating system, but also in the type of computer magazines

1
2 Chapter 1. Introduction

they buy, the websites they visit, the type of additional software (e.g., Microsoft Offices vs.
LATEX) they require, etcetera.
The networks above do not directly affect the value of the good firms produce and consumers
consume. However, indirectly, these kinds of networks are oftentimes of the utmost importance.
Recall the example just given about the firms in Silicon Valley. Almost all these firms witnessed
tremendous growth, mainly because they were in the right place at the right time. To give
another example; the value of Betamax video recorders for consumers diminished heavily, because
almost all other consumers bought VHS-type of video recorders, resulting in a severe supply cut
of prerecorded Betamax videotapes.

1.2 The impact of networks

So, the occurrence of networks is rather common in the economy, but does that mean that we
should pay specific attention to it. In other words, does the presence of networks affect more
traditional economic and business theories? The answer to this question is undoubtedly ‘yes’.
The basic rationale behind this answer is that the creation of networks is costly, and that usually
these costs have to beared unequally by various economic agents (by which agent depends on
the type of network).
The reason that there is usually only one big railway operator in a country is that it is
too costly (and too inefficient) to maintain more than one railway network by a multiple of
operators. The same accounts for the provision of water and electricity. Being a provider for
mobile telephones requires considerable investment in aerials. Networks created by clustering
of firms require costs as well, usually in the form of land rents. Compare, e.g., land prices at
the Amsterdam Zuidas with landprices in North-East Groningen (in the northern rural part of
the Netherlands). And finally, consumer networks require initial costs as well. Producers have
to invest by, e.g., below average costs pricing, to lure customers away from competitors and
incumbents. Consumers, on the other hand, have to invest in compatibility. Recall the example
again given above regarding the choice between a VHS and a Betamax video recorder. Most
people will opt for the VHS system, at the moment it is known that the VHS system will be the
prevailing system, even when Betamax might be the superior system. Note that it is oftentimes
simply not known which system will prevail, which makes consumers rather hesitant to invest in
a new technology. Sometimes, consumers have to invest in using a good as well—apart from the
direct costs of a purchase. A direct example of this would be that consumers need additional
equipment (such as hardware) to be able to consume the good (software). An indirect example
would be that consumers oftentimes need to invest time and effort to get used to specific software,
like working with a complex text editor (read: Emacs or VIM), learning a new statistical software
program (R) or learning to type on a QWERTY keyboard. All this does not require additional
monetary costs, but the time one needs to invest might be considerable.
1.3. A typology of networks 3

The consequence of these costs structures—where economic agents have to make an initial
investment in the network before being able to reap the benefits at a later stage—is that the
traditional economic paradigm of perfect competition does not hold anymore.1 Instead, markets
where networks play a significant role2 are typically characterized by the absolute domination
by one or only a few firms, such as the NS, Microsoft, Intel, Facebook, Whatsapp, etcetera. And
oftentimes, the presence of one or only a few firms is even justified from an economic perspective.
Think about it. Creating a rail network requires a lot of investment and should therefore enable
firms to exercise some market power in order to earn back their initial investment. Market
power works as a carrot on a stick to make that initial investment. The challenge is of course to
determine how much market power should be allowed to firms and that is exactly what most of
this syllabus is dealing with.3

1.3 A typology of networks

From an economic perspective it is useful to make a distinction between the following two
network characteristics:

Tangible vs. Non-tangible Pipelines, roads, and rail are typical examples of physical net-
works. You can touch them. Economists refer to these networks as tangible networks.
These networks are rather stable in nature. Usually, roads and rails remain where they
are, as well as pipelines and cables. On the other hand there are non-tangible networks.
These networks are not truly physical, at least not the lines (the edges) between the
connection points (the nodes). Examples of these kinds of networks are ATM machines,
consumer networks, and postal services. Some of these networks are still rather stable
(one does not easily remove a post office), though others—especially the networks between
consumers—are rather unstable. A consumer simply starts to belong to a certain network
when she purchases the particular product, though, as we have seen above, she might still
have to bear some indirect costs.

Owned vs. Unowned A second important characteristic of a network is whether a network is

owned or not and who owns it. For example, the rail infrastructure in the Netherlands
is completely owned by Prorail. Dutch roads are mainly owned by the government just
as the infrastructure for electricity. The network for mobile telephones are owned by a
couple of private providers, just as well as the provision of electricity. ‘Owned’ is usually
also accompanied with paying for the use of a network. If you use the rail network, you
1
We use the term paradigm here, but the notion of departure from perfect competition has already been for a
long time common knowledge for economists. Treatment of different market forms is only seldomly taught to
undergraduate economic and business students.
2
Throughout the syllabus we also use the term network markets for these kinds of markets
3
These are actually rather hot topics in policy making. Recall the recent discussion about the liberalization of
the Dutch electricity network and the pressure on Microsoft to reveal part of the source code for its operation
system to competitors and third party software developers.
4 Chapter 1. Introduction

have to pay a fee. The same accounts for using the mobile telephone network. Using a
road network usually does not involve paying for its directly use.4 On the other side, some
networks are not owned at all, most notably consumer networks, networks between firms
and networks between friends. Because there are still costs to be made (and hopefully
gains as well), these kinds of networks may profit some and harm others when costs (and
gains) are not properly allocated. Economists refer to these kinds of effects as ‘network
effects’ or ‘network externalities’.

Every combination of the two types of characteristics is possible and, as Figure 1.1 shows,
networks may also differ in the degree of tangibleness or ownership. For example, inland
waterways are in principle owned by the government, but individuals do not have to pay for the
use of the network.5 Therefore, we consider inland waterways not completely owned. Continuing
with infrastructure, we can say that it is usually tangible, except for network services like bus
lines. Thus, it really depends on each network market itself, whether it is tangible or owned and
to what degree. The clustering of firms—or economies of agglomeration—is a bit of an outlier
here, because it is not really a good. However, we can say that the network formed by firms
itself is at least definitely not owned or tangible. In Chapter 2 we focus more on characteristics
of networks and ways to represent them graphically.
As already mentioned above, market power is essential to network markets. With tangible
owned networks, there are usually monopolists (such as the NS), while non-tangible and owned
network markets are teemed with oligopolies (e.g., in the mobile telephony sector).6 With
consumer networks it is, a priori, not to say whether oligopolies or monopolies occur. The
tendency at least is towards a smaller amount of firms (look at the market for operating systems,
game computers, on-line auctioning sites, etcetera). Chapter 3 deals with market power, its
causes and consequences, and its specific relation with networks.
Most owned network markets belong to a specific class of sectors, usually referred to
as network sectors. A common denominator of these network sectors is that they were all
completely governmental owned until a few decades ago. Since the beginning of the 1980s the
Dutch government decided to privatize some of these markets. Some of these privatizations are
now regarded to be very successful (mail, telecommunications, radio and television), other are
regarded to be less successful (taxi services and to a lesser extent railway services). Chapter
4 looks closer into these network sectors, the reasons for privatization, and appropriate forms
of deregulations. We specifically focus on the economic rationale for reasons (not) to privatize
and—consequently—(de)regulate, all with an emphasis on the economic consequences of the
presence of networks.
4
This is of course the case for the Netherlands. Other countries have already implemented toll roads, such as
France and lately the UK in London. Besides, roads are indirectly funded by taxes.
5
Of course, the maintenance of inland waterways are maintained by taxes. However, somebody who uses the
river twice as much as somebody else, does not have to pay more for its use. Actually, this accounts for some
other network markets as well, most notably roads and, strangely enough, tap water provision for which you only
have to pay a fixed amount in the Netherlands.
6
Again, this division is not set in stone. Exceptions on this rule happen rather frequently.
1.3. A typology of networks 5

Owned
Telecommunications
Water & elecitricy

Infrastructure
Postal services

Non-tangible Tangible

Inland Waterways
Economies of
agglomeration

Consumer networks

Unowned

Figure 1.1: A typology of network markets

One of the network markets where the network is tangible and widely regarded as a network
where users do not have to pay for its use, is the road network. Here, users who use the road
network extensively, do not have to pay more (through taxes) than users who use the network
only now and then. Because the use of the road network has grown rapidly during the last
decades, it has become congested for at least some parts of the day. Economists argue that
this congestion is mainly caused because using the network is too cheap for some of the users,
causing a—as already mentioned above—network effect. Namely, network users do not take into
account that their actions influence other network users as well. In this case, the presence of an
additional road user means that the road has become a little bit more crowded. With only a
small amount of road users, this effect is negligible. Unfortunately, this effect becomes rather
serious when many users decide to use the road at a specific time (peak hours), which leads to
serious traffic jams. Note again that this effect primarily originates from the fact that people do
not have to pay directly for road use. Chapter 5 discusses this effect, offers possible solutions
from an economic perspective and some policy issues as well.
One of the last types of network markets as given in Figure 1.1 are consumer networks,
being identified by the identical good consumers purchase. Thus, there is a network of Microsoft
Excel users, of milk drinkers, and of DVD owners. Every good has its accompanying consumer
network, only for some goods they are more important than for others. Surely, if the number of
milk drinkers does not exceed a certain amount, milk production will probably stop. However,
6 Chapter 1. Introduction

milk drinkers usually do not have the tendency to learn from each other. The opposite is true
for high-tech goods, specifically ICT goods, where Microsoft Excel users learn from each other
and increase the value of the product. Moreover, the larger the market share of a technological
advanced product, the more additional and complementary services and goods are created.
Thus, the larger the market share of DVD players the smaller the number of prerecorded VHS
video tapes are offered for rent in your local video store7 — indeed they have almost completely
disappeared at the moment. Oppositely, services and goods complementary to the consumption
of milk are rather limited, and its consumer network is therefore as well regarded to be less
important. Shapiro and Varian (1999) deal extensively with consumer networks.
Finally, in Chapter 6 we extend the theory of Shapiro and Varian (1999) to the aggregate
macroeconomic domain and investigate how much the advent of information and communication
technology (ICT) has changed the national economic landscape. In more detail, we look here
at the functioning of ICT as a general purpose technology (GPT), the way ICT affects key
economic variables as inflation, growth and the business cycle, and implications for economic
policy.

1.4 Summary
This chapter shows that networks are everywhere. They occur in a wide range of
forms, applications, and goods. Networks themselves are not free; they usually cost
something—and sometimes even a substantial something albeit only in terms of
time). However, other agents might benefit greatly from the occurrence of networks.
Therefore, when networks are important (for example in transportation), perfect
competition is usually not the correct market form anymore to model markets.
Instead, these markets are typically associated with firms exercising a great deal of
market power. From an economic perspective, there are two important (and related)
characteristics associated with networks: whether they are tangible and whether they
are owned. The next chapter will continue with the various form networks might
have.

7
I know, its an old example. But the principle remains (arguably has become more stronger). Replace, e.g.,
VHS video tapes with mobile phone apps and you catch my drift.
Chapter 2

Networks and Graphs

2.1 Introduction

This section focuses on networks, ways to represent and analyze them, and their economic
implications. The first subsection deals with graph theory, a mathematical tool to analyze
networks graphically. Using graph theory, it is possible to research network optimization; i.e.,
to calculate the most optimal route a bus should make in order to pick up most passengers.
Because of the complexity of this subject, we treat this rather superficially in Section 2.4.1
Finally, in the last section, we look at the economic impact of networks. How is it exactly that
the presence of networks influences standard economic theory.

2.2 Graph theory

In most areas dealing with some kind of network involved, graph theory is used to study the
properties of networks. Informally, a graph is a set of objects called nodes (or vertices) connected
by links called edges which can be undirected or directed (assigned a direction). Typically, a
graph is designed as a set of dots (the nodes) connected by lines (the edges).
Structures that can be represented as graphs are ubiquitous, and many problems of practical
interest can be represented by graphs. The link structure of a website could be represented by a
(directed) graph: the nodes are the web pages available at the website and there’s a directed
edge from page A to page B if and only if A contains a link to B. A road network could be
represented by a graph as well, where the cities (or to be more precise; those places where roads
cross) are the nodes and the roads themselves are the edges. Similarly, airports and airline
connections form networks, just as firms with their corresponding contracts with suppliers and
clients. Finally, individuals often form networks, be it with family, friends, or neighbors. Several
relations between node A and node B are possible. The most important types are treated in the
1
Arguably, for operation management students it is more important to understand the principles behind these
problems than the real technicalities, which is more for the operation research people.

7
8 Chapter 2. Networks and Graphs

subsection below, after which we continue with some general properties of networks in the next
subsection.

2.2.1 Symmetric networks

Symmetric networks are networks where the edges are not directed. Examples of such undirected
arcs include A and B being relatives, neighbors, friends and colleagues of each other. To start
with, consider the very simple symmetric network depicted in diagram 2.1.

A B (2.1)

C D

In diagram 2.1, there are four nodes (A, B, C, D), with four undirected edges: namely
(A, B), (A, C), (A, D), (B, D). As already mentioned, these nodes and edges may represented
anything but for the time being we assume that the nodes represent cities and the edges bus
connections. Thus, city A and B are connected by a bus line, and city C and D are not. We
may also represent this graph in matrix format:

A B C D
A 0 1 1 1
B 1 0 0 1 (2.2)
C 1 0 0 0
D 1 1 0 0

The matrix in equation (2.2) is now filled with zeros and ones, where a one indicates that there
is connection between the corresponding nodes. The convention is that a node does not have a
connection with itself (otherwise it would be stipulated in the graph by a circular arc). Thus,
the matrix has zeros on its diagonal.2 Using ones basically means that each link received a
similar weight. However, we may also assign different weights to the connections in the network
of 2.1. For example, we may attribute to each connection the distance of that connection, or,
sticking to our bus example, the frequency or number of passengers on that link.
The matrix in (2.2) is also known as a first order contiguity matrix. Namely, those nodes
that lie adjacent to each other (in terms of connections) are shown. Suppose, we like to figure
out which destinations we can reach from a node if we have exactly one transfer. Then, we need
the so-called second order contiguity matrix, which can be calculated by multiplying the first
order contiguity matrix with itself (see appendix 2.A for some detail):3
2
The matrix appears to be symmetrical as well. It is left to the reader to figure out why.
3
For third and higher order contiguity matrices the procedure is analogous.
2.2. Graph theory 9

     
0 1 1 1 0 1 1 1 3 1 0 1
     
1 0 0 1
 × 1 0 0 1
 = 1 2 1 1
 

1
 (2.3)
 0 0 0 

1 0 0 0 

0 1 1 1

1 1 0 0 1 1 0 0 1 1 1 2

Now note that the final matrix in (2.3) consists of higher numbers than ones on the diagonal
matrix. These numbers indicate in how many ways one can return in two steps to the node
of departure. Thus, from node B one can go to node A or D and back to node B again.
Often, the numbers on the diagonal are changed to zero, because they are not very informative.
The off-diagonal ones now indicate that there is a connection with one transfer between the
corresponding nodes.

2.2.2 Asymmetric networks

Diagram 2.1 already gave an example of a symmetric network. But consider now that we have
asymmetric relations between the nodes. Meaning that node A has a connection with node B
but node B has no connection with node A. A might for example be the mother of B, which
usually indicates that the reverse is not possible. The connection A → B might as well reflect
a one-way traffic street and so on. Basically, in asymmetric networks people make another
distinction: being hierarchical and non-hierarchical.
We start with non-hierarchical networks. These are networks that are directed (with arrows),
but where detours are possible. This means that node A could be but is not necessarily the
superior of node B. A large example of a non-hierarchical network is the Internet. Here,
webpages point to each other, but not necessarily symmetrically. For example, consider again
diagram 2.1, but now in a directed form, such as in diagram 2.4.

O A_ /B (2.4)

 
C D

Here, nodes A, B, C, and D may be considered as websites and the arcs as hyperlinks. Note
that there is a circuit possible (A → B → D → A). Moreover, node A has a hyperlink to node
C, while the reverse is true as well. This means that A and C have a symmetrical relationship
by change. However, because this is an asymmetric (or directed) network, both arrows have to
be depicted. The asymmetry also pops us in the matrix that reflects the graph, being:
10 Chapter 2. Networks and Graphs

 
0 1 1 0
 
0 0 0 1

1
 (2.5)
 0 0 0

1 0 0 0

Matrix (2.5) has to be read as follows: the rows reflect the nodes of origin and the column the
nodes of destination. Thus an edge from A to B results in the entry on the first row and the
second column filled with a one. The same story now applies for the second order contiguity
matrix. Thus:
     
0 1 1 0 0 1 1 0 1 0 0 1
     
0 0 0 1 0
  0 0 1 1
  0 0 0

1 × =  (2.6)
 0 0 0
 1 0 0 0
 0 1 1 0

1 0 0 0 1 0 0 0 0 1 1 0
Note that this second order contiguity matrix is again asymmetric
The other type of asymmetric networks is a hierarchical network. Hierarchical networks
are characterized by node A being superior to node B. Boss-worker relations are examples of
hierarchical networks. Other examples include pyramid games, tournaments and family trees.
Hierarchical networks typically have the following shape:

A (2.7)

~ 
B C D

~   
E F G H I

Obviously, hierarchical networks have corresponding matrices with only zeros below the diagonal
(now why would that be?). Basically, these three types of networks (symmetric networks,
non-hierarchical asymetric networks and hierarchical networks) represent all types of networks
there are. The next subsection deals with the various properties of networks.

2.2.3 Properties of networks

When dealing with a network, one would like to know certain properties or characteristics of
the network. For example, how large is the network and how well is it connected? To answer
the former, one can use the König or accessibility index of a certain node. Basically this index
gives the minimum number of edges one has to pass when one goes to the node being furthest
away. For example, the König index of node B in diagram 2.1 would be 2, because it take two
edges to go to the furthest node, being C (B → A → C). Similarly, the König index of node B
2.2. Graph theory 11

in diagram 2.4 would be 3, because the edges are now directed and it takes 3 edges to go to the
furthest node, which would again be C (B → D → A → C). Thus, to calculate such an index
for a particular node, we need to calculate the shortest paths to all other nodes and select the
largest one (thus we need to take the maximum of a set of minima). This procedure can quickly
become rather complex, for which we need specifically designed algorithms. Some of these are
discussed in section 2.4.
The König or accessibility index has several applications in the real world. For example, the
König index reflects in how many steps it is possible to reach any particular person—or, say, the
American president—in the world. Surprisingly, the König index from any person in the world
to any other person in the world is remarkably stable and is usually estimated to lie around
7. In transport, applications are abundant. For example, take the number of stations you can
reach with N transfers from station Arnhem, which is:
• 60 stations with no transfer,

• 150 stations with one transfer, and

• 50 stations with two transfers.

So with two transfers one can reach all stations in the Netherlands from Arnhem, which is again
surprisingly low.
Generally, the König index can be used as a measure of centrality in a specific network. The
lower the index, the more central the node can be considered. Consider once again the network
in diagram 2.1. Here, the König index is 2 for nodes B, C, and D, and 1 for node A. Thus,
node A can be considered as the most central node in the network. In transport terms, such a
node can then be considered, e.g., as the most suitable place as a distribution center. On the
other hand, the maximum of all König indices in a network gives the number of links in the
shortest path between the two most remote nodes. This number is also called the diameter of
a networks and gives an indication of the form of a network. The diameter can, e.g., be used
to measure the diffusion of information or technology in a network. If the diameter is large,
diffusion will be rather slow. If the diameter is small, diffusion will be rapid.
The Beta index is a useful measure of connectivity, and is defined as the number of edges
divided by the number of nodes—that is, for a fully connected network. Thus, for the network
in diagram 2.1 the Beta index is 1, and for the network in diagram 2.4 the Beta index is 5/4.
For undirected networks, the Beta index has an important implication: namely, it measures how
many circuits are possible in the network. When the Beta index is smaller than 1, no circuit is
possible, when the Beta index is exactly one, one circuit is possible, and when the Beta index is
larger than one, more circuits are possible. Thus, the network in diagram 2.1 has exactly one
circuit (Beta index is one): namely (A → B → D). For asymmetric graphs, this rule of thumb
does not apply that easily (check this for yourself in diagram 2.4). However, the Beta index
is still a measure of connectivity. Thus, the larger the Beta index, the higher the connectivity
within a network.
12 Chapter 2. Networks and Graphs

Number of
node pairs
connected

Number of arcs removed

Figure 2.1: Network connectivity and network vulnerability

Why would such a Beta index be important? Basically, because when there are circuits
possible in a network, then this makes a network less vulnerable for disturbances on key
connections. Consider a major malfunction on train stations in Sneek and in Utrecht. If
something happens in Sneek (city in the rural part of the Netherlands), other train passengers
in the Netherlands usually do not notice. Totally the opposite is true for the train station in
Utrecht.4 In general, the more circuits are possible, the less vulnerable networks are for sudden
shocks such as accidents, energy fallouts, terrorist attacks, and the like.
To elaborate on this point, consider Figure 2.1. Basically, Figure 2.1 depicts that, usually, there
are a number of key connections and nodes within each network, that, when removed, heavily
decrease the number of connections/circuits within a certain distance. The steeper the line in
Figure 2.1 decreases, the more dependent a network will be upon certain key nodes. For example,
remove node A in diagram 2.1 and the network stops being connected.
Figure 2.2 illustrates this by showing the cable network, used primarily for the Internet in
the United States. Clearly, the network is well connected through the United States, with one
exception. The East-west connection is primarily maintained by two edges, the one going via
Denver, and a south one going via Austin/Houston. This indicates that on these two edges the
Internet is vulnerable to sudden shocks (terrorist attack, earthquakes, and so forth).
To give an application of the use of circuits in transport economics to depict and analyze
network competition, consider the following network:

1
A 2 B (2.8)
3

4
Actually, many circuits are possible on the train network in the Netherlands. For example, the station of
Utrecht can be bypassed via Rotterdam. However, this usually costs passengers an additional hour of travel time.
Moreover, creating such bypasses requires several hours of logistic preparation.
 Page 1 of 1
2.2. Graph theory 13

Figure 2.2: Bandwidth providers in the US Source: http://www.telegeography.com/.

The network in 2.8 shows two nodes (A and B) and three edges (1,2, and 3). Assume that the
nodes are train stations and the connections three similar railway operators (admittedly, a bit
unrealistic), which are identified by their market share, which happens in this case to be equal
to a third for all three railway operators.5 The market shares are identical, because the network
is symmetric and we have assumed similar railway operators. Thus prices, trains and levels of
service are identical. But what happens when we slightly change the network configuration, as
in 2.9?

1
A 2 B (2.9)

3 3

There are now two direct routes from A to B, and one indirect route via C. Everything else
still being equal, this entails that route 3 would be more costly for the railway operator (the
train has to make an additional stop) and that customers have a lower willingness to pay for
route 3 (traveltime is higher). Consequently, the market share of route 3 decreases if the railway
operator does not lower his prices.6 Thus, direct routes are preferred above indirect routes, by
both operator and customer. However, creating and maintaining a connection is costly, which
5 http://www.telegeography.com/ee/free_resources/images/ib2004-fig04.gif
Often, market 26/10/2005
shares in those kind of networks are characterized by the famous Herfindahl index (H), which
12 12 2
can be calculated by the sum of all the squared shares. Which would in this case be H = 3 + 3 + 13 = 19 . It is
left to the reader to figure out what the highest and lowest possible value for H is.
6
Thus, what will happen with the Herfindahl index?
14 Chapter 2. Networks and Graphs

means that trade-offs have to be made about the number of edges and the number of indirect
routes.

A B (2.10)

C D

Specific forms of networks are created when not all nodes are connected, such as in diagram 2.10,
where node C is not connected to the network. Such networks occur rather frequently. High
speed rail is not directly connected to the normal rail network. Worse, rail networks of some
countries are not even connected at all, because of different rail gauges. Networks may not be
connected because there are national borders or because nodes are islands. From an economic
perspective, unconnected nodes may be caused by a lack of supply or a lack of demand (which
causes what exactly is often a difficult question). For example, the new high speed train in the
Netherlands connects Amsterdam and Rotterdam with Brussels. However, Den Haag is not
directly connected, with as arguments that speed and frequency would be too much hampered if
the high speed train would make a detour via Den Haag.
There are other examples of unconnected networks. For example, in most countries in the
world it is the case that households are not always connected to the water and energy supply
network—let alone to the cable and telephone network. The cable and natural gas network do
not even reach everybody in the Netherlands. Often, this is because of specific circumstances,
but usually there are economic arguments for this unconnectiveness. Namely, the marginal7
costs of connecting the last customers are oftentimes very high. Consider households on very
remote locations. The costs of connecting these households to sewage, water, electricity, and
cable are that high that alternatives (solar energy, water pomps) are usually much cheaper.
Finally, there are specific (transport) networks which are connected to each other at certain
transfer points. These networks are the road, rail, aviation, and water (inland shipping, short &
long sea shipping) networks. Networks that are a combination of these various types of networks
are called multimodal networks. The reason that they exist is that, first, every network in itself
has only a limited scope of penetration. For example, inland shipping is sometimes preferred
because of cost considerations, but—as Figure 2.3 shows—the inland waterways network is not
able to reach all potential destinations (e.g., cities), and thus additional road freight transport
has to be used. The second reason is that modalities have their own specific transport advantage.
Water networks are good in transporting bulk goods, while air transport is perfect for fast

7
We come back to this concept of marginal costs later in the syllabus. At his point, marginal may be read as
‘extra’.
2.2. Graph theory 15

transport. If you need to first transport bulk goods, after which they have to distributed to a
large number of cities, it is best to use a multimodal network (e.g., water and road transport).

2.2.4 Intermezzo: the symmetric connections model

The following example comes from Jackson (2008) and is rather insightful in the way network
structures and socioeconomic outcomes relate. In this case, we model social relationships (e.g.,
friendships) in a linear fashion. Relationships offer benefits in terms of favors, information, and
the like, and also involve maintenance costs. The trick here is that individuals also benefit from
indirect relationships. The “friend of your friend” is also of value for you, but to a lesser extent
than a direct relationship. The same is true for the “friend of a friend of a friend”, where we
assume that benefits deteriorate with the distance of the relationship. This deterioration is
represented by a factor δ, which lies between 0 and 1. For instance, in a network in which player
1 is linked to 2, 2 is linked to 3, and 3 is linked to 4, player 1 receives a benefit δ from the direct
connection with player 2, an indirect benefit δ 2 from the indirect connection with player 3, and
an indirect benefit δ 3 from the indirect connection with player 4. The payoffs to this network of
four players with three links is depicted in Figure 2.11. Note that costs are only incurred when
players have a direct relationship.

1 2 3 4 (2.11)
δ+δ 2 +δ 3 −c 2δ+δ 2 −2c 2δ+δ 2 −2c δ+δ 2 +δ 3 −c

Define now the utility of players in such a networks as the sum of all the benefits minus the
costs. Then we can say that the network is efficient when it maximizes the total utility of all
players in the network. There are now three possibilities:

Costs are very low In particular, if c < δ − δ 2 then adding a link between two persons always
increases total welfare. Note that the benefits of a new link accumulates to δ − δ 2 because
individuals get at most δ 2 from each sort of indirect relation.

Costs are intermediate When costs rises, so that c > δ − δ 2 but c does not become too high,
then it turns out that the most efficient network structure is a hub-spoke system. The
intuition behind this lies in the fact that a hub-spoke system has the minimal amount of
links. Namely, one person has n − 1 links, all others have only 1 link.

Costs are too high If costs of a connection are too high then no network is formed.

Of course, the example above is very stylized (with symmetrical nodes, links, distances, costs
and benefits). However, it already shows the connection between economic performance and
network structure and the importance of being central in a network. The latter ensures, namely,
that indirect connections have low König or accessibility indices.
16 Chapter 2. Networks and Graphs

Figure 2.3: Dutch inland waterway network. Large dots represent container terminals, small
dots goods transfer points.
2.3. Real-world networks 17

2.3 Real-world networks

At the beginning of the 1990s, a curious game started at US west coast campuses, called the
Kevin Bacon game. The game focused on connecting the actor Kevin Bacon with every other
actor (be it alive or already dead) by using the smallest number of connections possible. As an
example from Wikipedia with the ‘actor’ Elvis Presley: Elvis Presley was in Change of Habit
(1969) with Edward Asner and Edward Asner was in JFK (1991) with Kevin Bacon. Thus Asner
has a Bacon number of 1 and Elvis Presley has a Bacon number of 2. It turns out that all actors
can be related to Kevin Bacon in 6 or (usually far) less steps, hence the other name of the game:
“the six degrees of Kevin Bacon”.
This is a remarkable feature of this large existing network.8 The connectivity is very high—
and thus the König index is very small – for most (or even all) nodes in this specific network.9
Actually, it turns out that most real-world networks, such as the Internet, social networks (e.g.,
Facebook or Linkedin), sexual contacts and even gene networks possess this characteristic: the
network is extremely densely connected.
This section looks a bit deeper in this phenomenon from a somewhat more technical perspec-
tive. To do so, we consider three types of networks in somewhat more detail.

2.3.1 Random networks

Mathematicians have more or less the same game as moviefans, but do not use Kevin Bacon,
but Paul Erdös instead. Erdös (1913–1996) was a well-known Hungarian mathematician who
wrote hundreds of articles and, more important, had a large number of coauthors. Thus, every
mathematician (and even some economists) now have an Erdös number, which frequently can be
encountered on personal websites and resumes. Moreover, Erdös actually pioneered the study of
networks, inventing graph theory (with the edges and the vertices) as explained in this chapter.
Erdös invented as well (together with Alfred Rényi, who has an Erdös number of 1) the theory
of random networks.
Random networks are networks with no obvious structure. A possible condition could be that
all nodes have to be connected. Usually, random networks are constructed as follows. Number
each node i (i ∈ {1, . . . , N } with N the total number of nodes). Then with probability p there
is an edge or connection between node i and node j.
It then turns out that such networks have interesting characteristics. First of all, the
distributions of connections is asymptotically normal (that is, they follow a bell-shaped curve).
Secondly, the König index of random networks is usually rather small (and it does not really
matter when additional connections are added to the networks). Thus, random networks already
8
Easibly checkable by using the IMDB database on the internet (http://www.imdb.com/.
9
One may of course wonder: why Kevin Bacon? Presumably, because Bacon played in many different movies
and genres and he could therefore be seen as a very central node. Actually, again using the IMDB database,
researchers have found that the average Bacon number is actually rather large for Kevin Bacon. The average
‘Bacon’ number for other actors such as Rod Steiger, Dennis Hopper, Donald Sutherland and Marlon Brando is
much lower (Ball 2004).
18 Chapter 2. Networks and Graphs

approximate real-world networks. However, there is one characteristics that real-world networks
have that random networks do not possess. Namely, in real-world networks there are so-called
hubs: nodes that are central in the networks and have an out of proportion share of all connections
(some individuals or firms are more central then others: compare, e.g., Clint Eastwood with
Madonna in the network of movie actors.). Therefore, researchers constructed in the 1990s
another type of network: the small-world network.

2.3.2 Small-world networks

Consider first the following ordered symmetric network.

• (2.12)

• •

• •

• •

•
First see that the average path length (the König number) is relatively rather large in network

(2.12). Therefore, this network does not resemble a real-world network where connectivity
between the nodes is typically large.
If we now randomly reconnect edges one by one, then the network becomes increasingly less
structured, as can be seen in the following network where we have reconnected randomly six
edges.

• (2.13)

• •

• •

• •

•
This network possesses already those characteristics that real-world networks have: firstly, they

are well connected and, secondly, some nodes have more connections (the hubs) than other
2.3. Real-world networks 19

100 ln P (k) ≈ −2 ln k.

Pr(connected to k nodes)
10−2

10−4

10−6

10−8

100 101 102 103 104

Connected to k nodes

Figure 2.4: Logarithmic depiction of a powerlaw with γ = 2

nodes. It is now not difficult to see that if we continue this process we end up with a random
network. As Watts and Strogatz (1998) prove, rewiring only a relatively small number of edges
in an ordered network already gives us the desired real-world properties: large connectivity (it is
a small world after all) and a high degree of clustering (some nodes are way bigger than other
nodes). Returning to the question: why Kevin Bacon? The answer is now simple: Why not?
Namely, in small-world networks everybody is in the centre of the network. And although some
nodes are better connected than others, relative differences between connectivity are actually
small.

2.3.3 Scale-free networks

So are we there yet? Have we now finally found out how real-world networks look like? In 1999
Albert et al. (1999) used a digital robot to search all websites on the Internet and to register how
many in- and outcoming hyperlinks each website had. What they found was that the number of
incoming links for each website followed (asymptotically) a power function. In mathematical
terms:

P (k) ≈ k −γ , (2.14)

which entails that the fraction P (k) of nodes in the network having k connections to other nodes
is equal to an inverse function. Thus when k increases, this fraction goes down rather quickly.
For networks it seems that γ is usually in the range 2–3. Log-linearize equation (2.14) and we
end up with:

ln P (k) ≈ −γ ln k. (2.15)

Figure 2.4 depicts this logarithmic relation. Hence, the name scale-free network. The relation
20 Chapter 2. Networks and Graphs

between the number of links and the number of nodes is linear (albeit a logarithmic one).
It turns out that scale free networks closely resemble small-world networks. However, scale
free network typically exhibits much larger extreme hubs than small-world networks. Is has
been demonstrated that large real-world networks (such as the internet) can be better described
by scale free networks then by small-world networks. Thus, scale free networks are just as well
characterised by a large degree of connectivity and clustering. The latter even more than for
small-world networks.
You may wonder; why bother? The answer lies in the topology of scale-free networks. When
edges are randomly removed from the network, the network remains functioning. Just as
removing randomly websites from the Internet. The Internet remains functioning. However,
when a few special nodes (the main hubs, or in the case of the Internet, the main servers) are
removed, the whole network collapses. Thus, scale free networks are rather vulnerable to highly
targeted attacks (e.g., by terrorists or hackers).

2.3.4 Summary
This section dealt with the explanation and description of the small-world phe-
nomenon. Why is it that real-world networks are so connected? We first explained
this by introducing the concept of random networks, which describe the high degree
of connectivity actually rather well. However, random networks do no explain the
high degree of clustering some nodes (hubs) exhibit in real life. Therefore, we looked
into the concept of small-world networks and finally in the concept of scale free
networks. The former describes smaller networks (e.g. between friends) rather well,
the latter large networks (such as the internet).

2.4 Network optimisation

The previous sections dealt with identifying and explaining properties of existing real-world
networks. But what if we have to construct a new network, such as a new sewage network
or new railway tracks? Are there any guidelines for doing this as cheap or most efficient as
possible? Are we able to construct optimal routes for buses, postmen, and the like? This section
deals, rather superficially, with those questions and looks a bit further into the issue of network
optimization. The first subsection deals with some elementary forms of network optimization.
The second subsection discusses the evolution of (existing) networks.

2.4.1 Minimum spanning tree: Prim’s algorithm

Creating (and using) networks is costly. Think of the railway and road network, but also of
the route a postman should walk to deliver his mail. If this route is unnecessarily inefficient,
2.4. Network optimisation 21

Figure 2.5: Network with costs between nodes

a
7

8
5 b c
7
9 5
15
d e
6 9
8
11
f g

delivering the mail can take several hours extra, day in day out. The first question is therefore
whether we can connect all nodes with minimum possible costs. Consider the following undirected
network with costs (or time) between the nodes.
Consider the set of nodes (a,b,c,d,e,f and g) as displayed in the unconnected network 2.5. If we
want to create a connected undirected network with minimal construction costs, then a very
simple algorithm looks like:

1. start with an arbitrary node, name this network V ;

2. Find that node that is not in V with minimal distance to network V and connect that
node and the corresponding edge with V . V now includes the new node and edge.

3. Are all nodes connected? If not, return to 2.

This algorithm is a so-called ‘greedy’ algorithm, because it takes the first arc with the lowest
weight (e.g., cost, time or distance) it encounters. Usually, this algorithm is called Prim’s
algorithm or the minimum spanning tree.10 This algorithm works very fast on a computer and
is actually optimal. Note that the number of edges in such a network is always one less then the
number of nodes. Namely, if we have edges arcs than necessary, then it is always possible to
remove one arc from the network and still have a connected network.
If we apply this algorithm on the network in diagram 2.5, then we get for the first step the
network as displayed in 2.6:
which ends up with the complete minimal spanning tree in 2.7.

10
See for an animated webpage of the working of this algorithm: http://students.ceid.upatras.gr/~papagel/
project/prim.htm.
22 Chapter 2. Networks and Graphs

Figure 2.6: Step 1 of Prim’s algorithm starting at d

a
7

8
5 b c
7
9 5
15
d e
6 9
8
11
f g

Figure 2.7: Mimimal spanning tree

a
7

8
5 b c
7
9 5
15
d e
6 9
8
11
f g
2.4. Network optimisation 23

Figure 2.8: Fully connected network

a
7

8
5 b c
7
9 5
15
d e
6 9
8
11
f g

2.4.2 Fully connected networks

The minimal spanning tree networks usually occur, when all nodes have to be connected,
but where direction is not an issue. Public utility companies, like gas, water, and electricity,
usually have these kind of networks. Note that this network is specifically designed to minimize
constructions costs, but what if we want to minimize user costs? Then, it is most likely that
each user has immediate access to all other users, and not via several transfer points as in a
minimal network as in 2.7. Thus, we end up with a fully connected network as in diagram 2.8.

Classic textbook examples of fully connected networks are telephone networks and footpaths on
a campus. However, it is doubtful whether these are truly fully connected networks. Telephone
networks work with certain transfer points and footpaths on a campus only connect all buildings.
Fully connected networks are scarce in the economy because of the already mentioned cost
considerations. Obviously, it is very costly to connect everyone with everyone simultaneously.11

2.4.3 Hierarchical connected networks

If we would want to form a hierarchical connected network, then we chose a central node such
that the total user costs of serving people from this node are minimized (assuming that each
edge has an equal weight). This would leave us then with the following network:

11
Consider the case of the Internet, where every web page should then have a direct hyperlink to every other
web page in the world.
24 Chapter 2. Networks and Graphs

A (2.16)

B C D

E F

These kind of networks are usually associated with location-allocation models. For example, one
can use this kind of network to establish the building site for a shopping center. Airports and
airline operators are also characterized by hierarchical networks (the hub and spoke networks),
where only a limited set of airports have connections to all other smaller airports. Other examples
of hierarchical networks are city structures within countries, where one city is the largest one
with most public functions (France and Mexico have these kind of city structures).

2.4.4 Routing problems

When dealing with directed networks, network formation is slightly different and much more
complex. These kind of problems resembles routing problems: which route is most optimal,
usually in terms of time, for someone to deliver packages, persons, websites, routes in warehouses,
etcetera. Minimal spanning tree algorithms are not adequate anymore, because they do not
take into account that one then has to travel back and forth—on the same link and back to the
origin.

2.4.4.1 Travelling Salesman

The best known problem in this respect is the so-called traveling salesman problem, where a
person—the salesman—should find a minimal route such that all nodes are visited once and
returns to the origin. In our original set of nodes (diag 2.5) this would look like:

Ak (2.17)

Bs /C F
D


E /F
2.4. Network optimisation 25

Although this problem looks quite easy at first sight, with more nodes it becomes rather complex
to find the optimal solution. In fact, this problem is de facto example of a NP-hard problem (Non-
Polynomial-hard); a problem which cannot be solved in polynomial time. With each additional
nodes within the network, the time it will take to solve the problem grows exponentially. Thus,
with NP-hard problems, it is very difficult to find optimal solutions and we have to resolve
to suboptimal but reasonable solutions by using heuristics. Unfortunately, most routing and
scheduling problems in transportation and distribution networks are NP-hard.
Note that even though in all these problems it is already hard to find an optimum. This
becomes even more difficult, however, when one is allowed to add additional nodes, such as
crossroads for example. Then even minimum spanning trees become mathematically very
hard—read: NP-hard—to calculate.

2.4.4.2 Shortest path algorithm

A problem which we fortunately can solve within polynomial time is finding the shortest path
between a node and each other node in the network. This shortest path algorithm is also called
Dijkstra’s algorithm, after the Dutch computer scientist Edsger Dijkstra who conceived it in
1956. It works as follows:

1. Begin with the source node, and call this the current node. Set its value to 0. Set the
value of all other nodes to infinity. Mark all nodes as unvisited.

2. For each unvisited node that is adjacent to the current node, do the following. If the value
of the current node plus the value of the edge is less than the value of the adjacent node,
change the value of the adjacent node to this value. Otherwise leave the value as is

3. Set the current node to visited. If there are still some unvisited nodes, set the unvisited
node with the smallest value as the new current node, and go to step 2. If there are no
unvisited nodes, then we are done.

To illustrate this algorithm consider the network in 2.9. Here we are in step 1. We first
selected the source node (a) and named that the current node. We set its value to zero and all
other nodes at ∞. The other nodes are marked as unvisited (blue).
The next step would be to calculate the value each unvisited node adjacent to the current
node. If that is lower than ∞, which it usually is, we change the value to this value. We then go
the node with the lowest value and mark the current node as visited (red). This would look like
diagram 2.10.
This algorithm now continues until all nodes have been visited. For real life purpose, this
algorithm is relatively fast (not NP-hard) and used (in modified forms) for all kinds of routing
problems (in warehouses, in distribution centers, and between servers on the world wide web).
26 Chapter 2. Networks and Graphs

Figure 2.9: Dijkstra’s algorithm: initalisation

a:0
2

8
4 b:∞ c:∞
7
6 3
2
d:∞ e:∞

Figure 2.10: Dijkstra’s algorithm: step 1

a:0
2

8
4 b:2 c:∞
7
6 3
2
d:4 e:∞
2.4. Network optimisation 27

Adoption

Pilot phase Rapid Large scale Degradation

introdu- use
ction

Time

Figure 2.11: Pattern of innovation adoption

2.4.5 Network evolution

Networks, of course, do not pop up randomly. Usually, there is one or a small amount of nodes
which at a certain point in time form connections between each other and add more nodes to its
network. With most types of networks there is some kind of invention (such as a steam or a
combustion engine) which leads to further innovation and eventually to some kind of adoption
or implementation (in this case trains and cars) of the new technology. In this subsection, we
stay in the realm of transport, but most of this story applies to other areas as well (energy
and microchips lead to computers and mobile phones, etc.). In general, when an innovation is
successful, its life cycle resembles Figure 2.11. First, there is a pilot phase where the innovation
is used by only a selected group of users. There after, the innovation catches on and is rapidly
being used by large groups in society, until most people uses the innovation. After a certain time,
the innovation looses its competitive advantage to newer innovations and gradually degrades.
In Section 6 we continue with innovations, however, at this points we are mainly interested in
the consequences for the size of networks. Table 2.1 displays these life cycles for three major
means of transport in the Netherlands. Train is already in the phase of degradation (after 1930
no major innovation has been done anymore), cars are in the phase of maturation, and aviation
is still in the phase of rapid introduction, although most signs point to the possibility that the
exponential growth in air transport is over. But what about the size of their corresponding
network then? Basically, networks grow the most when the life cycle has passed the rapid
introduction phase, like aviation, and could even expand during the maturation phase (roads).
During the degradation phase however, network decrease as happens nowadays with the rail
network, where some rail connections have already been abandoned (e.g., to IJmuiden).
The discussion above centers on major innovations and their accompanying size of their
networks. But what about individual connections or edges? When are we going to construct
28 Chapter 2. Networks and Graphs

Table 2.1: Examples of breakthrough innovations and their life cycle

Pilot phase Rapid introduction Maturation Degredation

Train 1839–1850 1850–1890 1890–1930 1930–present

Car 1880–1920 1950–1980 1980–present
Aviation 1900–1950 1950–2005

a new connection? Before we offer a decision rule, we note one particular feature networks
exhibit. Namely, new connections tend to reinforce the position of nodes that have already
strong positions. A new canal will reinforce the position of the inland waterway network, and a
new rail track will reinforce the whole rail network, mainly because the larger size of the network
induces more people to use it. The more people uses Whatsapp the merrier.12 That is why
controllers of the whole network have an advantage in the sense that they have a better overview
of the costs and benefits of the networks. To clarify this last point, consider the network in
diagram 2.18.

A 1 B 2 C (2.18)

Here, there is a firm, say bus operator 1, who operates on bus line 1 (A–B). There is another
firm, bus operator 2, who considers to open market 2 (B–C). Note, that this would not only
open market 2, but also the indirect bus line A–C. Thus, bus operator 1 would benefit from
opening bus line (B–C), because of extra passengers on bus line 1. However, these benefits are
not considered by firm 2. It might well be that the benefits of opening bus line 2 are negative for
bus operator 2, but that the benefits on line 1 would compensate for this. However, this would
not be taken into consideration and the (overall) profitable bus line 2 would not be opened.
When bus operators 1 and 2 formed an integrated firm, this would not have happened (and the
bus line would have been opened). So, an integrated firm has welfare advantages in the field of
network expansion.
Now, as a decision rule one should take the following into account. First look at the extra
benefits of opening an additional line. This includes the net receipt on the extra lines plus
the extra net receipts on the existing line. There after one should subtract benefits with the
annualized construction costs of the extra line.

2.5 Economic implications of networks

After reading the previous section, one might be left with the idea that this is al quite nice, but
that consequences for the economy and firm behavior are still rather vague. However, we have
12
Well, in theory that is. Sometimes, negative effects play a role, like congestion.
2.5. Economic implications of networks 29

Euro’s

Costs per passenger

Profits per passenger

Benefits per passenger

L*
Size of a busline network

Figure 2.12: Profitability of a bus operator and the size of the busline network

already seen some economic consequences above. For example, from a firm perspective, it is most
favorable to construct a network with minimum construction costs, while from a consumer’s
perspective fully connected networks are preferred (as long as the consumer does not have to
contribute for the construction of the network). Moreover, as we have seen with the example of
bus lines, firms have a large incentive to own as much of the network as possible, in order to
reap most profits. This in combination with the fact that arcs or connections tend to become
more profitable when the size of the network increases leaves us with the idea that as few firms
as possible want to control networks as big as possible. To clarify this further, see Figure 2.12.
Imagine that a bus operator only operates on one busline, say busline 1, and that only 10
passengers per day find it attractive to travel on this line. Then costs per passenger for the
busline are very high (costs here include all short run costs: gas, personnel, overhead, etc). Now,
the busoperator opens another busline, say 2, which increases demand on busline 1 as well.13
Thus, costs per passenger on busline 1 decreases, meaning that the benefits increase (or losses
decrease) if we assume that the benefits per benefits remain the same.14 . Then, there is a specific
network size, say L∗ , where costs and benefits per passenger are equal. If the network grows
larger than L∗ , the bus operator will make a profit per passenger. Note that in Figure 2.12,
profits remains stable after a certain network size. When the network reaches a certain size, it
does not matter anymore for most consumers whether the network expands or not. Think of a
network of city buses, which is in principle confined to that particular city. But, most often,

13
Demand increases because a wider reach of the total busline network; passengers on busline 1 can now transfer
to busline 2 as well.
14
This is not a very heroic assumption, because whether a bus moves 10 or 20 passengers, the ticketprice remains
the same
30 Chapter 2. Networks and Graphs

Table 2.2: Average distance of any house in the Netherlands to:

Road 5 meters
Bus, tram, or metrostop 350 meters
Railway station 3,700 meters
Express way exit 3,900 meters
Major airport 70,000 meters

we observe that costs decrease in network size. One reason for this is what we have explained
above: a larger network creates new demand.
There is another reason why larger networks are more cost efficient. Large costs are involved
with creating the network itself, the so-called fixed costs. To open a busline network, first buses
have to purchased, drivers trained, and busstops constructed. Think also of the rail network whih
needs railway tracks, the road network which needs roads, and so forth. The more consumers
use such a network, the less the total costs per consumer are, because fixed costs do not vary in
the number of users. Basically, you have to build the network only once, independent of how
many users will use it.
Note that this argument is slightly different than the previous argument. The first argument
deals with the size of the network, the second argument deals in principle with the number of
users of that network. Transport economist typically refer to the first argument as economies to
scale (networks size) and to the second argument as economies of density (number of users of
that network). There is some confusion in the literature, but the bottomline here is that size
matters, whether it is in the network (the size of the edges) or on the nodes.
Examples of economies of density in the nodes of the networks, are the ever increasing size
of (air)ports and container terminals. It seem that there is especially an ungoing concentration
process leading to larger nodes in transport networks. The size of the arcs or connections between
those nodes can be increased by larger frequencies (lightrail) or larger vessels (containerships).
We have seen above as well that there seems to be a trade-off between the number of nodes
and the number of arcs within a network. Most often, networks are not fully connected, as Table
2.2 indicates. So, why aren’t there more connections? Obviously, because this would be too
costly. Increasing the network further does not create the desired economies of scale anymore. In
other words, the construction costs of the extra link does not yield the benefits to be profitable
(from the perspective of the whole network).
For example, if we know that the Netherlands has 6,000,000 dwellings, then the potential
number of interactions would reach the astonishing figure of 36,000,000,000,000.15 There is no
15
That is, if the network is directed and households are allowed to have an connection with themselves. Assume,
we have n households, then the potential number of interactions is of course n2 . If we do no allow for connections
on the diagonal, the number of potential interactions is n(n − 1). Finally, if we assume that the network is
undirected, then the number of potential interactions is 21 n(n − 1).
2.A. Matrix multiplication 31

network with this size that is fully connected. Mail for example works with distribution centers.
Thus, each household is connected to a distribution center and not to every other household.
The same accounts for the other utilities, as gas, water, and electricity.

2.A Matrix multiplication

A matrix X can be seen as a rectangular box filled with numbers xij , where x denotes a number
and i and j are indices that run from 1 to I or J, respectively. Specifically, i stands for the i-th
row of X and j for the j-th column (thus xrow,column ). So, in general we have
 
x11 x12 ··· x1j ··· x1J
 
x21 x22 ··· x2j ··· x2J 
 .. .. .. .. 
 
.. ..
 . . . . . . 
X=
 . (2.19)
 xi1 xi2 ··· xij ··· xiJ 
 . .. .. .. .. .. 
 .. . . . . . 
 
xI1 xI2 ··· xIj ··· xIJ

For matrix multiplication we now use the so-called inner products of vectors (usually denoted by
a 0 ). Say, x = [x1 x2 x3 ] and y = [y1 y2 y3 ], then x0 y = x1 y1 + x2 y2 + x3 y3 . Assume that
we have the following matrix multiplication

XY = Z, (2.20)

then the number on the i-th row and j-th column of Z is calculated by the inner product of the
i-th row of X and the j-th column of Y. For example:
" #" # " # " #
1 3 2 4 1×2+3×1 1×4+3×6 5 22
XY = = = =Z (2.21)
4 4 1 6 4×2+4×1 4×4+4×6 12 40
32 Chapter 2. Networks and Graphs
Chapter 3

Market Power and Networks

3.1 Introduction

As already mentioned in Chapter 1, sectors dominated by large networks rarely resemble the
ideal economic case of perfect competition. On the contrary, usually these sectors are dominated
by one (monopoly) or a few firms (oligopoly). As we shall see, such market structures result in
considerable market power for those firms. Economists typically regard these market structures
as market failures and therefore justify government intervention in these markets.
Market power may arise because it may be beneficial from cost perspectives that only one
firm operates on the network, such as water provision, electricity, rail and gas companies. Such
market structures are usually referred to as natural monopolies, because it seems to make sense
that only one firms develops and operates the network. For instance, two seperate rail networks
in one country does not appear to be very effective. These kinds of sectors are usually termed
network sectors and will be explicitly dealt with in Chapter 4.
Thus, in most markets where large networks exist, some degree of market power exists. This
degree may vary from a pure monopoly (e.g., when a single company supplies all rail services),
to oligopolistic competition (e.g., in the computer operating systems market) and monopolistic
competition (e.g., in road freight transport).
Market power generally gives rise to market failures, and hence also calls for some form of
government intervention, something we treat in Chapter 4 in more detail. A solid economic
analysis of the implied welfare effects from market failures, and of the desirability of different
possible types of government intervention, requires a careful identification of the fundamental
reasons for the existence of market power, and of the implied consequences for firms’ and
consumers’ behaviour in the market considered.
We will begin the chapter by briefly reviewing some general economic aspects surrounding
market power. We will then move on to discussing important sources of market power in network
markets.

33
 34 Chapter 3. Market Power and Networks

3.2
56 Market power: an introduction
Before considering the effects of market power in network markets, this section first briefly
3.2. Market
reviews power:
some generalan introduction
basic economic aspects surrounding the existence of market power.
Before considering the effects of market power in transportation markets, this section first
3.2.1 A spectrum
briefly reviews of market
some general basicstructures
economic aspects surrounding the existence of market
power.
Micro-economics textbooks distinguish a number of different possible market structures. These
market structures differ in particular with respect to the number of suppliers that are active in
the market, and consequently with respect to the degree of market power that they will have.
3.2.1. A spectrum of market structures
As a rule of thumb, the greater the number of suppliers, the smaller their market power will be,
Micro-economics
and textbooks
the more competitive distinguish
the market a number
is said of different
to be. Figure 3.1 showspossible market
the most structures.
important four
These market structures differ in particular with respect
types of market structures that are generally distinguished. to the number of suppliers that are
active
On thein the
far market,
left-hand and consequently
side of Figure 3.1, with
we respect
find thetoworkhorse
the degree of of market theory,
economic power that they
namely
will have.
perfect As a rule
competition, of thumb,
where the greater
a very large numberthe of number
very smallof (‘atomistic’)
suppliers, the smaller
firms offer antheir market
identical
power will
product. be, and
Markets forthe more competitive
agricultural products theare market
typicallyis modeled
said to be.by Figure 3.1 shows
this market form. the most
Perfect
important four
competition can types of market
only arise providedstructures
that therethatare
arenogenerally distinguished.
fixed costs in setting up or closing down a
firm, andOn the production
when far left-handtakes
sideplace
of Figure
under3.1, we find
constant perfect
returns to competition, where a free
scale. Consequently, veryentry
large
number
and of very
exit will small that
guarantee (‘atomistic’) firms offer
in equilibrium, an identical
zero economic product.
profits This
will be market
realized form can
(positive only
profits
arise provided
would attract extra that firms,
there are no fixed
negative costs
profits in setting
would cause upsomeor closing
to close down
down).a firm, and when
An individual
production
firm takes place
has no control over under constant
the market price:returns to the
setting scale. Consequently,
price below the marketfree entry
priceand wouldexitlead
will
guarantee
to that in free
losses (because equilibrium,
entry driveszerothe
economic
price downprofits will be cost,
to average realized
and (positive
hence profitsprofitsdownwould
to
attractand
zero), extra firms,above
a price negative
the profits
marketwould cause some
price would make to theclose down).switch
consumers An individual
immediately firm has
to
nocompetitors.
its control over the Thismarket
is the price: setting the price
most competitive typebelow the market
of market priceThe
structure. would lead tosupply
industry losses
(becauseforfree
function thisentry
marketdrives the price
structure down totoaverage
corresponds cost, marginal
the long-run and hence costprofits down to
(mc)—these arezero),
the
and aof price
costs the lastabove
unit the market price
produced—, whichwould
is (duemake
to thethe consumers
constant returnsswitch immediately
to scale) simply equal to toits
competitors.
the average cost This is the and
function, mostwhich
competitive type
is therefore of horizontal.
also market structure.
The market The equilibrium
industry supply will
function have
therefore for thisthemarket structure
attractive propertycorresponds
that marginalto thebenefits
long-run marginal
equal marginal cost, which
costs: theismarket
(due to
the constant returns to scale) simply equal to the average cost function, and which is therefore
equilibrium is efficient. Note that this market form occurs because of the absence of fixed costs.
also horizontal.
When The marketfixed
there are substantial equilibrium will therefore
costs in setting haveperfect
up a firm, the attractive
competitionproperty that mb
will never be=
mc:prevailing
the the market equilibrium
market form. is efficient.

Perfect competition Monopolistic competition Oligopoly Monopoly

Most competitive Least competitive

Adapted from McCarthy (2001)

Figure 3.1 A spectrum of market structures
Figure 3.1: A spectrum of market structures. Source: McCarthy (2001)
Under monopolistic competition, there are still many firms, but they have some possibilities to
Under monopolistic competition, there are still many firms, but they have some possibilities to
set their own price, typically because they offer a (perhaps slightly) differentiated product.
set their own price, typically because they offer a slightly differentiated product. These firms
These firms thus each face a downward sloping demand function. However, free entry and
exit will still drive economic profits to zero. Moreover, there are still so many firms that there
is no strategic interaction between individual firms.
In an oligopoly, the number of firms has become so small that firms do act
strategically. In making their output and/or price decisions, they explicitly take into account
the decisions made by their individual competitors, and respond optimally to these decisions.
3.2. Market power: an introduction 35

thus each face a downward sloping demand function. However, free entry and exit will still
drive economic profits to zero. Moreover, there are still so many firms that there is no strategic
interaction between individual firms. An example is the market for clothing, which is represented
by a large number of chains of stores. Each chain has its own make of clothing, which no other
store sells. But the clothing in each chain is similar enough to that supplied by other chains
that there is considerable competition.
In an oligopoly, the number of firms has become so small that firms do act strategically.
In making their output and/or price decisions, they explicitly take into account the decisions
made by their individual competitors, and respond optimally to these decisions. Firms may also
incorporate the expected reactions of their competitors to their own decisions, when making
their own decisions. The classical example of an oligopoly is the automobile industry, which has
perhaps a dozen manufacturers in Europe and three in the United States, and their numbers
are still diminishing. Price changes by one manufactures typically invokes reactions by all other
manufactures.
A monopoly, finally, is the least competitive market structure: there is only one firm active
in the market, and by assumption there is no threat of entry. This is a convenient position,
because the firm can freely choose its profit-maximizing price and output level, without having
to worry about the risk of losing a market share to a competitor. Examples of monopolies can
typically be found in those sectors where firms are operating large physical networks, such as
electricity and water pipelines.
The effects of market power upon the behavior of firms can easiest be illustrated by considering
a monopolist who faces a downward-sloping demand function, and who will therefore maximize
profits by equating marginal cost, mc, to the revenues from the production of the last unit
produced known as the marginal revenues, mr.1 Because marginal revenues are not equal
to marginal benefits, mb, the resulting market price and output level will not be efficient,
and—equivalently—will not maximize social surplus.2
Figure 3.2 illustrates this for a monopolist who face the following inverse demand function,
D = 100 − N , and who has constant marginal production costs mc equal to ¤40. N may now
denote the number of passengers carried by an airline on a certain route, or the number of
consumers who bought Windows XP. For a linear inverse demand function D, the corresponding
marginal revenue (mr) function has the same intercept as D, but a slope twice as steep.3 With
a constant mc, the monopolist’s output N m is therefore exactly half the competitive output
N c (which is found where mb = mc); and the corresponding monopolistic price pm exceeds the
competitive price pc = mc whenever the demand function is downward sloping. The loss in
1
As long as the marginal revenues are larger than the marginal costs (mr > mc), total profits can be increased
by expanding output, since the extra revenues from doing so (mr) will exceed the extra cost (mc). The reverse is
true for mr < mc.
2
Equation (3.1) below will tell us that mr = D + N D0 . This means that mb = D > mr = D + N D0 (recall
that D0 < 0 because the inverse demand function is downward sloping).
3
The reader may verify this as follows. An inverse demand function D = d0 + d1 N means that the total
revenues are R = N D = d0 N + d1 N 2 . The derivative of R with respect to N gives mr = d0 + 2d1 N .
36 Chapter 3. Market Power and Networks

100

80
pm
60

p c = mc
40
D=
20 mb
mr
N
10 20 30 40 50 60 70
Nm Nc

Figure 3.2: The monopolistic and competitive equilibria compared

social surplus due to monopolistic rather than competitive supply is given by the shaded triangle,
which can be derived by determining the total benefits lost, and the total costs saved, when
going from N c to N m . We will look at social surplus more intensively in Chapter 5.
We can also characterize the monopolistic equilibrium mathematically. Because the inverse
demand function (which we can also denote by D(N )) gives the price p and hence the average
revenues for a certain output level, total revenues R can be written as R(N ) = N D(N ).
The marginal revenues are then the derivative of R with respect to N . The product rule of
differentiation tells us that the marginal revenue is therefore equal to:4

mr = D(N ) + N D0 (N ) (3.1)

The term D(N ) gives the revenues over the last unit sold; the term N D0 (N ) gives us the lower
revenues over all other units sold, because the price has to be lowered to sell that last unit. We
can rewrite equation (3.1) in a different form by noting that differentiating the inverse demand
function to the number of users equals differentiating the price to the number of users, and by
substituting the expression for the demand elasticity :

∂N p 1 p 1 N
(N ) = = 0 ⇔ = D0 (N ) (3.2)
∂p N D (N ) N (N ) p
Note that  in will generally depend on N , and certainly does not need to be constant a demand
function. When we substitute equation (3.2), as well as D(N ) = p, into (3.1), we find:

4 ∂R(N )
More elaborately, the product rule tells us that the marginal revenue may also be denoted as ∂N
=
∂N
∂N
D(N ) + N ∂D(N
∂N
)
, which simplefies into equation (3.1).
3.2. Market power: an introduction 37

 
1
mr = p 1 + (3.3)

The profit maximization output is then found to where mr = mc applies, or:
 
m 1
p 1+ = mc (3.4)

where pm is the monopolist’s optimal price. Equation (3.4) shows how the monopolistic mark-up
of price over marginal cost increases as demand becomes less elastic (i.e., when  approaches -1
from below5 ). This mark-up is often expressed as the ‘Lerner index’, and is used as an indication
for the amount of monopoly power:

pm − mc 1
=− (3.5)
pm 
The qualitative welfare effects of market power under a Cournot oligopoly or monopolistic
competition will be comparable to those shown in Figure 3.2, although the relative difference
between the competitive prices and outputs, and those under market power, will typically be
smaller under these other market structures than under a purely monopolistic supply. It is the
discrepancy between marginal revenues and marginal benefits that causes suppliers with market
power to offer less than the efficient output level, and to charge a corresponding price above
mc.6 The associated inefficiency reflects the market failure from market power.
As an exercise, the reader may verify the correctness of the profit maximizing equilibrium
shown in Figure 3.2, at N m = 30 and pm = 70, for D = 100 − N and mc = 40. This requires
checking that mr according to equations (3.1) and (3.3) is indeed equalized to mc.

3.2.2 Price discrimination

The above analysis of pricing under market power assumes that the firm can only charge one
single price. But when a firm has market power, and provided it can identify its consumers
and prevent reselling between them, the firm may often find it profitable to engage in price
discrimination. Price discrimination exists when a different price is charged to different consumers
of the same product; or more generally, when the relative profit margin differs over units sold.
Note that not every form of price differentiation therefore implies price discrimination: when
also the marginal costs differ over sub-markets, a lack of price differentiation would in fact reflect
5
A profit maximizing equilibrium cannot occur for an inelastic point elasticity, with −1 <  ≤ 0, since total
revenues would then increase for a reduction in quantity supplied and a corresponding rise in the price charged.
6
A notable exception to this result is an oligopoly under Bertrand competition, which assumes that oligopolists
compete with prices rather than quantities. In the simplest case with identical firms, identical products, and
constant average and marginal production cost, the equilibrium will then entail efficient prices, equal to marginal
cost. The intuition is that the oligopolist with the lower market price can capture the entire market. Each firm
will therefore find it profitable to set its price below that of its competitor(s), until the price becomes equal to
marginal cost for all firms.
38 Chapter 3. Market Power and Networks

price discrimination. Various types of price discrimination can be distinguished, which we will
now briefly review.

3.2.2.1 First-degree price discrimination

An extreme form of price discrimination is ‘perfect’ price discrimination (or ‘first-degree price
discrimination’ or ‘personalized pricing’), in which case the firm charges a different price for every
unit sold, with the unit-specific price set equal to the willingness to pay for that particular unit.
Perhaps counter-intuitively, when a monopolist engages in this extreme form of exploitation
of market power, this would not induce additional efficiency losses compared to the standard
monopolistic outcome shown in Figure 3.2. In contrast, it would take us to the efficient outcome
N c . The intuition is that a monopolist, under perfect price discrimination, would sell every unit
along the horizontal axis of Figure 3.2 against the price implied by the D = mb function. Because
the monopolist would no longer have to reduce the price for other, ‘previous’, non-marginal units
sold when selling a next one, he will find it profitable to serve the market as long as D > mc, so
up to unit N c .
The resulting social surplus will be equal to that under perfect competition, but the distribu-
tion of welfare will have changed significantly: the consumer surplus is entirely ‘skimmed’ by
the monopolist, and has fallen to zero. Therefore, although the outcome may be desirable in
terms of efficiency, it need not be so in terms of distribution.

3.2.2.2 Third-degree pricing

In practice, perfect price discrimination is usually impossible. But a monopolist may still succeed
in charging different prices to different groups of consumers. This is called ‘third-degree price
discrimination’ or ‘group pricing’. From the firm’s perspective, this becomes attractive as soon
as demand elasticities differ over sub-groups of customers. The Lerner index in equation (3.5)
shows that if a firm can discriminate between groups of users, it would use a higher mark-up for
users with a less elastic demand.
This is illustrated in Figure 3.3, where two groups are considered, with different demand
functions. Group A has the same demand function as used in Figure 3.2, and consequently the
same profit maximizing price and quantity is found. Group B’s inverse demand function DB
has a smaller intercept (60) but the same slope (-1) as that of A. The same therefore holds for
mrB . At every N , group B’s demand is therefore more elastic than group A’s demand7 , and
consequently a lower profit maximizing price is found if the same mc applies to both groups.
The reader may again verify the profit maximizing output (10) and price (50) for group B by
calculating mr according to equations (3.1) and (3.3) and comparing the result to mc.
Successful price discrimination requires that customers can be identified as belonging to
certain groups, and re-sales can be prevented. In transportation markets, this is often the case.
7
This underlines that demand elasticity is related but not identical to the slope of the demand function.
3.2. Market power: an introduction 39

100

80
pA
60
pB
40 mc
DA
20
DB A
B mr
mr
N
10 20 30 40 50 60 70
NB NA

Figure 3.3: Third-degree price discrimination by a monopolist

For example, in public transport, customers can for instance easily be separated by age,
which suggests that discounts for elderly—with a typically more elastic demand for public
transport—may be part of a profit maximizing strategy rather than just a nice gesture. Because
transportation products can not be stored, it is often also possible to separate by trip motive:
inelastic demand for commuting peak travelers can easily be exploited by using discount tariffs
outside the peak—although of course not every price differential between peak and off-peak
tickets needs to reflect genuine price discrimination, as also the marginal cost may differ when
capacities are binding in the peak and non-binding outside the peak. And the use of multiple
classes in trains also allows price discrimination. Surely the marginal cost may again differ when
first-class seats offer more convenience. But a derived benefit for the profit maximizing firm is
that it will typically be business travelers and higher income groups that will choose first class,
and both groups can be expected to be relatively price insensitive. Ticket prices can thus be
differentiated more strongly between classes than what would be appropriate on the basis of
cost considerations alone.

3.2.2.3 Second-degree price discrimination

Having discussed examples of ‘first-degree’ and ‘third-degree’ price discrimination, ‘second-degree’

price discrimination or ‘versioning’ of course still remains to be considered. This is the situation
where individual choose between various versions or quantities of a product. The rationale is that
each consumer pays more or less the price for the product he is willing to pay for that product,
so that firms are more able to skim the market. For example, it is customary in public transport
that season’s tickets are on offer, which allow the purchase of trip tickets at a discounted price.
Consumers are of course free to decide whether or not to buy a season ticket, but when they
40 Chapter 3. Market Power and Networks

do so, they voluntarily make themselves subject to a two-part tariff, where consumers pay a
lump-sum fare and next a price per unit purchased. Supermarkets (bonus cards) and airlines
(frequent flier cards) offer client cards, where discounts are possible, when customers purchase or
fly more.
In general second-degree price discrimination frequently occurs in markets with consumer
networks. For example, software is usually offered in various versions (demo’s, professional
editions, etcetera), so that consumers may sort themselves out in users who would just like
to ‘taste’ the product and professional users. In essence, this is also the main difference with
third-degree pricing. In third-degree pricing groups are easily identified by the firm where in
second-degree pricing consumers identify themselves by the versions of products they buy.

3.2.3 Monopoly rents

Figure 3.2 showed the profit maximizing output decision for a monopolist, and the resulting loss
in social surplus compared to competitive pricing, but did not yet show us how large this profit
will be. To determine this, we have to add another cost function, namely the average production
cost ac. When there are substantial fixed cost in setting up a firm—which is usually the case for
a monopoly in a sector dominated by a large (physical) network—this ac function will typically
have a shape such as the one shown in Figure 3.4. For this function, we have assumed that
mc is still constant and equal to ¤40, as in Figure 3.2, but that in addition fixed cost equal to
f c = 600 applies. The average cost function shown thus becomes ac = f c/N + mc; or, for the
numerical example, ac = 600/N + 40.
At N m , the average cost acm amounts to 600/30 + 40 = 60, and the total profit can be
determined as N m (pm − acm ) = 30(70 − 60) = 300. This total profit is shown as the shaded
area. Note that it could also have been calculated as total revenues R (= 30 − 70 = 2100) minus
total cost C (= f c + N m mc = 600 + 30 × 40 = 1800).
The absence of competition thus allows the monopolist to earn excess profits. These are often
referred to as ‘monopoly rents’. Unfortunately for the monopolist, there are various reasons
why at least some of these excess profits may vanish over time. Because these reasons are often
relevant also in network markets (and especially in network sectors), we will briefly address
them. We stick to the example of a monopoly, but similar arguments would again often hold for
oligopolistic firms.

3.2.3.1 Rent seeking

One reason why a monopoly may exist is when government policy simply restricts or forbids
entrance of competitors into the market (see also Section 3.3.2 below). In such cases, the
monopolist of course has a clear interest in making sure that the policy is not changed, as this
would endanger the profits made. Activities that reduce the probability of a policy change
may include all sorts of lobbying activities, such as hiring professional lobbyists, diners with
3.2. Market power: an introduction 41

100

80
pm
ac m
ac
40 mc
D
20
mr
N
10 20 30 40 50 60 70
Nm

Figure 3.4: Monopoly profits or rents

government representatives, information gathering, etc. In the extreme, the monopolist would
find it worthwhile to invest an amount of money up to the size of the monopoly rent in activities
that seek to guarantee maintenance of the monopoly position. Such behavior is called ‘rent
seeking’. Rent seeking typically does not affect the marginal conditions on the market in which
the monopolist is active, and hence will not affect the equilibrium. But it will reduce the overall
size of overall monopoly profits. Although some people may personally benefit from rent seeking
behavior, it is generally considered to be a non-productive waste of resources, which means that
it creates another economic inefficiency due to monopoly power.

3.2.3.2 Rent sharing

It is often widely known when a firm holds a monopoly position, and consequently has the
ability of earning monopoly rents. Other economic agents may use this knowledge when dealing
with the monopolist, in the hope of extracting some of these rents to their own benefit. A good
example could be labor unions. For competitive firms, labor unions will typically trade off the
direct benefits of higher wages for their members against the risk of the firm’s going bankrupt,
in which case the resulting loss of jobs would of course be considered as an undesirable outcome
also for the union’s members. For monopolists, a union may realize that there is more room for
negotiating higher wages and more favorable fringe benefits, as the monopoly rent would make
the firm less vulnerable to bankruptcy. It is therefore not uncommon to find relatively highly
organized and strong unions in monopolistic markets, such as network sectors. Any successful
attempt to negotiating better contracts than in competitive markets would give an example of
‘rent sharing’.
Rent sharing in the first place reduces the eventual size of the monopoly rent. Secondly,
when rent sharing raises the marginal cost—as in the example of higher wages—-it would also
42 Chapter 3. Market Power and Networks

lead to an (even) higher price and lower quantity than in the monopolistic equilibrium without
rent sharing (consider the effects of upward shifts of ac and mc in Figure 3.4). Rent-sharing
may then aggravate the efficiency losses from monopoly power.

3.2.3.3 X-inefficiency

Because a monopolist does not have to worry about losing market shares to competitors, the
incentive to operate efficiently is typically smaller than for competitive firms. The resulting ‘X-
inefficiency’ may for instance result from a lower labor productivity throughout the organization,
a less ambitious management, a smaller incentive to seek the best suppliers for inputs used,
greater budgets for prestigious but unproductive activities (‘gold-plating’), and a smaller incentive
to implement product and process innovations. As with rent-sharing, X-inefficiency will reduce
the size of the monopoly rent, and may also raise marginal cost and hence lead to a higher price
and a smaller output.

3.2.3.4 Capitalization

A final reason why monopoly rents may dissipate over time might be capitalization. This refers
to the situation where the right to operate a monopoly is to be acquired, for example through
winning a concession in an auction, or when a certificate is to be purchased from an exiting
monopolist (e.g., for local monopolies). Again, the monopolist would find it worthwhile to bid an
amount up to the size of the expected monopoly rent in order to acquire the monopoly position.
Some of the effects are similar to those of rent seeking: the excess profits will be reduced or
even disappear, but the price and output level are less likely to be affected (assuming that the
right to operate the monopoly simply goes to the highest bidder, and the chance of winning
the right is not affected by the monopolist’s intended prices and quantities). A difference with
rent seeking is that capitalization involves a transfer rather than a waste of financial resources,
which makes the efficiency losses smaller. Moreover, if an auction for the monopoly position is
sufficiently competitive, it may lead to a decrease of rent-sharing and X-inefficiency, because
excess profits have disappeared already before production has commenced.
We can thus distinguish four reasons why monopoly rents may dissipate over time. It is
important to realize that when this occurs, the possibilities for policies that aim to make the
monopolist move closer to the competitive equilibrium are reduced. In the extreme event where
all monopoly rents have evaporated, there will be only one point where the monopolist can
operate without making losses, and that is the monopolistic equilibrium. Only in the longer
run, when for instance rent sharing and X-inefficiencies can be reduced, would it then become
possible to avoid losses for the firm when moving closer to the competitive outcome.
3.3. Sources of market power 43

3.2.4 Summary

Various types of market structure can be distinguished. The usual ones are—in order of decreasing
competitiveness—perfect competition, monopolistic competition, oligopoly and monopoly. When
a firm has market power, it will often seek to maximize profits by equating marginal revenue to
marginal cost. Because efficiency requires marginal benefit to be equal to marginal cost, the
resulting outcome will typically not maximize social surplus.
A firm with market power may find it profitable to engage in price discrimination, especially
when the firm can identify its consumers, prevent re-sales, and when demand elasticities vary
over sub-markets. For firms in network markets, these conditions are often satisfied, and price
discrimination is indeed often observed. Although price discrimination reflects the exploitation
of market power, it certainly need not necessarily aggravate efficiency losses from market power.
In contrast, perfect price discrimination may even lead to the competitive outcome in terms of
quantity supplied, although the distribution of welfare will be different than under competitive
supply.
Market power typically gives a firm the possibility of making excess profits, but these rents
may often dissipate—at least to some extent—in the longer run, due to rent seeking, rent sharing,
X-inefficiency or capitalization. Especially the first three of these may aggravate the efficiency
losses due to market power.

3.3 Sources of market power

The continuing existence of excess profits under market power of course raises the question why
there would not be any competitors that enter the market in response to these profits. What
would be the relevant barriers?
Three categories of such barriers are often distinguished, namely economies of scale, gov-
ernment policy, and control of critical inputs. Each of these are relevant to the explanation of
market power in all kinds of network sectors, such as electricity, water, and telecommunication,
and in markets dominated by consumer networks, such as ICT. To understand the behavior
of firms with market power better, and to make a more solid analysis of actual and potential
government responses to such market power, it is important to know the underlying source(s)
of the firm’s market power. This section therefore discusses these in some greater depth, with
the emphasis on the first two categories. The control of critical inputs, such as in ‘natural’
monopolies is discussed further in Chapter 4.

3.3.1 Market power arising through economies of scale and scope

A fundamental economic reason for the spontaneous emergence of market power in a free market
could be that a larger firm can produce a certain level of output more efficiently than what
could be realized by many small firms. In such cases, one speaks of economies of scale (when
44 Chapter 3. Market Power and Networks

the average costs for a single product fall in the amount of output produced by a single firm) or
economies of scope (when the average costs for related products fall when they are produced
jointly by a single firm). Entrance of the market then becomes difficult, as its success depends
on whether the entrant can manage to start off at a size that allows a sufficient exploitation
of economies of scale and scope, and that hence enables the firm to offer a competitive service,
without making losses, as soon as possible.
We will consider a number of sources of economies of scale and scope in transport below,
beginning with those that are comparable to what could be observed also in other types of
markets, and next moving on to those that are more specific to transport.

3.3.1.1 Sources of economies of scale and scope in transport

Probably the most important reason why economies of scale may arise concerns the existence
of substantial fixed costs. Figure 3.4 already illustrated how the combination of fixed costs
and constant marginal costs causes the average costs to fall as output expands. The discussion
below offers reasons for the existence of economies of scale and scope in transport but are easily
extended to other kinds of network markets.
In transportation markets, such fixed costs may arise from the necessity of having an
infrastructure, before operations can commence. And infrastructure will typically have a
minimum technical scale before it can be used: before any train could be run between Amsterdam
and Haarlem, the entire track of course has to be present. Likewise, no take-offs can occur
from an airport unless there is at least one runway (and one terminal). In the long run, and
with a sufficiently large demand for the use of the infrastructure so that the minimum possible
scale of infrastructure is exceeded, infrastructure capacity may be regarded as a variable input.
Consequently, the need for infrastructure capacity need not always lead to significant economies
of scale for an industry’s long-run average cost function. Nevertheless, when the minimum
possible scale of infrastructure is relatively large compared to demand, infrastructure costs may
be a reason why economies of scale would occur at the industry level, even in the long run.
And in the short run, the existence of given infrastructure costs would of course mean that the
average costs for an operator who owns the infrastructure will fall as output increases, for the
simple reason that these fixed infrastructure costs can be divided over a larger output.
A second reason why economies of scale may occur in transport is that a larger demand
allows the use of, for instance, longer trains, or bigger aircraft. Economies of scale would then
occur when the cost per passenger (or per ton of freight) go down if more of them are transported
simultaneously. This is often the case. For example, the size of the crew will typically not grow
in proportion with the number of passengers: for a bigger train or airplane, still only one driver
or two pilots are needed. Larger trains and airplanes may also be more efficient in terms of fuel
use, maintenance, and capital costs (per passenger or ton).
Thirdly, and not unimportantly, a greater output often means that the same train or airplane
can be used more intensively over the day. This reduces the per-unit-of-output fixed capital
3.3. Sources of market power 45

costs from owning (or leasing) of the train or aircraft. It is not without reason that a famous
saying in the airline industry tells the airline to ’keep the aluminum in the air’.
Fourthly, a larger transport company of course allows specialization of its employees. Financial
management, service scheduling, maintenance, ticket handling, flying or driving, and other
activities are—comparable to what is often the case in other industries—typically performed
more efficiently by specialist employees.
Fifthly, there may be important economies of scope in transport. One example is the use of
the same airplane for transporting passengers, freight, and (express) mail. The additional cost
for reserving space for freight or mail in the aircraft’s freight compartment is often lower than
the cost that would be made for a separate flight, and need not conflict with the room available
for passengers. Another example arises when demand fluctuates over time. An example would
be a convertible aircraft, that can be used for tourist charter traffic during summer holidays, and
for freight transport during the rest of the year. An aircraft can be used for summer destinations
in summertime and winter sport destinations during winter time. Similarly, at the level of
infrastructure, it is not uncommon that rail tracks are used for passengers during daytimes, and
for freight trains at night. Yet another example of economies of scope would be the alternate use
of the same train or aircraft (and its crew) for two different destinations, from the same origin,
when demand on either origin-destination pair is insufficient to justify continuous services. The
train or aircraft could then first make a return trip between cities A and B, then between A and
C, then again A and B, and so forth.
Each of the above examples may contribute to the existence of scale and scope economies in
transport. Most of these considerations, however, are similar to arguments that might also be
relevant for any other economic sector. This is not the case for a different type of economies of
scale and scope in network markets, to which we will turn below. These stem from the network
structure of those markets themselves. The resulting network economies may be among the most
important sources of market power, and hence deserve some extra attention. The discussion
focuses in this case on transportation markets, however, may again have implications also for
network markets other than transport, such as telecom companies, the Internet, cable television,
electricity, etc.

3.3.1.2 A simple form of network economies

A simple example of network economies can be taken from the Dutch railway network. The
example involves two long distance connections, namely between Groningen and Amsterdam, and
Enschede and Rotterdam, that are scheduled to stop simultaneously in the city of Amersfoort.
Figure 3.5 shows the relevant network.
By exploiting network economies, the Dutch railways are capable of offering 4 connec-
tions (Groningen– Amsterdam, Groningen–Rotterdam, Enschede–Amsterdam, and Enschede–
Rotterdam) at the cost of 2 connections. The network structure of the industry thus allows the
supplier to realize a particular type of economies of scope: four long distance origin-destination
A simple form of network economies
A simple example of network economies can be taken from the Dutch railway network. The
example involves two long distance connections, namely between Groningen and Amsterdam,
and Enschede and Rotterdam, that are scheduled to stop simultaneously in the city of
46
Amersfoort. Figure 3.6 shows the relevant network. Chapter 3. Market Power and Networks

Groningen
Amsterdam

Amersfoort
Enschede
Rotterdam

Figure 3.6 A simple example of network economies

Figure 3.5: A simple example of network economies

pairs can be offered, while only two have to be actually produced. And the supply of services
between each of these cities and the city of Amersfoort are provided alongside the long distance
connections, without any further cost.
There is one important side-condition that must be fulfilled to make this type of network
economies work, and that is that the cost per passenger-kilometer fall as larger trains are used.
Otherwise, one might as well run four separate trains between the four city pairs, and separate
trains to connect Amersfoort. As we saw above, however, this side condition is typically satisfied.

3.3.1.3 A second example of network economies: ‘hub-and-spoke’ networks

A second example of network economies can be taken from the aviation industry. Suppose that
we would have a situation where an airline wants to provide services between four cities, labelled
A — D, implying 4 × 3/2 = 6 city pairs. One possibility of doing this would be to offer direct
connections between each city pair. This would mean that the airline has to offer 6 different
return flights. The resulting fully connected network is shown on the left-hand side of Figure 3.6.
Alternatively, the firm might offer services only between its home city, say city A, and each
of the other cities. Travelers that do not have city A as an origin or destination, could then still
make their trip, but would have to make a stop-over in the ‘hub-city’ A. Three different return
flights are then sufficient to connect all city pairs in the resulting hub-and-spoke network. Again,
if the average costs per passenger mile fall with the size of the aircraft, as is typically the case in
reality, the exploitation of such economies of scope (offering 6 connections by producing only 3)
may give rise to substantial cost savings for the airline. Most of the greater carriers therefore
nowadays operate hub-and-spoke networks.
The potential advantages of using a hub-and-spoke network are easily seen to increase in the
size of the network. With n cities, a fully connected network would require n(n − 1)/2 return
flights (starting from each of the n cities, n − 1 destinations are possible, but each return flight
of course connects 2 cities). A hub-and-spoke network only requires n − 1 return flights. Figure
3.7 shows how the implied difference in the required number of return flights increases rapidly,
and at an increasing rate, as the number of cities in the network, n, increases. Economies of
scope through the use of hub-and-spoke operations may therefore remain relevant also for bigger
airlines.
different return flights are then sufficient to connect all city pairs in the resulting hub-and-
spoke network. Again, if the average costs per passenger mile fall with the size of the aircraft,
as is typically the case in reality, the exploitation of such economies of scope (offering 6
connections by producing only 3) may give rise to substantial cost savings for the airline.
MostSources
3.3. of the greater carriers
of market powertherefore nowadays operate hub-and-spoke networks. 47

A B A B

C D C D

Figure
Figure 3.6: 3.7 A fully
A fully connected(left)
connected (left) versus
versus aa hub-and-spoke
hub-and-spoke(right) network
(right) network

The potential advantages of using a hub-and-spoke network are easily seen to increase in the
# ret urn f lights

size of the network. With n cities, a fully connected

300
F u l l y network
c o n n e c t e d would require n!(n–1)/2 return
250

200

150

100

50 Hub -and-s poke

n
5 10 15 20 25

Figure 3.7: The number of required return flights for a fully connected versus a hub-and-spoke
network

3.3.1.4 Economies of scale through frequency effects

A final important potential source of economies of scale that is relevant for scheduled transport
services arises through the increased frequency of services that often results from a larger number
of passengers. A higher frequency need not only reduce the capital cost per passenger as trains
or planes are used more intensively over the day, but may in itself also attract more passengers
because the service becomes more attractive. In this case, the economies of scale are not directly
reflected in the operator’s average cost function, but nevertheless may indirectly contribute to
its profitability, because a result will be that the passengers’ willingness to pay for the service
increases. We can illustrate this using a simple example, where a tram service connects two
points, and potential customers do not know the exact time table, but do know the frequency f ,
expressed in the number of services per hour.
Because the exact departure times are unknown, the potential customers will take into
account the expected waiting time at the tram stop in calculating the full generalized price of
the service. This expected waiting time is equal to 1/(2f ) hours: in the worst case, a tram has
just departed and the traveller will have to wait for 1/f hours for the next one; in the best
case, he can just catch the tram and has a zero waiting time. All intermediate waiting times
48 Chapter 3. Market Power and Networks

occur with equal probabilities, so that the expected (average) waiting time equals the average
of the two extremes, which is 1/(2f ) hours. If the waiting time is valued at votw , this simply
means that the implied average waiting cost amounts to votw /(2f ). When the operator serves
more passengers, and in doing so increases the frequency, it can capture the resulting frequency
benefits – the decrease in votw /(2f ) – by raising the ticket price correspondingly. The economies
of scale, that originally benefited the passengers, then become genuine economies of scale for the
firm.
Of course, the existence of such benefits does not require the schedule to be truly unknown.
With a known schedule, comparable benefits will be realized because the average passenger can
travel closer to his or her most desired departure or arrival time.
We finally note that frequency benefits and hub-and-spoke benefits may reinforce each other.
If, in the example of Figure 3.6, the airline switches from a fully connected to a hub-and-spoke
network and keeps the same total number of passengers, an alternative to using bigger aircraft
would be to increase the frequency on the remaining three flights. This may contribute to the
firm’s profitability for the reasons just explained. The firm can thus choose between cashing in
on the advantages of hub-and-spoke operations via higher frequencies, or via bigger aircraft. A
profit maximizing firm would in the long run seek the optimal combination of these two.

3.3.1.5 Implications

We have seen many reasons why economies of scale and scope may exist in markets dominated by
networks. Some of these do not strictly require supply from one single firm. Also with competing
airlines, a consumer could make a ‘hub-and-spoke’-type of trip, by using a different airline for
both segments. But connecting flights will then typically not be scheduled optimally from the
consumer’s perspective (hub-and-spoke carriers generally schedule arrivals at and departures
from the hub airport in a number of ‘banks’ over the day), ticket purchasing may become more
complicated, the total ticket price may be higher because the ticket prices for the two direct
flights will exceed that for the entire indirect flight, and luggage may get lost more easily. And
also with competing tram operators, the customer would value a higher frequency, independent
of whether these trams are run by a single or by multiple firms. But firms in competition need
not set departure times equally spaced, so that the benefit in terms of an increased frequency
may be smaller that for one operator when the total number of services increases. Moreover,
competing firms would have a lower incentive to optimize frequency from this perspective, as a
part of the implied benefits would be reaped by their competitors rather than by themselves. In
general, therefore, also these economies of scale and scope will often be exploited most effectively
when only one or a few firms are active in the market.
This creates an interesting dilemma for governments. On the one hand, there are advantages
in exploiting economies of scale. On the other hand, this typically requires the presence of only
one or a few operators, who will consequently have a significant degree of market power.
3.3. Sources of market power 49

We will see in the next chapter which policy lessons for dealing with such dilemmas can be
identified from the economic perspective.

3.3.2 Market power arising through government policies

Given the inefficiencies associated with market power, it may seem odd that market power can
sometimes result from government policies. Nevertheless, this frequently occurs, and there may
be economic arguments—sometimes valid, sometimes not—for governments to restrict access to
markets, and to accept the resulting market power for existing firms as an undesired by-product.
This section discusses a number of considerations that may underlie such policies.

3.3.2.1 Market instability

An important motivation for restricting entry in transportation (and other) markets has tradi-
tionally been the fear for unstable markets. Market instability can be defined in many ways. For
our present purpose, it suffices to declare a market unstable when there exists no equilibrium in
which it is simultaneously true that there is no incentive for outside firms to enter the market,
and no incentive for existing firms to leave the market. In short, there exists no stable equilibrium
in terms of the number of firms active in the market.8
We can illustrate this by making a small change to our model in the previous section, namely
by replacing the constant mc by a rising mc. As shown in Figure 3.8, the average cost function
will now be ‘U-shaped’, because the mc function by definition cuts ac in its minimum, acmin .
The output level at which this occurs, N e , is sometimes referred to as the firm’s ‘efficient scale’.
This is somewhat confusing, as N e relates to cost efficiency only, and does not consider overall
economic efficiency – which would ask for the consideration of the benefits from consumption,
too. We will therefore call N e the firm’s cost-efficient scale.
We have chosen the mc such that the firm in monopoly would choose the same output level
as before, namely N m = 30, and hence sets the same price, pm = 70. The profit, indicated
with the shaded area, will now be somewhat higher than in the previous example, because
the ac function is now different. Note that profit maximization generally does not induce the
monopolist to produce at the cost-efficient scale N e . This underlines once more that not only
cost considerations, but also demand characteristics affect the monopolist’s output decision.
What will happen if entry in this market is free, and an identical firm might enter? The
candidate entrant would certainly be tempted to do so. Knowing that his acmin is below pm = 70,
a clear potential for profits appears to be on offer, for instance by entering the market, choosing
a production level N e , and setting the price slightly below the incumbent’s price. The existing,
‘incumbent’ firm will find it very difficult to prevent such entry by means of pricing. Entrance
8
quite different definition of market instability is for instance employed in the so-called ’cob-web model’,
traditionally illustrated in the context of ’pig cycles’. The cob-web model shows how a standard competitive
market may be unstable, and fail to achieve the equilibrium, when supply is relatively elastic, demand is relatively
inelastic, and the quantity supplied in a given period is determined solely on the basis of the market price in the
previous period, without any forward-looking behaviour on behalf of the suppliers.
50 Chapter 3. Market Power and Networks

€
mc ac
100

80
pm
ac min ac 2
40
D
20
mr
N
10 20 30 40 50 60 70
Nm Ne N#

Figure 3.8: An unstable market when entry is unrestricted

would always appear profitable for the entrant, unless the incumbent would set a price equal to
acmin also when being a monopolist. But at the corresponding demand N # , this would generate
a loss, because ac at N # exceeds p = acmin . Even if this loss can be avoided by restricting
output and somehow rationing demand to N e , a price p = acmin would of course by definition
take away all profits, and it is unlikely that a monopolist would choose this price even if no
potential competitor is ‘in sight’.
However, once the entrant has commenced operations, it is impossible in our example to
avoid losses for at least one of the two firms, independent of the type of competition that will
arise (e.g., Bertrand competition, Cournot competition, or even collusion). This is illustrated by
the second, dashed ac function, called ac2 . It gives the minimum possible average production
cost with two firms in the market. It accounts for the double fixed costs that will exist with two
firms in the market, and furthermore uses the minimum possible marginal production cost when
two firms are present (which is obtained by assuming that for every N , production is divided
equally over both firms). Because it lies entirely above the inverse demand function, which gives
average revenues at each N , there is no equilibrium with two firms possible without losses for at
least one of them.
After entrance, therefore, one of the two firms will sooner or later have to terminate operations
because of losses, one firm might survive, and the cycle may start all over again. The market is
unstable, because excess profits are generated with one supplier, inducing entrance; but losses
are unavoidable with two suppliers, inducing exiting.
To avoid such wasteful cycles, the government may restrict entrance into the market. Such
fears for market instability were among the key motives for strict entrance regulation in aviation
markets that was in place for most Western countries over a significant part of the previous
century. Airlines are often believed to have a number of characteristics that would make aviation
3.3. Sources of market power 51

markets prone to instability. Among these characteristics, the most important ones are that
an individual supplier’s cost-efficient scale will often be relatively large compared to market
demand, and that fixed costs per carrier on a market may often be relatively large compared to
marginal costs.

3.3.2.2 Wasteful competition, price wars and destructive competition

Even if the market would be stable with two competitors, the government may still forbid entry
when it fears wasteful competition. In the context of public transport and aviation markets,
wasteful competition is usually defined as the situation where multiple firms serve the same
market, while the capacity of one of these firms would already be adequate to serve market
demand. Any duplicate services would then be socially wasteful, at least from the perspective
of the costs involved. Wasteful competition becomes more likely when each entrant incurs
substantial fixed costs, which can be avoided when only one firm is present, and when the
economies of scale for one firm are so strong that the market will not be served at lower average
costs when multiple firms are present. Although one might argue that if these scale economies
are so important, excessive entrance by small firms becomes less likely anyway, governments
may nevertheless see this as a motivation for limiting access by law, and may prefer to achieve a
more efficient output decision by the incumbent firm through direct regulation instead of by
competition.
Moreover, under these same conditions where wasteful competition may occur, price wars
could easily arise as an undesired additional by-product of wasteful competition. When firms face
substantial fixed costs, fierce competition may lead them to set prices at a level that, although
covering the variable cost, are not sufficiently high to also cover the fixed cost.
To see why, let us return to the example of Figure 3.4, and imagine two identical firms that
face a given fixed cost (f c) in addition to the constant marginal cost of ¤40. Under Bertrand
price competition, where the firm with the lower price captures the entire market, both firms
would find it profitable to set their own price below the competitor’s price, as long as their own
price exceeds ¤40. This occurs because the following pay-offs are relevant:

1. With a price above the competitor’s price, no output is sold and the firm’s profit (in fact a
loss) will amount to −f c;

2. With a price equal to the competitor’s price, and assuming that market demand will then
be divided equally over both firms, the firm’s profit will be 12 N (p − mc) − f c;

3. With a price below the competitor’s price, the firm serves the entire market, and the profit
will be N (p − mc) − f c.

As long as the firm’s price p exceeds the marginal cost mc, the third of these options yields the
highest profit, or at least the smallest loss, and will therefore be the preferred option.
52 Chapter 3. Market Power and Networks

Under Bertrand competition, the price war will thus end with both firms charging the same
price p = mc, and both firms will be making losses as the fixed costs are not covered (note
that under Bertrand competition, this market is therefore unstable). Under more extreme
conditions, a firm may even temporarily lower its price below its own marginal cost, with the
aim of imposing such large losses on the competitor that he will eventually have to leave the
market. Such practice is sometimes referred to as destructive competition, because the goal is to
destroy the competitor, rather than just maximizing instantaneous profits (as under Bertrand
competition). A related motivation for this strategy may be to build up the reputation of being
a tough competitor, which may prevent later entrance from other firms into that same market,
or in other markets in which the firm is active. Governments may find such price wars and
destructive competition not in the long-run interests of the consumers (nor the firms), and may
seek to avoid them simply by restricting entry.
Fears for wasteful competition, price wars and destructive competition were also important
reasons for the regulation of most network sectors. We emphasize again that such fears need not
always be grounded.

3.3.3 Summary
Market power in network markets may arise from a wide variety of sources. Government policies
may be among these, and may for instance be motivated by fears for market instability, wasteful
competition, price wars, or destructive competition. Another possible source of market power
concerns the control of critical inputs. For transportation, this may occur at least in the short
run if infrastructure capacity is (nearly) fully utilized.
Economies of scale and scope may provide fundamental economic reasons for the spontaneous
emergence of market power in free network markets. Entrance of the market then becomes
difficult, as its success depends on whether the entrant can manage to start off at a size that
allows a sufficient exploitation of economies of scale and scope, and that hence enables the firm
to soon offer a competitive service, without making losses
Economies of scale and scope may arise for a wide variety of reasons. We discussed fixed costs
of infrastructure, economies of train or aircraft size, a more intensive use of trains and aircraft
over the day, specialization of employees, and various examples of economies of scope. A specific
type of economies of scope concerns network economies, for which hub-and-spoke networks are
an important example in transportation markets. Finally, frequency benefits for consumers may
lead to economies of scale for the firm if it succeeds in capturing these via increased ticket prices.
These economies of scale and scope create an interesting dilemma for governments. On the
one hand, there are advantages in exploiting economies of scale and scope. On the other hand,
this typically requires the presence of only one or a few operators, who will consequently have a
significant degree of market power and may set prices well above marginal cost.
Chapter 4

Network Sectors: Privatisation and

(de)regulation

4.1 Introduction
In the 80s a formal start was made on the privatization and deregulation of network sectors in
the Netherlands. Prior to 1989 the then called Staatbedrijf der Posterijen, Telegrafie en Telefoon
(PTT) (State-owned Mail, Telegraphy and Telephony Company) already faced competition in
a number of sub-markets, such as parcel post and express mail. In 1989 this situation was
regulated by law, which meant the PTT formally lost its monopoly in several sub-markets. At
the same time the PTT became independent, which was followed by (partial) privatization in two
stages in 1994 and 1995. The main objectives of this operation were to increase freedom of choice
for the consumer, increase the efficiency of the former state-owned company and supply products
and services with a better price/quality ratio. This development coincided with the current
trend in many Western industrialized countries, whereby the market mechanism is introduced in
markets that, traditionally, were operated by public monopolies. In the mid 90s this policy was
intensified in the context of the MDW operation.1
The privatization of the PTT does not stand alone. Since the 80s many of the sectors in which
public monopolies operated now faced a change in legislation which implied that the government
has less to say in decisions made by the management. Some of the many examples in the case of
network sectors are the electricity market, the gas market, the water market, the Dutch railways,
the cable market and the regional transport companies. Some of these privatizations have been
largely successful (PTT), others less so (rail). Therefore, the economic and public debate focuses
especially on the critical success factors of privatization and deregulation. In this chapter we
explicitly deal with the following questions
What are exactly network sectors and why should they be privatized? What does economic
theory teach us about the desired market form and ownership proportions in different types of
1
Market Mechanism, Deregulation and Legislative Quality.

53
54 Chapter 4. Network Sectors: Privatisation and (de)regulation

network markets, taking into account the social welfare effects of the different options? And
finally, but definitely not less important, what are the implications for the necessary control,
enforcement and monitoring objectives? Or, how are we able to deregulate these sectors the
best we can.
This chapter will use the theoretical framework of market forms as explained in Chapter
3. Besides those market forms it offers a discussion about the concept of ‘natural monopolies’
as well—a market form closely related with network sectors. We start with a definition of
network sectors and an overview of networks sectors in the Netherlands. Thereafter we deal
with instruments of privatization and, subsequently, an exposition of instruments of regulation is
offered. As a final remark, we would like to stress that we intend to provide the reader with a first
impression of the literature that has accumulated over the last couple of decades. To give a full
account of all the nitty-gritty of privatization and deregulation probably requires several (rather
thick) volumes. The main purpose of this chapter, however, it to provide business and economics
students with a basic understanding of current and future developments in strategically and
economically very important sectors.

4.2 Network sectors in the Netherlands

Obviously, the first question asked is why certain sectors, such as natural gas, electricity and
telephony, are referred to as network sectors. The literature does not seem to give a conclusive
definition of network sectors. It is possible, however, to list a number of characteristics of network
sectors that most sectors normally referred to as network sectors will possess. We deliberately
refer to most sectors because it is quite possible, for each characteristic, to find a network sector
that does not have this characteristic. This section will, first of all, list different sectors that
are generally classified as network sectors. Next, three characteristics are discussed that each
network sector should, in principle, possess. For each of these characteristics it will be indicated
specifically which network sectors do not possess this characteristic.
With regard to the following sectors there is a general tendency to consider them network
sectors: Although these sectors differ widely between each other, it seems that most of these

Table 4.1: Network sectors. Source: Aalders et al. (2002)

Gas Electricity
Telecommunication Cable
Television Radio
Mail Drinking water
Waste water collection Waste water treatment
Rail City transport
Regional transport Airports
4.2. Network sectors in the Netherlands 55

sector possess the following characteristics:

1. The presence of (physical) infrastructures used by companies to supply products and/or

services to consumers;

2. The presence of economies of scale in (a certain section of) the network sector;

3. The presence of a natural or legal monopoly in (a certain section of) the network sector.

The first characteristic often mentioned in relation to network sectors is the presence of
(physical) infrastructures used by companies to supply products and/or services to consumers.
Nearly all network sectors appear to possess this characteristic, although there does appear to be
a dichotomy. There are sectors in which there is a real tangible network of arcs and nodes, such
as the electricity, gas and water sectors, and sectors where there is only a network of nodes, such
as mobile telephony (antenna masts), aviation (airports), mail (sorting centers and mailboxes),
radio and television (antenna masts) and bus transport (bus stops). The problem with using this
interpretation of ‘infrastructure’ is that nearly every sector in our economy becomes a network
sector. After all, every sector uses a network of nodes. For instance, the wholesale sector supplies
its products via a network of stores and the agricultural sector consists of a network of farms.
We must therefore conclude that there is not a single definition of the term ‘infrastructures’ that
provides an exact demarcation between network sectors and non-network sectors. Although the
presence of arcs is a sufficient condition (if a sector uses a tangible network with arcs this sector
is a network sector), it is not an essential condition (if a sector is a network sector it does not
necessarily use a network connected by tangible arcs).
The second characteristic that is often associated with network sectors is the fact that there
must be (substantial) economies of scale present in the network sector as a whole or in a part of
the network sector. The rationale is that, for cost efficiency reasons, the production of certain
products or services—e.g., gas transport via pipelines—must in that case be left to one company.
After all, if there are continuous increasing economies of scale, one company can always produce
more cheaply than a combination of two or more companies. In such a case we refer to the
term ‘natural monopoly’. In cases where the company only produces one product the presence
of economies of scale is a sufficient, but not an essential condition as, even if after a certain
production level the economies of scale change into disadvantages, there may still be a natural
monopoly. If, in contrast, the company produces two or more products, the presence of economies
of scale for all products is neither a sufficient nor an essential condition. After all, this may
involve relatively large negative synergy effects. Examples are problems with regard to the
management of large companies that produce very diverse products, or a competency conflict
with regard to the financing of investments in cases were the borrowing capacity of the company
is limited. Even if, within a certain sector, there are both economies of scale for all products
and synergy effects between all products, such a sector is not necessarily a natural monopoly.
For a more detailed discussion about natural monopolies, the reader is refered to appendix 4.B.
56 Chapter 4. Network Sectors: Privatisation and (de)regulation

The third characteristic is the fact that ‘somewhere’ in the network sector there must always
be a natural or legal monopoly. The entire network sector may be a natural monopoly, but
also a part of that sector, such as the tangible network in the electricity sector, which is up
to now still a legal monopoly. With regard to, for instance, the postal sector, there is a legal
monopoly because no company other than TPG is allowed to install postboxes and letterboxes.
Another example of a legal monopoly is bus transport. Because buses are only allowed to take
on passengers at bus stops—this is called the boarding rule—a company that has a permit for
installing and managing bus stops has a legal monopoly. This characteristic need not apply
to all network sectors. Firstly, for instance, there may not be a natural monopoly but rather,
e.g., a natural duopoly or, more commonly, a natural oligopoly. An example of such a situation
could be the postal sector if the decision were made to lift the legal barriers with regard to
postboxes and letterboxes. There could, for instance, be a U-shaped cost function, which means
that the economies of scale in the sorting and delivery of mail can, at a specific point, turn into
diseconomies of scale when congestion effects take place. An example of a natural oligopoly
can perhaps be found in the mobile telecommunication sector, where parts of the network are
constructed jointly in order to achieve cost savings. It must be noted here that the question
whether there is a natural monopoly or a natural duopoly does, apart from the costs, also
depend on the demand. For instance, in the expectations for the UMTS (Universal Mobile
Telecommunication System) auction the mobile telecommunication sector was perhaps a natural
oligopoly and in the current expectations it is perhaps a duopoly. From this example we can
derive the more general point that a legal monopoly need not necessarily go hand in hand with
a natural monopoly—although this can be often the case. For instance, in developing countries
there is, in most cases, no boarding rule, resulting in the fact that there are many companies
offering bus transport services (and buses seemingly randomly stopping everywhere).
The discussion of above shows that it is possible to give common characteristics for a number
of network sectors, but that in all cases these characteristics are either too restrictive or not
restrictive enough. However, for reasons of simplicity, we assume that all sectors referred to
above are network sectors. Although many network sectors have already (partly) been privatized,
it may be insightful to look at the firms the Dutch government still (partly) owns. Table 4.2
provides an overview of the firms in the Netherlands which are still (partly) owned by the Dutch
government.
4.2. Network sectors in the Netherlands 57

Table 4.2: Firms and the degree of ownerships of the Dutch government. Source: De Volkskrant,
13 Juli 2005

Firm % Firm %

Energie Beheer Nederland 100 KG Holding 100

Bank Nederl. Gemeenten 50 NOB Holding 100
KPN 19 Saranne 100
TNT 10 Prorail 100
Luchthaven Schiphol 75.8 Twinning Holding 100
Nederlandse Spoorwegen 100 Noordelijk Ontwikkelingsmij. 100
Tennet 100 Industriebank LIOF 94.3
Westerscheldetunnel 95.4 DLV Adviesgroep 82
Ultra-Centrifuge Nederland 98.9 Groningen Airport Eelde 80
Nederlandse Munt 100 Ontwikkelingmij. Oost Ned. 57.6
Connexxion Holding 100 Luchtvaartterrein Texel 65.3
SDU 100 Brabantse Ontwikkelingsmij. 64.5
Ned. Waterschapsbank 17.2 Nozema 59
Nederlandse Gasunie 10 Nederlandse Pijpleidingmij. 50
Air France – KLM 14.1 Luchthaven Maastricht 34.8
Nederlands Inkoop Centrum 100 AVR Chemie CV 30
ADC Archeologisch Diensten Centr. 100 NIB Capital Bank 14.7
FMO 51 Eurometaal Holding 13.3
CF Kantoor v. Staatsobligaties 100 MTS Amsterdam 5
Covra 100 Thales Nederland 1
De Nederlandsche Bank 100

Note that Table 4.2 gives an overview of firms. It does not mention anything about public
services, such as hospitals and schools, which may be subject to privatization in the near future
as well.

Although there are several strong reasons to privative, the specific characteristics of network
sectors (most notably the occurrence of natural monopolies) usually force a government to still
exercise some control over these sectors. The government usually has several instruments to its
disposal, from which the next section discusses the most important ones.
58 Chapter 4. Network Sectors: Privatisation and (de)regulation

4.2.1 Summary
This section dealt with the question what exactly network sectors are. We provided
a list of network sectors in the Netherlands and some common characteristics these
network sectors usually exhibit. It turns out that network sectors are usually (i )
accompanied by large (physical) infrastructures, (ii ) display substantial economies
of scale somewhere in the production range due to the large fixed costs needed to
construct the network, and (iii ) usually display some sort of natural monopoly. This
section ended with a list of firms in the Netherlands who are still partly owned by
the Dutch government and are usually operating in a network sector.

4.3 Government regulation instruments

This section deals with the instruments available to the government in order to exercise control
over a certain sector/firm. The instruments available to the government for the regulation of
network sectors can be classified into five types:

1. changes to the external structure;

2. changes to the competition modality;

3. price or tariff regulation;

4. changes to the ownership proportions; and

5. additional regulation (quality requirements, UDV etc.)

Below, we briefly discuss these five types.

Changes to the external structure If the supervisory authority decides to change the ex-
ternal structure of a certain sector, this obliges companies in this sector to either merge or
dispose of parts of their business. Four types of change can be distinguished in this process:
horizontal integration, horizontal division, vertical integration and vertical division. In the
case of horizontal integration there is a merger between two companies that operate in the
same market. Horizontal division refers to dividing one company into two or more smaller
companies. The smaller companies operate in the same market(s) as the original company.
Vertical integration refers to a merger between two companies that operate in the same
industrial column whereby the output of one company serves as the input for the other
company, in other words: it involves adjoining markets. Vertical division takes place if a
company that operates in two adjoining markets is divided into two (or more) companies
that no longer operate in the same market, whereby the output of one company/companies
serves as the input for the other company/companies.
4.3. Government regulation instruments 59

Changes to the competition modality In the event of changes to the competition modality,
there is a change in the way in which companies compete in a sector. We distinguish
five competition modalities: competition in the market, competition for the market,
competition between markets, competition by comparison and no competition. To indicate
a change to competition modality the terms deregulation and regulation are normally
used. These terms indicate a specific change to competition modalities as a result of a
government decision. For instance, a market is deregulated if the government decides to
increase the extent to which companies must take the performance of other companies into
account, with the objective of increasing the level of social welfare. In this context we also
use the term liberalization. Deregulation does not necessarily mean that the number of
regulations companies must take into account is reduced, but rather that the character of
the regulations changes: companies must now take the decisions of other companies into
account to a greater extent. A market is regulated if the extent to which companies must
take the performances of other companies into account decreases as a result of a policy
change on the part of the government. We also refer to ‘deliberalization’. Once again it
is not so much the number of regulations that changes but rather the character of the
regulations.

Tariff regulation Tariff regulation is involved if the supervisory authority announces regula-
tions a company must take into account in determining its pricing structure. Well-known
examples of tariff regulation are ‘rate of return’ regulation2 and ‘price cap’ regulation3 .
We can distinguish between direct and indirect tariff regulation. There is direct tariff
regulation if the company cannot influence the price level of the individual goods. In that
case it may be that the supervisory authority (i ) determines the prices in advance; (ii )
approves the prices retroactively; or (iii ) formulates regulations that fully determine the
prices. There is indirect tariff regulation if the government allows the company room to
determine at least part of the prices of the individual goods.

Changes to the ownership structure In the case of changes to the ownership structure as
a result of regulation, the question is whether a company comes into the hands of the
government or the private sector. We refer to privatization (of a company) if the ownership
of the company is transferred from public to private hands. Privatization, therefore, is
entirely separate from deregulation. As an example, it is possible to only deregulate a
certain sector and not privatize it or, conversely, to only privatize the sector and not
deregulate it. The merits of privatization must therefore be considered separately from
deregulation. However, it is possible to have an interaction between deregulation and
privatization. There can be full or partial privatization. In the latter case we talk about
a company with mixed ownership. It is also possible to have both public and private
2
Loosely speaking, rate of return regulation means that a firm is not allowed to make a profit above a certain
profit margin.
3
Price cap regulation entails that the supervisory authority sets a maximum price.
60 Chapter 4. Network Sectors: Privatisation and (de)regulation

companies operate in the same market. Nationalization (of a company) is involved if the
ownership of the company is transferred from private to public hands. Nationalization is
completely separate from (de)regulation as well. We refer to ‘renationalized’ if a company
is once again nationalized after it was previously (partially) privatized.

Additional regulation Finally, the government can impose additional requirements on the
companies that operate in a sector. Examples are universal service provision (UDV),
standards, quality requirements and requirements with regard to interconnection and
obstacles to competition (‘competitive bottlenecks’). With regard to standards we can
think of, for instance, requirements relating to number portability and requirements relating
to the quality of the gas and electricity that producers import into the transport network.
With regard to quality requirements we are talking about requirements that have the
nature of a public good, in other words: it is technically not possible to supply products
and/or services of a different quality level to individual consumers. Examples are the
quality of drinking water, the average number of hours a consumer may be left without
electricity or gas during the year and the maximum number of delays in train journeys.

By using these instruments the supervisory authority can influence the level of so-called
social welfare in a (network) sector (see the next section). In order to achieve the predetermined
objective of an improvement to the social welfare level there often needs to be a package of
measures, as the individual instruments can influence more than one objective. For instance, if
the supervisory authority were to introduce ‘price cap’ regulation to a market this would give
companies an incentive to reduce the quality. If, from a social point of view, the quality of a
product in a certain network sector is (very) important, it would make sense to introduce a
minimum quality requirement or, if such a requirement already exists, to upgrade it.

4.4 Social welfare

In order to be able to assess which form of government control is desirable from a social perspective
we use the economic concept of social welfare. This concept weighs up the advantages and
disadvantages that are the result of a certain policy change—for instance regulation—in terms
of social welfare. An example will clarify this. Let’s say that, in the current situation, a certain
network sector is not being regulated and the government wonders whether it would be desirable
to regulate this sector. Then, for both situations, it is possible to calculate the social welfare
in both situations. The situation with the highest corresponding social welfare is then the
most preferred one. The term social welfare can be defined as broadly as possible, however we
distinguish only the following three dimensions for reasons of clarity: consumer surplus, producer
surplus and government revenue. In addition to the level of social welfare, the distribution of
social welfare is relevant as well. Before we can further define the dimensions of social welfare it
must be noted that economic theory does not give a definite answer to the question in which
4.4. Social welfare 61

way these three dimensions, and their distribution, must be weighed up against each other. This
is, and always will be, a political choice. It is however possible to argue that, from a standpoint
of, for instance, economic efficiency or equal distribution of welfare, certain dimensions (must)
weigh heavier than other dimensions. We now first give an overview of the dimensions that
play a role in determining the level of social welfare. Thereafter, we provide an overview of the
distribution aspects that play a role.
With regard to social welfare we distinguish the following three dimensions:

Consumer surplus The first dimension that is relevant to social welfare is the consumer
surplus of the used product. This consumer surplus is a standard to measure the welfare
of consumers. Loosely speaking, the consumer surplus equals the difference between the
maximum price consumers are willing to pay for a certain product and the price they
actually pay. Let’s say, for instance, that a consumer is willing to pay a maximum of
¤500 for a mobile telephone and that this mobile telephone costs ¤300. This means the
consumer surplus is ¤200. This implies that, if the price of a product decreases, the
consumer surplus increases (and vice versa). If there are a number of consumers, the
surpluses of all these consumers can be added together. Let’s say, for instance, that there
are two consumers with surpluses of ¤200 and ¤100, respectively. This results then in
a total consumer surplus of the used product ¤300. With regard to consumer surplus a
distinction must be made between private and public goods.

producer surplus The second dimension that is relevant to social welfare is the producer
surplus of the used product. This producer surplus equals the surplus profit achieved
jointly by all companies operating in a certain market. In this context, ‘surplus profit’ is
defined as the revenues minus costs. As explained in Chapter 3, in a situation where there
is perfect competition the producer surplus of the used product is zero: in this case there
is no surplus profit. If companies have market power (recall Figure 3.1 again) the surplus
profit—and therefore the producer surplus—is greater than zero. Generally, the larger the
market power, the larger the producer surplus (this reveals the incentive why firms usually
have a desire to grow).

Government revenue The third dimension that is relevant to social welfare is government
revenues and costs, as policy changes result, in addition to changing the consumer and
producer surpluses, usually in changes in government revenues as well. Firstly, a change in
regulation for a certain market will generally result in different prices, costs and profits.
Because the government levies certain taxes, such as VAT, income tax and corporate
income tax, government revenue will also change. Secondly, changes in the regulation of,
for instance, the natural gas market, will generally result in a different level of natural gas
revenue. Thirdly, the change of ownership structure (privatization and nationalization)
will also result in changes to government revenue. Fourthly, changes in regulation can also
62 Chapter 4. Network Sectors: Privatisation and (de)regulation

result in changes to regulation costs. Depending on the change of regulation, all these
changes can have either a positive or negative effect on the government coffers.

As mentioned above, apart from the level, the distribution of social welfare is relevant as well.
This distribution can relate to both the distribution between any of these three groups, for
instance the distribution between producers and consumers, and the distribution within each
group, for instance the distribution between consumers with low and high incomes. With regard
to distribution, three dimensions can be distinguished: by group, by space and by time.

Distribution by group This can relate to the distribution of welfare between consumers with
high and low incomes, but also to the distribution of welfare between producers and
consumers or between the own country and other countries. These distribution aspects
can be relevant as changes in regulation often do not have identical results for everyone.
For instance, a certain tariff change may have the effect that consumers who frequently
use a service will be paying relatively less for the service, whereas consumers who use a
service relatively infrequently will pay more. If the consumers who frequently use a service
are also the consumers who have a higher average income, the tariff change will result in
welfare being distributed less equally. If society considers an equal (income) distribution to
be important, this will then result in a reduction of welfare. Again, it is important to stress
that the relevance of an equal income distribution (or of a certain distribution of welfare
between producers and consumers, or between the own country and other countries) is a
political issue.

Distribution by space With regard to distribution aspects by space we must consider the
effects of imposing an universal service distribution (UDV). If companies have an UDV
obligation this means that they must offer the same service to all customers at the same
price, regardless of their location. In some situations a distinction can be made between
regions. Examples of an UDV can be found in the postal sector, where mail items to
a maximum weight of 20 grams are always delivered at a rate of 85 cents, regardless of
the dispatch location and destination. Another example can be found in the electricity
sector, where all consumers in a certain region pay the same amount per month for their
connection, regardless of whether they live in a rural area, a village or a city. The main
characteristic of an UDV is the fact that consumers who are subject to an UDV and
live in villages or rural areas pay less for the service they receive than they would in a
situation without an UDV. An UDV, therefore, influences the distribution of welfare by
space. Whether an UDV is advisable from a social point of view depends on the costs
and benefits of an UDV. Several studies have shown that the costs associated with an
UDV must not be ignored. These costs are generated by the fact that the customers
are not faced with the costs their decisions incur. In fact, an UDV creates implicit price
discrimination as discussed in subsection 3.2.2, because consumers do not have to pay for
4.4. Social welfare 63

the true marginal costs for their service.4 The ultimate consideration as to whether or not
an UDV is advisable from a welfare perspective is once again a political matter.

Distribution by time With regard to distribution aspects by time we must consider, for
instance, the effects of regulation changes on welfare over time. As an example, a change
in regulation will often result in disadvantages in the form of transition costs, whereas the
advantages are not realized until later. These advantages and disadvantages at different
times can be weighed up against each other by discounting the effects. After all, today’s
euro can be invested, which means it will be worth more in the future than it is worth
today.

As a criterion, an unweighted sum of the consumer surplus, the producer surplus and
government revenue is often used. Sometimes, however, distribution effects are too relevant
to be not taken into account. Again, we stress here that distribution effects are ultimately a
political choice. If the government—or society in general—feels that it is very important that
train services in peripheral areas are to be maintained, although they are in no way profitable,
then that must be taken into account as well.

4.4.1 Social welfare and regulation

This subsection deals with three points for attention that are relevant to the level of social welfare
should the government decide to regulate. Firstly, there is the interaction between regulation on
the one hand and innovation and technological development on the other hand. Secondly, there
is the influence of the historical situation the government has to deal with. Thirdly, there is the
potential existence of transaction costs.

The interaction between regulation and technological development A change in reg-

ulation may affect the technological development in a sector. In this development we must
distinguish between process innovation and product innovation. With regard to process
innovation it is a known fact that under ‘price cap’ regulation, more process innovation
takes place than under ‘rate of return’ regulation. The reason for this is the fact that
under ‘price cap’ regulation companies are allowed to keep a larger proportion of their
profits. Another example of the influence regulation can have on process innovation can
be found in the telecommunication sector, where the technological development increased
after deregulation of the market. This has resulted in the application of countless new
technologies (such as fibre optic cable) and the introduction of new services (such as
mobile telephony). However, regulation can also slow down the technological development.
For instance, regulation of the infrastructure based on a ‘price cap’ can result in delays
in the introduction of new processes and technologies. After all, until a company has
4
If the NS did not have to comply with the UDV most of the peripheral rail services would already have been
abandoned long ago. This implies that railway customers in the Randstad are subsidizing railway customers in
the peripheral regions.
64 Chapter 4. Network Sectors: Privatisation and (de)regulation

recouped its investment in the old infrastructure it will not be inclined to realize a new
and better infrastructure. With regard to product innovation, the change from ‘rate of
return’ regulation to ‘price cap’ regulation for existing products, combined with an absence
of regulation for new products during a limited period of time, results in a substantial
acceleration in the introduction of new services and products. The reason for this is the
fact that this form of regulation makes it more attractive for companies to introduce new
products. On the one hand the new service can be introduced much faster (as it is only
preceded by marginal testing) and on the other hand the company can charge market
prices for a certain (although limited) period. One important point with regard to both
process and product innovation is the fact that, although regulation in the Netherlands
affects the speed with which new products and services are introduced, it does not affect
the technological development itself. After all, this technological development takes place
mostly in companies that produce for a world market (for instance Nokia for the devel-
opment of mobile phones). A change in regulation in the Netherlands will hardly, or not
at all, affect the speed with which Nokia develops mobile phones, as the Dutch market is
too small. With regard to technological development, therefore, the Netherlands is really
a ‘free rider’. A change in regulation in the Netherlands will therefore mainly affect the
introduction of new processes and products and not so much the technological development
itself.

Table 4.3 provides an overview of the importance of innovation and technological devel-
opment per sector. This classification is important, as it provides an insight into the
effects of a change in regulation on the introduction of new processes and products. If the
relevance of technical development is low, the introduction of new processes and products
need not be taken into account in the current regulation. If, in contrast, the relevance of
technological development is high, the introduction of new products and processes must
most certainly be taken into account in the decision to regulate. Table 4.3 shows that,
for most network sectors, the relevance of technological development is low for both the
infrastructure and the products and services supplied via this infrastructure. Only for
two sectors is the relevance of technological development of both the infrastructure and
the products and services high. This concerns the telecommunication sector (fibre optics,
mobile telephony, Wireless Application Protocol, Wireless Local Loop, etc.) and the cable
sector (Internet use, video on demand, telecommunication, etc.). For only two other sectors
is the relevance of technological development of the infrastructure average. These sectors
are the postal sector (new sorting technologies) and the railways (high-speed lines). With
regard to the services and products there are additional sectors in which there is a question
of the relevance of technological development that may be described as average. In this
context consider drinking water and waste water (new water treatment technologies) and
mail (electronic direct mail). In addition the electricity sector is testing the multiple use
of its networks for the purpose of, for instance, sending information.
4.4. Social welfare 65

Table 4.3: The relevance of technological development in network sectors. Source: Aalders et al.
(2002)

Technological development of . . .
Sector infrastructure products and services

Gas low low

Water
drinking water collection low average
waste water collection low low
waste water treatment low average
Electricity low low–average
Telecommunication high high
Information
cable high high
television low low
radio low low
Mail average average
Public transport
railways average low
City and regional transport low low
Airport low–average low

The influence of the historical situation The historical situation can have a significant
influence on the effects caused by a change in regulation. In this context the influence of
the so-called absolute barriers is of particular importance. These barriers are created by
the fact that the incumbents have an inherent advantage over newcomers in the market.

A first example of absolute barriers is the existing long-term contracts that incumbents
imposed in the past on suppliers or consumers. This makes it difficult, if not impossible,
for a newcomer to enter the market. For instance, Belgian electricity company Electrabel
entered into a contract with the distribution companies shortly before the deregulation
of the electricity market. This contract was valid for 10 to 15 years, which means the
Belgian market for private users remains closed to competitors until the contract expires.
Another example of absolute barriers concerns the contracts between Gasunie and the
companies that extract gas in the Netherlands. These contracts are often valid for 30 years.
Because gas in the Netherlands can be extracted relatively cheaply (owing to the relatively
short transport distances on the one hand and the presence of a very cheap gas field in
Groningen on the other hand), Gasunie competitors do not have access to the same inputs
on the same conditions.

A second example of absolute barriers is the fact that, in the starting phase, the incumbents
66 Chapter 4. Network Sectors: Privatisation and (de)regulation

often have very large market shares. This legacy, which results from the fact that in
the past these parties were often a public monopoly, can thoroughly disrupt the market
forces. In cases where there is excess capacity at the moment of the change in regulation,
the incumbent also has the option to use this excess capacity to deter newcomers by
threatening a price war as soon as they enter the market.

Transaction costs Transaction costs are costs that agents (consumers, companies and author-
ities) must incur to keep the economic system going. Examples of transaction costs are
solicitors’ and consultants’ fees for drawing up contracts. Transaction costs are relevant
because they can differ (considerably) under different circumstances. For instance, the
costs associated with entering into a contract for the purchase of electricity in a regulated
market will be much lower than the cost of entering into a similar contract in a deregulated
electricity market. The reason for this is the fact that, in a deregulated electricity market,
many more issues must be expressly stipulated in the contract (whereas before they were
arranged by implication). Consider, for instance, the purchase of reserve capacity at certain
times, the financial covering of risks, purchasing electricity at different times and different
costs, etc. Of course, in regulated markets these matters are set out in contracts as well.
However, the detail in which contracts now regulate these elements often increases strongly
in a deregulated market.
Transaction costs are not stand-alone costs and have therefore not been included as a
separate dimension of welfare. Transaction costs form part of either the consumer surplus,
the producer surplus or the government revenue. If the transaction costs increase, welfare
in one or more of these dimensions will be reduced. For instance, if companies need to
incur transition costs either the producer surplus is reduced (if they do not charge these
costs on) or the consumer surplus is reduced (if they do charge these costs on).
Transitions costs can in some case be quite important. Transition costs incurred by
companies in particular can be considerable, as can regulation costs. However, these costs
must be offset against the welfare benefits that can be achieved through monitoring. One
important point in this context is the fact that, when the price elasticity of demand is small,
the potential welfare benefits are greater. Because in a number of network sectors there is
a relatively low degree of demand elasticity (e.g., for electricity, gas, water and rail), the
provision of information by companies—and therefore the monitoring of companies—will,
from a welfare perspective, soon be required.

4.5 Market failure and government failure

By using regulation the government can try to influence social welfare. Regulation is advisable
from a welfare perspective if the following two conditions are met. Firstly, there must be a valid
reason for intervening in the functioning of the market. This reason exists if the market is unable
4.5. Market failure and government failure 67

to independently achieve the optimum level of welfare. If this is the case we refer to the term
‘market failure’. By intervening the government can try to increase the level of welfare. Secondly,
welfare must increase as a result of regulation. In some cases this condition will not be met.
In these cases we refer to the term ‘government failure’. This section will provide an overview
of the main forms of market and government failure, and of the relevance of such failures for
network sectors.

4.5.1 Market failure in network sectors

Market failure occurs if the free functioning of the market mechanism does not result in an
efficient outcome. The five main causes of market failure are: market power, external effects and
public goods, information problems, specific capital goods and equivalent lease revenue. Each of
these causes can affect both the scope and the distribution of social welfare. These five causes of
market failure are briefly described below.

Market power Market power exists if market parties are able to influence the price. There
are many reasons why there may be market power in a certain market. One important
cause is a situation where there are so few companies operating in a market that it can
no longer be called a competing market. In such a situation there may be explicit or
implicit collaboration+++++. The chances of implicit collaboration are relatively high if
market parties regularly encounter each other in a standard situation. This is the case, for
instance, in the electricity production market. The existence of a natural monopoly or
oligopoly is an important reason why a limited number of companies or only one company
operate(s) in the market. Such a situation is one of the main causes of market failure in
network sectors. Apart from too small a number of companies, market power can also be
‘promoted’ by the presence of so-called competition-limiting factors. Examples of these
factors on the supply side of the market are the presence of entry barriers and, on the
demand side of the market, transfer costs and the relative power position of consumers.

External effects and public goods The existence of external effects can also result in market
failure. External effects are involved if certain effects associated with the production of
a specific product or service are not expressed in the private costs and revenue of the
producer. In other words, the social costs and benefits differ from the private costs and
benefits. External effects can be either positive or negative. In the event of positive
external effects, free functioning of the market mechanism creates a production level that
is too low from a welfare perspective. In such a case the level of social welfare is higher
than the sum of the consumer and producer surplus. Examples of positive external effects
in network sectors are the network effects that occur between people who have a telephone
connection. If a consumer is connected to the fixed telephony network this increases the
level of welfare for all consumers who are connected to this network, because it means they
can get in touch with more people. In the case of negative external effects the market, in
68 Chapter 4. Network Sectors: Privatisation and (de)regulation

contrast, creates too high a level of production. In that case the level of social welfare is
lower than the sum of the consumer and producer surplus. Examples of negative external
effects in network sectors are environmental pollution and security of supply.

Public goods are products or services of which everyone consumes the same, because
these goods are not exclusive. It is therefore impossible for one consumer to use more
public goods than another consumer. Examples of public goods in network sectors are the
safety of nuclear power plants, the safety of drinking water and the quality of the services
provided by the Nederlandse Spoorwegen (Dutch Railways). If the quality of a certain
product increases this will, in itself, have a positive effect on social welfare. Whether
the level of social welfare actually increases obviously depends on the cost of this quality
improvement. Table 4.4 provides an overview of the importance of quality in the various
network sectors. The table makes a distinction between the importance of quality in
relation to the infrastructure and in relation to the service/product that is supplied via
the infrastructure. In a number of cases an infrastructure is marked ‘essential’, as a poor
quality of this infrastructure can cause major detriment to either the economy (electricity)
or public health (gas, cable, television and radio). For instance, a loss of gas pressure can
result in fatal accidents, as most gas-fuelled equipment in the Netherlands does not have
automatic safety valves. This means that gas could escape at the moment the pressure
is restored. The importance of a good quality of the infrastructure for cable, television
and radio is ‘essential’ because they provide information during national emergencies. In a
number of cases the quality of a service is also classed as ‘essential’. This is the case in
the gas (once again because of accidents), water (public health) and electricity (economic
importance) sectors. Table 4.4 shows that for all network sectors the quality of at least
one of the two dimensions (infrastructure or product/service) is significant if not essential.
The government has a clear responsibility here.

Information problems The existence of information problems is a third potential reason for
market failure. Without the availability of sufficient information on which market parties
can base their decisions, market results will not be efficient. Insufficient information could
result in companies producing products that are too high or too low in quality, or companies
using inefficient technologies. Insufficient information among consumers could result in
consumers in network sectors unwilling to change their supplier, because they might worry
this will be at the expense of the speed with which they receive assistance in case of faults
etc. In addition to insufficient information being available, an asymmetric distribution
of information can also cause market failure. In extreme cases this could result in the
disappearance of markets that are desirable from a welfare perspective. Regulation can
ensure efficiency benefits in two ways. Firstly, better information provision can result in
the market parties being better informed with regard to their purchasing decisions, which
creates a better ‘match’ between consumers and producers. Secondly, establishing minimum
4.5. Market failure and government failure 69

Table 4.4: The relevance of quality in network sectors. Source: Aalders et al. (2002)

Quality of the . . .
Sector infrastructure products and services

Gas essential essential

Water
drinking water collection significant essential
waste water collection significant n.a.
waste water treatment significant significant
Electricity essential essential
Telecommunication significant high
Information
cable essential average
television essential average
radio essential average
Mail average significant
Public transport
railways essential significant
City and regional transport average significant
Airport essential average

standards can result in consumers who are (insufficiently) informed being protected against
results that might have negative consequences for them, such as the purchase of a poor
quality product.

Specific capital goods Specific capital goods are involved if investments have such a specific
character that the alternative investment options are extremely limited or result in lower
revenues. Specific investments occur frequently in network sectors. Consider, for instance,
investments in electricity, gas, water, cable and telephony networks. However, other
investments can also be ‘specific’ in nature. One example of such investments concerns the
train carriages of the Nederlandse Spoorwegen (NS). Because trains in the Netherlands use
a different voltage than trains in the surrounding countries, a sale of these train carriages
will entail conversion costs. This results in reduced revenue. If the percentage of specific
capital goods in a sector is relatively high, so-called ‘hold-up’ problems can occur.5 This
means that insufficient investments are realised, because investors are worried that the
sunk costs cannot be recouped.6 Such a situation results in underinvestment, which is
undesirable from a welfare perspective.
5
These ‘hold-up’ costs described here are rather similar to the switching costs described later in both the
course and in Shapiro and Varian (1999).
6
Sunk costs are costs that have been incurred in the past. These costs are not relevant for current decisions.
70 Chapter 4. Network Sectors: Privatisation and (de)regulation

4.5.2 Government failure

We talk about a government failure when regulation on the part of the government is not
effective. The two main causes of government failure are: information asymmetry and the
so-called ‘principal-agent’ problem. The same rule that applies to market failure—namely that
it may affect both the scope and distribution of welfare—applies to government failure. We will
briefly look at these two causes.

Information asymmetry To answer the question what the socially desirable result looks like
in practice, a lot of information is required. Consider, in this context, information about
the cost function of companies and the demand functions that companies are confronted
with in the markets in which they operate. In addition, information is also required about,
e.g., relevant technological developments and the (likely) costs of such developments. The
companies that the government tries to regulate will often have this information available
but, because the company interest does not correspond with the public interest, they may
be reluctant to provide the government with this information. In this context we refer
to ‘asymmetrical information’: the government has information arrears compared to the
companies they want to regulate. There are two consequences. First, the government may
try to collate this information, which could entail considerable costs. Second, there is a
risk that the government has insufficient information to determine the socially desirable
outcome. This may result in the formulated policy objective not being correct from a
welfare perspective and the government intervening in the market process in a suboptimal
way. In some cases it is even possible that, as a result of government intervention, the
level of social welfare is reduced rather than increased. The importance of asymmetrical
information in network sectors increases as:

• fewer companies operate in the market. This makes it more difficult for the supervisory
authority to obtain information by comparing companies with each other; and
• the dynamics in a sector are greater, for instance because the level of technological
development is high. This makes it more difficult for the supervisory authority to
obtain insight into the cost functions of companies and the demand for products and
services in a market.

For many network sectors the rule applies that both the demand and cost functions
change only marginally from year to year. One important indicator of this effect is the
technological development. If this development is low or average, this will allow the
supervisory authority to obtain a good impression of a certain sector over the course of
time. At present, asymmetrical information only appears to be a significant problem in
the telecommunication and cable sectors, because the technological development in these
sectors is relatively high.
4.5. Market failure and government failure 71

A specific form of information asymmetry occurs with regard to the verifiability of public
goods or interests. For instance, it may be difficult to determine the quality of a product,
which means that, under certain forms of regulation, companies have an incentive for
reducing quality. This form of asymmetrical information can occur in virtually all network
sectors, as quality is very important in these sectors (see Table 4.4). The essential question,
however, is whether the quality aspects of the goods supplied in network sectors are easily
determined or measurable. This is an area of concern for many network sectors. Let’s say,
for instance, that we want to measure the quality of the rail sector. To do so, we could
collate information about the number of faults and defects in points, tracks and trains.
This gives us a impression of the quality of the railway. But it is a delayed image! At the
time when data shows us that there are a lot of faults, the level of, e.g., maintenance has
usually been insufficient for a number of years. This means that the supervisory authority
only obtains a retrospective insight into the quality of the supplied product. Because the
faults may have been caused by neglect, it may take many years before the quality of the
service is restored. One example of this is, once again, the Nederlandse Spoorwegen. It
will take many years before the NS will once again be able to meet its punctuality targets.
The greater the social importance of quality, the more it will need to be taken into account
in the regulation of network sectors. The verification of universal service provision (UDV)
will, in general, not cause any problems as it concerns the question whether a company is
charging customers in a certain region the same price. This is relatively easy to check.

The ‘principal-agent’ problem The condition that, in formulating its policy, the government
should use social welfare as the standard, is not always met. This form of government
failure is a result of the way in which the decision process is designed. Mostly there is a
complicated structure with different principal-agent relationships.7 This may have the
effect that the decisions made by the government are not always aimed at achieving the
social welfare optimum, but also at realizing other objectives. One possible cause of this is
the role played by interest groups. These groups try to influence government policy toward
their preferred direction by influencing the policymakers. Their objective is to create
benefits (rents) that can be distributed among the members. Because it is easier for some
groups in society to unite than for others, this type of influencing results in an outcome
that may differ considerably from the welfare optimum. A second possible explanation is
the fact that executive authorities (such as a supervisory authority or a ministry) may be
trying to achieve their own objectives. This is a direct consequence of the ‘principal-agent’
problems that exist within the government. In this case the responsible politicians (the
principals) have insufficient means to control the actions of civil servants (the agents) and,

7
In economic science the term ‘principal-agent’ refers to the relationship between a boss (the ‘principal’) and his
subordinate(s) (the ‘agent’). An important theme in this literature is the fact that the boss does not know exactly
what his subordinates do, because he only has imperfect instruments to monitor the subordinate. Realistically,
this is therefore also a situation of information asymmetry.
72 Chapter 4. Network Sectors: Privatisation and (de)regulation

where necessary, redirect these actions. The core of this problem can be summarised as
follows :

“Hence, agencies have an opportunity to engage in ‘shirking’—consciously failing

to pursue the policy objectives that elected political leaders would desire.”

They can do this in different ways. For instance, it is possible that the civil servants
responsible for policy implementation make insufficient efforts to do so, because this
prevents unpleasant conflicts with the market parties involved. In addition, (the most
senior) civil servants may have their own political agenda which, when it comes to important
points, differs from that of their political principals. This may result in decisions being
made that are favorable for specific groups in society. Another possible cause is the fact
that decisions made by civil servants may be aimed at promoting their personal careers.
In this case acquiring power and respect for an individual civil servant is considered more
important than optimizing social welfare. Finally, it is possible that civil servants may try
to represent the public interests to the best of their abilities, but that their perception of
these public interests is limited. This may be the result of their educational background.
For instance, economists will focus mainly on efficiency, whereas lawyers will focus more
on legal policy aspects. It has been shown that transparency of the decision-making
process produce the most effective means of reducing ‘principal-agent’ problems. However,
increased transparency cannot fully prevent ‘principal-agent’ problems.

4.6 An example: the energy sector in the Netherlands

The whole discussion above tends to be a bit theoretical, without many links to government
practice. In other words, are we able to apply the theory of above on a recent example? To this
end, we look at bit further in the recent (de)regulation of the Dutch energy market.
In 1996, the European Commission published her guidelines for the liberalization of the energy
market, stating that all member states should have a free energy market for households in 2007.
Basically, this means that all households have freedom of choice of their energy provider. The
Dutch government decided to introduce a total free market energy market in July 2004.8 . The
main objectives of this liberalization were lower energy prices and eventually more innovation,
because customers were free to choose their best and cheapest energy supplier.9 Apart from the
fact that only a small amount of customers actually have switched from energy provider (and
often not even to the cheaper one), there are at the moment large concerns with respect to the
status of the energy networks.
During the first round of privatization in the energy sector, the national distribution network
was separated from the old state-owned company and transformed in a new firm (completely
8
The choice for sustainable or ’green’ energy was already liberalized in 2001
9
See also http://www.kiesenergie.nl/vragen/.
4.6. An example: the energy sector in the Netherlands 73

governmental controlled as well) called Tennet. This network ensures transmission between
energy plants and the regional distribution networks and all international connections as well.
Local and regional distribution networks are still owned by the energy providers, such as Essent,
Nuon, Eneco, and Delta. The regional distribution networks connect with the national network
of Tennet and the local distribution networks connect with customer (the households). Note
that the energy providers are still partly owned by regional governments (such as the province
of Gelderland in the case of Nuon). To completely privatize the energy sector, the Dutch
government has stated as necessary condition that the regional and local networks and the
provision of energy should be separated from each other (vertical division). There are two
reasons for this condition. First, government desires certainty in energy supply and, second, the
government does not want any impediments for competition by the vertically integrated energy
companies.
In the case of the Dutch energy sector, two specific advantages and two specific disadvantages
of vertical division can be given. To start with the advantages:

1. Vertical division may prevent abuse of market power. Namely, private incumbents on the
network market can prevent newcomers to enter the market. If there is a public network
operator, then the energy network can be used by any energy provider at normal market
prices.

2. Vertical division prevents the use of cross-subsidization. Namely, when integrated, stable
cash flows from the network companies can be used for other activities as advertising to
keep or increase market shares. Because network companies are heavily subsidized10 , cross-
subsidization leads to unfair competition. Regulation alone is usually not an attractive
option, because monitoring is hard due to the lack of transparency in the accounting
system. Division leads in any case to a higher transparency.

On the other hand, two disadvantages can be raised for the division of the Dutch energy sector.

1. The classical argument for integration is synergy between production processes (in this
case between distribution (network) en supply (the provider)). However, from international
(UK) experience we know that synergy on these markets is rather limited.

2. A second and more important argument is the presence of unequal market power on the
European energy market. Because foreign (read German and French) energy firms are
often heavily subsidized, the size of these firm is usually relatively large. Energy experts
expect that eventually the European energy market will converge to an oligopolistic market
(with four of five players). Because of the important strategic function of energy, this is
a undesirable scenario for the Dutch government. On the other hand, it is questionable
10
Energy providers purchased the regional of local networks for much less that it was worth at the time of
privatization
74 Chapter 4. Network Sectors: Privatisation and (de)regulation

whether vertical integration will offer dutch energy providers enough market power to
withstand fierce European competition.

Thus, it seems that, from a national view, vertical division is desirable. The question,
however, remains whether the regional and local distribution network companies should be
privatized or not.

4.A Integration versus division

In this chapter we have discussed briefly the concept of changes in the external structure
(division versus integration) as a possible instrument of the government to regulate network
sectors. Because of the large attention this topic recently received in both media and scientific
publications, this appendix dives a bit deeper in this concept.

4.A.1 Reasons for vertical integration

In the past, vertical integration was explained mainly from a technological perspective. However,
since the seventies the explanation for vertical integration is sought more in the area of transaction
costs. This concerns the relationship between the direct and indirect costs associated with
internal production as compared against the direct and indirect costs associated with purchasing
intermediary goods and services in the market, or in the words of Perry (1989):

Since the primary alternative to vertical integration is contractual exchange, trans-

action cost economics examines the relative costs of contractual versus internal
exchange.

The level of the transaction costs is strongly influenced by the extent to which capital goods are
specific. As long as capital goods are not very specific11 , vertical integration does not result in
major transaction cost advantages. Not only is the market better able to minimize production
costs in this situation (for instance as a result of advantages of scale), the relationship between
consumer and supplier is not very specific in character either. The latter results in the fact
that there are no major costs associated with formulating a contract. If capital goods are,
however, (very) specific in character, vertical integration does offer advantages. In this case the
specific character of the goods means that there are no advantages of scale that make large-scale
production by specialist companies attractive. Furthermore, entering into a contract with an
external supplier is expensive, as, because of the specific nature of the transaction, such a
contract will have to be very detailed. There is also the fact that, in practice, it is virtually
impossible to enter into a complete contract, which may indirectly result in additional costs.
The ultimate conclusion is that companies—and therefore network companies as well—will
look for the most efficient structure. Let’s say, for instance, that in a certain area more electricity
11
This means that investments have such a general character that the alternative options for the use of these
investments are extremely large and generate at least an equally high revenue.
4.A. Integration versus division 75

is consumed than produced. This means the network manager will have two deploy more
expensive production units which were initially not eligible for production. This situation can be
remedied in two ways: additional transport capacity can be constructed and new power stations
can be built in locations that are closer to the demand. With the correct regulation a vertically
integrated company will have an incentive to choose the most efficient solution. If the company
is divided, this incentive is not always present. In this case new institutions must be developed
that will resolve these problems. Consider, for instance, a one-off contribution—that differs per
location—to the cost of the network expansion or an adjustment of the transport rates. This
way the information exchange between the production company and the transport company is
formalized. Which situation is preferable from a welfare perspective depends on the costs of
both options.

Another example—in which it is more difficult to devise new institutions—is the division
of the rail company into a network company and a transport company. In addition to the
installation of the tracks, the network company is also responsible for safety. The problem,
however, is that in the eyes of the travelers the transport company will soon become the main
responsible party. In this situation the network company has insufficient incentives to invest in
safety. The supervisory authority will have to take this into account, but the best way to do
this is not immediately evident, as a decision must be made with regard to the optimum safety
level, which means an evaluation situation is encountered. Within an integrated company this
decision is incorporated automatically, but in the case of division the issue of safety must be
allocated an explicit price. Another problem is that consultation is now required between the
two companies regarding the scope and timing of investments in the rail network, train carriages
and safety. Because these decisions must now be jointly made by several parties, this may slow
down the decision-making process. This means that the division of business units can create
coordination problems.

Vertical integration can also take place in response to market imperfections. The most
obvious market imperfection is the lack of sufficient competition. A well-known example in
this context is the existence of a bilateral monopoly. This means that both the supplier of an
intermediary product and the seller of the end product operate in a monopoly market. Because
both parties have market power, each company will use its own price/cost margin. The so-called
‘double marginalisation’ ensures that the joint profits are not maximized. Vertical integration
can resolve this problem. Another example is a situation in which a monopoly producer uses
different inputs, one of which is supplied by a monopolist. This may result in an inefficient
input mix, because in the production of the end product there is an incentive to replace the
monopoly input with input that is supplied with sufficient competition. Although this results in
a sub-optimum structuring of the production process, the associated costs can be exceeded by
the price advantages in the purchasing process.
76 Chapter 4. Network Sectors: Privatisation and (de)regulation

4.A.2 Vertical division of network sectors

If vertical integration provides advantages for the companies involved, this does not automatically
mean that vertical integration is also desirable from a welfare perspective. Especially in the case
of network sectors, there is an important disadvantage associated with vertical integration. After
all, the integrated company is in a position of power as it owns the network that other companies
must use to be able to compete. Only if there are sufficient substitution options, there such a
position of power is not a reason for concern. For instance, if rail transport encounters sufficient
competition from road and air transport, the opportunity for anti-competition behavior on the
part of a vertically integrated railway monopoly remains limited. If this is not the case, this
may result in the integrated company increasing its monopoly profits by not allowing competing
companies to use the existing network, or only allowing them to use the network at high costs
(‘raising rivals’ costs). The company could also negatively influence the quality of the service
position to third parties (which could cause a relatively high level of disruptions) in order to put
the competition in a disadvantaged position.
The advantages of vertical integration must therefore be weighed up against its disadvantages.
Although, in principle, it is possible to limit these disadvantages with the aid of regulation, in
practice this is far from simple. If vertical integration is undesirable from a welfare perspective
it may therefore be decided to divide the integrated company. In a vertical division of network
sectors a separation is made between the network on the one hand and the services supplied by
the network on the other hand. This section will first discuss the effects of vertical integration,
after which the next section will look at vertical integration from a welfare perspective.

4.A.2.1 Advantages of vertical division

Vertical division of integrated network monopolies can ensure that:

• the network becomes more accessible;

• the supervisory authority is able to obtain more and better information;

• cross-subsidization is no longer possible; and

• it is possible to benefit from ’dis-economies of scope’.

The first advantage of vertical division is that the network becomes more accessible. When the
network is separated from the company activities that can (in principle) be offered in competition,
the incentive to limit or complicate access to the network is reduced. Consider, in this context,
both the use of price-based methods (increasing the network access price for rivals) and the use
of non price-based methods (‘exclusion’). Especially in the case of asymmetric regulation, in
which the access prices are regulated differently than the end product prices, ‘exclusion’ may be
an attractive option for the incumbent. Vertical division means that competition in the different
sub-markets is likely to increase.
4.A. Integration versus division 77

The second advantage of vertical division relates to the information that is available to
the supervisory authority (monitoring). After the division of a vertically integrated network
monopoly it becomes easier to obtain reliable information about the network to be regulated. This
is because, once the network has been fully separated, the company in question no longer has any
options and incentives to internally reallocate the costs and profits for strategic considerations.
After division it will generally also be easier for the supervisory authority to structure certain
types of regulation in an efficient and effective manner. For instance, there is no longer any
problem with the general costs, which are difficult to allocate to individual operational activities.
The third advantage of vertical division is the fact that it prevents companies using cross-
subsidization from disrupting competition. In this context, consider both accounting and
economic cross-subsidization. Accounting for cross-subsidization refers to a situation in which
costs are moved between the regulated and the competitive activities. This may have a
detrimental effect on competition in the competitive market. Economic cross-subsidization
means that the company deploys its best production resources (for instance managers) in the
competitive sub-market. This means operations will not be optimal, resulting in welfare losses.
The fourth example of vertical division occurs when there are so-called ‘dis-economies of
scope’. In this case, separation of the various business units will increase the value of the resulting
companies. The reason why integration did take place in the past may lie in the advantages
offered by increased possibilities for anti-competitive behavior. This may enable the company to
increase its profits, despite the fact that vertical integration is accompanied by a loss in efficiency.
It is quite possible that the market power vertically integrated companies have in comparison
with their specialist competitors is more likely to be an important reason for integration than a
more or less random outcome thereof. Another reason why integration did take place in the
past is the fact that, with the technology that was available at that time, there were advantages
associated with integration.
To realize the advantages associated with vertical division it is important that sufficient
competition is created in the different markets. Although division ensures that the supervisory
authority can focus its attention on the part of the market that may be typified as a natural
monopoly, in itself this is not sufficient. If there is insufficient competition in the other sub-
markets, the positive effects of division for consumers will remain limited. Indeed, if the level of
competition does not exceed the effectiveness of the regulation that replaces it, it is possible
that, in the final balance, division has no positive effects at all. Obviously, in the case of perfect
regulation there are, in principle, no direct advantages associated with vertical division, because
the supervisory authority already has all the required information and instruments to achieve
the desired result.
However, in practice there will (nearly) always be imperfect regulation. If regulation is
imperfect, the costs associated with this form of government failure will, logically speaking, be
higher in the case of vertical integration. After all, compared to a situation of vertical division,
the integrated company has considerably more incentive for anti-competitive behavior.
78 Chapter 4. Network Sectors: Privatisation and (de)regulation

4.A.2.2 Disadvantages of vertical division

In addition to the aforementioned advantages, vertical division may also have some disadvantages,
as vertical division may result in:

• A loss of ‘economies of scope’ for a company;

• considerable costs; and

• problems relating to the lines along which division needs to take place.

The first disadvantage is the fact that, for the company in question, vertical division may result
in a loss of ‘economies of scope’. Although, in principle, it is possible to limit this loss with the
aid of detailed contracts, in practice this will be unlikely as, because of the dynamic environment,
entering into complete contracts is extremely expensive or even impossible.
The second disadvantage is the fact that the division of a vertically integrated company may
entail considerable costs. Although these are one-off costs, they do constitute an important part
of the cost/benefit analysis relating to the division of vertically integrated network companies.
‘Division costs’ also include the costs of establishing new institutions for the regulation of the
company or for the promotion of information exchange between different company units.
The third disadvantage is the fact that there is a range of problems associated with determining
the lines along which division must take place. For instance, if the supervisory authority has
imperfect information, companies have different ways of trying to influence the division to their
advantage. These lobbying activities may be advantageous to the companies in question but not
necessarily to the economy as a whole. In addition, the fact that companies are prohibited from
operating in certain segments of the market may result in these companies investing in relatively
inefficient technologies which will enable them to produce a permitted substitute. For instance,
if a telecom company is prohibited from supplying broadcast services with the aid of its network,
this could result in the telecom company in question looking to supply related services that are
permitted. It is also possible that the lines along which the division originally took place are no
longer realistic as a result of technological developments, as technological developments may
result in a natural monopoly ceasing to be a monopoly over the course of time. This would
mean that the external structure of a sector would have to be redetermined, which would involve
additional costs.

4.A.3 Effects on social welfare

In view of the above, the question when division of vertically integrated companies is desirable
and when it is not, is not easy to answer. Economic literature indicates that vertical integration
can have both a welfare-increasing and a welfare-reducing effect, but it is not yet clear what the
underlying fundamental factors are. On the subject of the choice between vertical integration
and vertical division Newbery (1999) says:
4.B. Natural monopolies 79

Each has it advantages and drawbacks. Vertical integration facilitates coordination

and may economize on transaction costs, but ensuring equal access may increase
regulatory costs and create other inefficiencies. Separation may avoid the regulatory
problem of ensuring equality of access but creates other regulatory problems in the
design of access pricing to retain coordination benefits.
Thus, each case requires a comprehensive analysis of the advantages and disadvantages associated
with division. See for such an analysis, e.g., Section 4.6 for an analysis of a vertical division in
the energy sector in the Netherlands.

4.B Natural monopolies

In this chapter we often speak about ‘natural monopolies’, a term which frequently occurs in the
media as well. Throughout the chapter, the central economic argument why we should regulate
an industry at all, is that the industry is characterized by a ‘natural monopoly’. However, what
is exactly a ‘natural monopoly’ and what does its occurrence mean for economic theory and
the organization of firms in network sectors? A formal definition of a natural monopoly could
be formulated as: “A natural monopoly means that the technology of certain industries or the
character of the service is such that the customer can be served at least costs or greatest net
benefit only by a single firm”.
Though this seems a rather straightforward definition at first sight, there are some difficulties
with the concept of natural monopolies, which we would like to elaborate on in this appendix.
First we discuss the ‘classical’ view on natural monopolies. Thereafter, we look at more
recent theories. We conclude with a small discussion concerning the difference between natural
monopolies and ‘normal’ monopolies.

4.B.1 The ‘traditional’ view on natural monopolies

As already mentioned, the question whether we can speak about a natural monopoly usually
hinges around the concept of constant returns to scale (or better: increasing returns to scale).
In its most basic form this means that average costs fall when market outputs increase. This
may happen when fixed costs are relatively very large—e.g., for the construction of a physical
network such as the sewer system. Figure 4.1 shows what happens.
In Figure 4.1 there is an inverse demand function (price as a function of the output) and a
continuously decreasing average cost function. To remain economically viable, the firm needs to
set a price which is at least as high as the average cost. If we now assume that there are no entry
and exit costs, then, if a firm sets a price pe higher than pY , a new firm enters the market setting
a price pb (pY < pb < pe) and drives the first firm out of the market while remaining economically
viable itself. Thus, the only production level that precludes entry from another firm would be Y ,
with corresponding price pY . In the traditional view the market is then said to be characterized
by a natural monopoly because competition within the market is not possible.
 has fixed costs (marketing costs, R&D costs, etc.). However, creating a (phys
network (pipelines, radio transmission masts, rail network) involves much larger f
costs. The following (simplified) figure show that depending on the fixed costs
thus average costs, one, two or more firms may be viable, given the dem
80 function.
Chapter 4. Network Sectors: Privatisation and (de)regulation

Price Price

Demand Demand

Average costs
AC1 AC2

Marginal costs Marginal costs

Y Output Y Outpu

Figure 1: One or multiple firms in the case of decreasing average costs.

Figure 4.1: The classical natural monopoly problem. Source: Braeutigam (1989).

But why bother, one would ask, because pY is set where the demand function intersects with
the average cost function, which happens to be the welfare optimum as well (Chapter 3). The
problem is that entry and exit of a firm is never costless and that there is a time dimension as
well. Firms could, e.g., enter a market, charge a price higher than average costs to cash in on
their market power and threaten to reduce price to a very low level (even less than average costs)
in the short run when other firms should attempt to enter the market. Thus, from a traditional
perspective, natural monopolies are based on the existence of economies of scale throughout the
relevant range of production on the market. These economies of scale might then lead to greatly
inefficient and even wildly fluctuating, unstable price, so that some sort of regulation is needed.

4.B.2 A ‘contemporary’ view on natural monopolies

The model of above received an increasing amount of criticism, centered around the multi-product
nature of firms operating in sectors with natural monopolies. First, it is insightful to see that
the notion of natural monopolies does not hinge upon economies of scale throughout the whole
range of production in the market. Consider the first diagram in Figure 4.2.
In Figure 4.2 there are no global economies of scale present, only on a certain (local) subrange
of the production output. If the firms average costs level decreases up to the production level Y ,
it is clear again that a single supplier could again serve the market at a lower unit cost than an
industry configuration with two or more firms. There is another difference with the previous
subsection and that is the term sub-additivity. Now, we speak of natural monopoly when a firm
can produce the whole range of output Y at the lowest possible costs. Note that sub-additivity
may as well refer to the production of a bundle of goods, instead of only one good.
As an example, consider the second diagram in Figure 4.2. In comparison with the first
diagram, the demand function is only slightly shifted to the right, meaning that for each price
demand has increased. In this situation we still encounter a natural monopoly (with price
pZ and output Z), although the firm does not produce anymore at lowest possible average
4.B. Natural monopolies 81

Price
Price Average
Average costs
costs Price
Price D_old
D_old Average
Average costs
costs

PZPZ
PYPY
PYPY D_new
D_new
Demand
Demand

YY Output
Output Y YZ Z Output
Output
Figure
Figure 2a2aand
Price and2b:
2b:A ANatural
(a) Natural Monopoly
Monopoly with
Averagewith increasing
increasing
costs demand.
demand.
Price D_old
(b) Average costs

PriceSubadditivity without (a) global economies of scale and with (b) a shifting demand
FigurePrice
4.2:
function. Source: Braeutigam (1989), Aalders et al. (2002). PZ
PY
PY D_new
costs. If the firm decides to produceDemand
at lowest possibleAverage
costs Z − Y production will not be
Ycosts
Average , costs
D_old
D_old D_new
D_new
produced—although being profitable. A new firm cannot produce only Z − Y , because the
high fixed costs prevent the new firm to be profitable. This means that total production costs
Y Output Y Z Output
of the sumPYPYof Y and Z − Y are higher than the production costs of Z, and thus we speak of
sub-additivity.Figure 2a and 2b: A Natural Monopoly
Y
with increasing demand.
Output
Y Output
Figure
Figure3:3:AAnatural
naturalduopoly.
duopoly.
Price

Figure
Figure2 2and
and3 3show
showthe
theoccurrence
occurrence of of
natural
naturalmonopolies
monopoliesof of
duopolies in the
duopolies case
in the of of
case
U-shaped
U-shapedaverage
averagecosts.
costs.InInfigure
figure2a2ait is clear
it is that
clear only
that one
only onefirm cancan
firm produce total
produce total
market
marketdemand
demandatatthe thelowest
D_old possible
lowest possiblecosts. In In
costs.
D_new figure 2b,2b,
figure thethe
demand
Average costs function
demand function is is
slightly
slightlyshifted
shiftedtotothe
theright.
right.There
Thereis isstill aa
still natural
natural monopoly
monopolynow, although
now, thethe
although firmfirm
can
cannot
notproduce
produceoutput
outputZ Zatatthe
thelowest
lowestcosts anymore.
costs anymore.It can of of
It can course produce
course produce
PY
output Y, but then a second firm has to produce output Z – Y at very
output Y, but then a second firm has to produce output Z – Y at very high costs. high costs.
Thus, Y is always
Outputlarger than the production
Thus,the
theproduction
productioncosts
costsofof
output
outputYYand Z–
and ZY– Y is always larger than the production
costs of Figure
Z. In 3: A natural
figure 3, the duopoly.function is doubled compared to figure 2a. In this
demand
costs of Z. In figure 3, the demand function is doubled compared to figure 2a. In this
situation we can speak Aofnatural
Figure
situation we can4.3:
speak ofa anatural
naturalduopoly.
duopoly. Two Aalders
Source:
duopoly. Two firms can
firms et produce
al.
can (2002).together
produce the
together the
FigureY2atand
market output 3 show
lowest the occurrence
possible costs. of natural monopolies of duopolies in the case of
market output Y at lowest possible costs.
U-shaped
As a final example, average
consider costs.
Figure 4.3.InHere,
figuredemand
2a it is isclear that only
increased one firm
to twice can produce
its original size, total
leading to question
The a case
marketofnow
a demand
natural duopoly:
is of course two firms
at thewhether
lowest canmonopolies
normal fulfill
costs.market
In showdemand
the sameat lowest possible
behavior
The question now is of course whetherpossible
normal monopolies figure
show2b,the the
samedemand function is
behavior
production costs.
as ‘natural’ In this case,
monopolies. natural duopoly would be optimal from a welfare perspective. It
slightly
as ‘natural’ shifted toThe
monopolies. theanswer
The right.
answer
is no. is
There Namely,
still a from
is no. Namely,
a welfare point
natural
from amonopoly
of view,
now,
welfare point
it
ofalthough
view, it the firm
is now easy to see that if market demand increases, market power of firms will decrease, and
may be can
desirable to maintain
not produce a ‘natural’
output monopoly.
Z at the lowest Breaking
costs up a ‘natural’
anymore. It canmonopoly
ofmonopoly
course produce
may be desirable
that, eventually, the marketto maintain a ‘natural’
form converges monopoly.
to the Breaking
case of perfect up a ‘natural’
competition.
output Y, but then a second firm has to produce output Z – Y at very high costs.
Thus, the production costs of output Y and Z – Y is always larger than the production
costs of Z. In figure 3, the demand function is doubled compared to figure 2a. In this
situation we can speak of a natural duopoly. Two firms can produce together the
market output Y at lowest possible costs.
82 Chapter 4. Network Sectors: Privatisation and (de)regulation

Thus,—in the case of only one good—there is a natural monopoly when there is sub-additivity.
Or, when there are economies of scale for all levels of production there is a natural monopoly.
The latter means simply that for all levels of production average costs fall when output increases.
In the case of two or more goods, unfortunately, it is not that simple. Consider two goods. If
there are economies of scale for both goods for all production levels but negative synergy effects
as well between those two goods, then it might well be that there is no natural monopoly, if
the negative synergy effects are large enough.12 To make things even worse, it is sometimes
very hard to discern whether something is one product or two (or even more). Take ‘energy’
for example. Energy is more expensive during peak hours and cheaper in the evenings and
weekends, causing different demand functions. Basically, this means that we can speak of two
goods, although both goods are usually aggregated for the sake of convenience.

4.B.3 Natural monopolies versus normal monopolies

The question is now in what way a natural monopoly exactly differs from a ‘normal’ monopoly.
The answer can be found in the concept of social welfare. If an industry or sector is characterized
by a natural monopoly, then—from a welfare perspective—it is desirable to have only one firm
on the market, because it produces at the lowest cost. When there is a normal monopoly that
would never be the case. Recall the discussion in Chapter 3. Social welfare will always be
suboptimal when there is a normal monopoly. Therefore, natural monopolies will be welfare
optimizing, and will usually be the desirable market form. However, as we have seen above, firms
may abuse its monopoly position, because there are entry and exit costs and there is a timing
dimension. Thus, governments have a large incentive to regulate natural monopolies—apart
from all kinds of normative and political reasons, such as redistribution of social welfare.

12
Those negative synergy effects may, e.g., be interpreted as internal competition within a firm.
Chapter 5

The Transport Sector: Roads

5.1 Introduction

“I will begin with the proposition that in no other major area are pricing practices so irrational, so
out of date, and so conducive to waste as in urban transportation. Two aspects are particularly
deficient: the absence of adequate peak-off differentials and the gross underpricing of some
modes relative to others.” These are not the recent words of a critic of contemporary Dutch
transport policies. Instead, these words were written in 1963, by a later Nobel Prize winner in
Economics, William Vickrey. He continued: “In nearly all other operations characterized by
peak load problems, at least some attempt is made to differentiate between the rates charged for
peak and for off-peak service. Where competition exists, this pattern is enforced by competition:
resort hotels have off-season rates; theaters charge more on weekends and less for matinées.
Telephone calls are cheaper at night... But in transportation, such differentiation, as it exists, is
usually perverse.”

Vickrey’s views on traffic congestion, and his plea for corrective prices to deal with it, are
shared by many economists but not by many road users. Therefore, this chapter focuses on this
particular type of transport network, for which use yet nobody is paying for; the road network.
Specifically, it will explain how a congested road can be analyzed from an economic perspective
as a—daily recurring—example of a failing market subject to ‘external costs’. The free market
outcome on such a market will be discussed and compared with the economic optimum. We
will next discuss how this optimum can be realized through the use of pricing policies, and will
compare the efficiency of such policies to that of non-pricing policies. The chapter will conclude
with a discussion of the important, somewhat paradoxical question of why economically optimal
(welfare maximizing) policies may often be so hard to implement in democratic societies. We
will begin, however, by presenting some relevant facts and figures on traffic congestion in The
Netherlands.

83
84 Chapter 5. The Transport Sector: Roads

5.2 Road traffic congestion in The Netherlands: some facts and

figures
Road traffic congestion is an important topic in contemporary transport policy debates, both
in The Netherlands and abroad. Before going into the details of the economic approach to the
analysis of traffic congestion, let us look at some background figures on traffic congestion in The
Netherlands, in order to get an idea of the size and dimensions of the problem at hand.

5.2.1 Development over time

The statement that traffic congestion has become worse over the recent past will probably not
come as a big surprise. Figure 5.1 shows some relevant trends and predictions. The relative
growth in vehicle-hours lost on Dutch highways (measured as time losses incurred because the
actual speed is below the speed limit) by far exceeds that in vehicle-kilometers traveled, which
in turn exceeds the growth in highway capacity (measured in total lane-kilometers).

250

200
Index (1990 = 100)

150

100

50

0
1975 1980 1985 1990 1995 2000 2005 2010 2015
Year

Highway capacity (lane-km's) Veh.-km's (highways, working days) Veh.-hours lost

Figure 5.1: Development in congestion on Dutch highways over time (passenger cars, 1980-2010).
Source: AVV (2003)

The data clearly suggest that the severity of congestion on Dutch highways is rapidly increasing,
and it is predicted to increase further in the years to come. This trend may be somewhat inflated,
as monitoring techniques have improved over the years, and a more complete coverage is realized
in the measurements. Nevertheless, an increasing trend is undeniable.
The trend alone does of course not yet tell us whether the overall size of the congestion
problem is significant. For instance, the total number of vehicle-hours lost in the year 2000 was
47 million. That sounds like a large problem, but is it? When dividing this figure by the number
5.2. Road traffic congestion in The Netherlands: some facts and figures 85

of inhabitants in The Netherlands, and assuming that there is on average only one person in a
car, we find an average of 3 hours lost per person per year. This means, on average, less than
half a minute per day for the average Dutch citizen. In other words, provided the instructions of
dentists are carefully followed, the amount of time spent brushing teeth in The Netherlands is
a multiple of the amount of time lost in highway congestion—and nobody complains about it.
Then why should we worry over traffic congestion—or shouldn’t we?
There are good economic reasons to worry about traffic congestion (and less so about teeth
brushing). First, in later sections we will talk about the market failure surrounding traffic
congestion, which provides the fundamental economic motivation for such worries. Next, the
valuation of time losses and travel time uncertainty is for most road users (which include
freight transport and business travelers) relatively high during peak road traveling. As a result,
the overall economic costs of congestion may be substantial1 . A further consideration is that
congestion on roads other than highways is not included in the data in Figure 5.1, so that
the overall scale of the problem is in fact larger than what appears from those data. More
importantly, however: unlike teeth-brushing, congestion is typically—and nearly by definition—
strongly concentrated. The time losses for those affected, therefore, may by far exceed the
average of less than a minute per day. Daily time losses of an hour or more (for return trips) are
no exception for such users, and the implied congestion costs are significant.

5.2.2 Concentration in space

One type of concentration of traffic congestion concerns the spatial dimension. Figure 5.2 shows
the distribution of the aggregate data in Figure 5.2 over the Dutch highway network. Clearly, by
far, most congestion occurs in and around the Randstad area, and in particular near its bigger
cities.
An implication of this spatial pattern is that policies aimed at coping with congestion should
preferably allow a differentiated use over space. This would make instruments such as fuel taxes
or annual licence fees less attractive candidates.

5.2.3 Concentration in time

Traffic congestion is of course also concentrated in time. Illustrative data are given in Figure 5.3,
which shows for the various days of the week the number of traffic jams by starting time-of-day,
over a full year. The general patterns found will not be surprising: one can easily identify the
morning and evening peak periods on working days, as well as relatively mild congestion levels
during weekends. The extent of the differentiation, however, is noteworthy.
This time pattern implies that congestion policies should ideally also allow for differentiation
over time, apart from the required differentiation over space. This is one of the potential pitfalls
1
In 1997, the economic damage due to traffic congestion on national highways was estimated to be in the order
of 1.7 billion guilders for The Netherlands. Given the increase in congestion since then, a value of ¤1 billion may
now be approximated.
86 Chapter 5. The Transport Sector: Roads

Figure 5.2: Severity of traffic jams (kilometer-minutes of traffic jams) on Dutch highways (2001).
Source: AVV (2002)

of road capacity expansion: the—often significant—investments will yield benefits only over
relatively short periods, as the extra capacity may remain idle for most of the time.

5.2.4 Summary

Traffic congestion is one of the greatest challenges faced by transport authorities in most modern
societies. Although the traffic congestion experienced by the average citizen in a country like
The Netherlands need not be alarming, it is particularly the concentration in space and time
that makes it a pressing problem. These characteristics at the same time pose high demands on
policy instruments for coping with congestion, since these should preferably allow a differentiated
use over place and time.
5.3. An economic market approach to the analysis of traffic congestion 87

Figure 5.3: Number of traffic jams on Dutch National highways by starting time (2001). Source:
AVV (2002)

5.3 An economic market approach to the analysis of traffic con-

gestion

Long before traffic congestion reached the levels we observe today, it had already attracted
attention from economists. Already back in 1920, Pigou applied microeconomic theory to
the analysis of traffic congestion, and laid the foundations for what has become the standard
economic model of road traffic congestion. His writings have, however, also attracted much
attention outside the field of transport economics, as the approach and conclusions are applicable
to ‘external costs’ in general. In brief, Pigou showed that when an external cost exists, a free
market will not attain an economically efficient outcome, and corrective taxation by a government
is called for to restore efficiency. The tax should reflect the economic value of the external cost at
the margin, and such taxes are now commonly referred to as ‘Pigouvian taxation’ in economics.
To understand the importance of this result, it is useful to recall that an external effect exists
when one actor imposes unpriced welfare effects on at least one other actor, as a by-product
of otherwise legitimate economic consumptive or productive behavior (crime and violence are
therefore usually not considered as external costs).2 Pigou’s analysis is thus relevant for issues
such as environmental pollution, noise annoyance, and smell, apart from the immediate relevance
for congestion policies.
2
In general, markets with networks where actors are highly interrelated exhibit usually large (and many)
external effects. Moreover, those external effects which are directly caused by network are usually called network
effects.
88 Chapter 5. The Transport Sector: Roads

In this chapter, however, we are primarily interested in the implications of Pigou’s analysis
for the economic regulation of traffic congestion. We will derive Pigou’s result step by step,
by presenting an economic analysis of traffic congestion for the simplest possible case. This
means that we will consider a single road, connecting a single origin to a single destination, with
nearly identical users who will only differ in terms of their willingness to pay for using the road.
There is one user per vehicle, and the terms ‘user’, ‘vehicle’ and ‘trip’ are used interchangeably.
We abstain from the often complex dynamics of traffic congestion, and will thus only consider
‘stationary state’ congestion, meaning that traffic speed and the intensity of use of the road are
constant over time, and along the road considered. We will furthermore consider a peak period
of a given duration.

5.3.1 Demand side

5.3.1.1 The ‘price’ for the use of a road: the concept of generalized prices

As for most economic goods, it is reasonable to assume that the demand for trips on a road will
increase if the price falls. In this respect, the economic approach to modeling road use is not
different from the modeling of any other market. What is somewhat different, however, is the
definition of the price. Normally, one only considers the monetary price of a product; say, the
quoted sales price in Euros in a shop. For trips on a road, this would be a somewhat unnatural
interpretation of a price: trips are not bought in a store, but are instead ‘produced’ by the
consumers themselves, through the purchase (or rental) of a vehicle, the expenses made for its
maintenance and insurance, the purchase of fuel, and the investment of valuable time necessary
to make the trip. Some of these cost elements are given in the short run, or independent of the
decision of whether or not to make a certain trip. Others are variable, and can thus be saved
when a trip is suppressed. This variable part of the costs will be referred to as the generalized
price (denoted p) for trips on the road that we will consider. This generalized price includes both
monetary expenses, for example on fuel, and the time required to make a trip. The adjective
‘generalized’ is thus used to remind us of the somewhat different interpretation of price in the
context of road use, compared to other markets.

5.3.1.2 Value of time

The different components of the generalized price cannot simply be added. For instance, fuel
expenses will be measured in Euros, and travel time in minutes (or hours). In order to make
money and time comparable, the so called value of time has to be determined.
The ‘value of time’ is the economic counterpart of the proverb ‘Time is money’. Time losses
due to congestion have a definite value for the majority of road users; otherwise, congestion would
not be a problem. The value of time aims to reflect the average amount of money that a typical
road user is willing to pay to avoid time losses while traveling. Empirical estimates of values of
time are sometimes obtained by studying actual behavior of individuals when they can make
5.3. An economic market approach to the analysis of traffic congestion 89

trade-offs between time and money (so-called ‘revealed preference’ studies). Alternatively, one
could analyze hypothetical behavior, for instance when confronting respondents of a questionnaire
with the same type of trade-off (so-called ‘stated preference’ studies). As shown in Table 5.1,
the resulting empirical estimates of the value of time may vary significantly over different types
of road users. Equipped with appropriate value-of-time estimates, the time required for a trip

Table 5.1: Estimated values of time in road use in The Netherlands in ¤/hr (1997). Source AVV
(1998)

Trip motive Gross monthly household income Average

Commuting Business Other 1365 1365–2275 2275–3410 > 3410

Value of time 6.59 22.81 4.55 4.23 5.09 6.27 11.55 7.36

can be translated into a monetary equivalent simply by multiplying the travel time by the value
of time. The result can then be added to other cost components, such as fuel expenses, in order
to calculate the generalized price of a trip, expressed in money terms.

5.3.1.3 Representing the demand for trips

The demand function for the use of our single road is now simply given by the relation between
the generalized price p (e.g., in ¤’s) and the number of trips demanded, N , during the peak
period considered. It is customary in economics to use the inverse of this function in graphical—
and mathematical—expositions. So, instead of showing N on the vertical axis, as a function of
p on the horizontal axis, the axes are interchanged. The resulting function is called the inverse
demand function, which will be denoted D in what follows.
Figure 5.4 shows an example. It is the inverse demand function that we will be using in the
numerical example used in this (and the next) chapter. It has a simple linear form:

D = d0 − d1 N (5.1)

where d0 and d1 are parameters denoting the intercept and slope of the inverse demand function,
respectively with current values for the numerical example of d0 = 100 and d1 = 1.
The inverse demand function of course conveys exactly the same information as an ‘ordinary’
demand function that shows N as a function of p. For example, the equivalent ordinary demand
function corresponding with the inverse demand function D in equation (5.1) is:

d0 1
N= − p (5.2)
d1 d1
implying N = 100 − p for the numerical example.
90 Chapter 5. The Transport Sector: Roads

p (!)

100

80 b

60
D = mb
40 a
20

N
10 20 30 40 50 60 70

Figure 5.4: The inverse demand (D) or marginal benefit (mb) function for the numerical example

5.3.1.4 The inverse demand functions and marginal benefits

An important interpretation of the inverse demand function is that it shows, for every possible
quantity, the marginal willingness to pay: the maximum amount of money that the consumer
of the last unit consumed is prepared to pay for consuming that unit. That is: if an ‘ordinary’
downward-sloping demand function tells you that the quantity demanded would be 30 at a
price of ¤70 (as in our example), it also tells you that the maximum willingness to pay for the
30th unit sold is exactly ¤70. Why should that be the case? Well, even a slight price increase
would leave the 30th unit unsold, from which we can conclude that the consumer of that unit is
apparently not prepared to pay more than ¤70. She is, however, prepared to pay ¤70; otherwise,
the 30th unit would not be sold at a price of ¤70. And she will of course typically also be willing
to buy that unit when the price falls below ¤70.

This interpretation of an inverse demand function provides an important link between demand
analysis and welfare measurement. A popular way of measuring the benefits from consumption
in applied economic research is namely based on the idea that the benefits attached to the
consumption of a certain unit, expressed in money terms, can be equated to the maximum
willingness to pay for that unit. In the example just given, the benefits of consuming the 30th
unit must amount to ¤70 when expressed in money terms. If the benefits were lower, the
consumer would not be prepared to pay ¤70 for that unit. If the benefits were higher, she would
also buy it when the price would exceed ¤70. A consequence is that the inverse demand function
shows the marginal benefits (mb) of consumption: at every quantity, it shows the benefits from
the consumption of the last unit consumed.
5.3. An economic market approach to the analysis of traffic congestion 91

5.3.1.5 From marginal benefits total benefits, consumer surplus, and social surplus

An important next implication is that the total benefits of consumption at a certain equilibrium
level of consumption are given by the area under the inverse demand function, bounded by the
two axes and a vertical line at the equilibrium quantity consumed. This area simply sums the
benefits of consumption for each of the individual units sold in equilibrium. In Figure 5.4, this
area is given by the sum of the rectangle a and the triangle b, for the case where the equilibrium
price would be ¤70, and the consumed quantity is 30 units.
Mathematically, this link between the inverse demand function and the total benefits (B)
from consumption is expressed as follows:
Z N
B(N ) = D(n)dn (5.3)
0

The term on the left-hand side denotes the total benefits B as a function of the total amount
R
consumed, N . The term on the right-hand side shows the so-called integral (the -sign denotes
the integral operator) of the demand function, between consumption levels of 0 and N . The
integral operator tells you to compute the size of the area under the function following the
operator, between the lower and upper limits shown below and above the operator.
Integration of a function is in fact the reverse procedure of the differentiation of a function.
So, if we want to find the B function consistent with the D function in equation (5.1), we ask
the following question: which function has equation (5.1) as its derivative? For a linear function,
this problem is easily solved, and we find:3

1
B = d0 N − d1 N 2 (5.4)
2
Equation (5.4) can be checked in two ways. The first is by taking its derivative with respect to
N , and checking that the result indeed reproduces equation (5.1). The second way involves a
geometrical approach. The reader may check that the area under D between 0 and any N in
Figure 5.4 can be determined as the rectangle of a size N d0 , minus a triangle that has a size of
1
2 N (d0 − (d0 − d1 N ). This immediately leads to equation (5.4).
The two components of the total benefit in 5.4, a and b, both have their own interpretation.
Rectangle a shows the total expenses: price times quantity (where the price in our road example
is the generalized price, and includes the value of time traveled). The difference between
total benefits B and these total ‘generalized expenses’, triangle b in the diagram, gives the
consumer surplus (CS). This aggregates, over all units consumed, the difference between what
the consumers would be willing to pay for each unit, and what they actually have to pay in
terms of generalized prices. Figure 5.4 thus shows that:
3
The derivation ignores one aspect of integration, and that is that any constant can be added to (5.4) and
leave the mathematical operation of integration correct (note that the constant would vanish when differentiating
the function again, so that equation (5.1) would be found as the derivative irrespective of the value of the constant
chosen). In the present example, the interpretation of this constant would be: the total benefits associated with
zero road use. A natural choice for the value of this constant is therefore zero.
92 Chapter 5. The Transport Sector: Roads

consumer surplus (CS) = total benefits (B) − total generalized expenses (N p)

The consumer surplus CS is closely related to the social surplus S, which is an important
indicator for welfare in applied economic research. The social surplus can be defined as the total
benefits from consumption, minus the total costs:

social surplus (S) = total benefits (B) − total costs (C)

The two surpluses are therefore identical when the total generalized expenses by consumers
are equal to the total social costs. In our analysis, this will be the case unless the government
imposes a tax on trips. A tax would add to the expenses of consumers, but not to the total
social costs. It is merely a transfer: each euro of taxes paid lowers the consumer surplus, but
increases the government surplus (GS) by the same amount. The irrelevance of tax payments
for the calculation of the social surplus in our analysis can alternatively be illustrated by noting
that the social surplus should include all surpluses in the economy, and is therefore also equal to
the sum of consumer surplus and government surplus:

social surplus (S) = consumer surplus (CS) + government surplus (GS)

The two definitions of social surplus of course differ in interpretation, but should give the
same answer when calculated for a specific situation.4

5.3.1.6 Summary

As for any market, we can analyze the demand for trips on a road by the use of demand functions.
An important aspect is that ‘the price’ now also includes non-monetized components, such as
the travel time required. By multiplying the travel time by the value of time, we can express this
travel time in a monetary equivalent, and add it to other monetary costs such as those for the
purchase of fuel, to obtain the generalized price for using the road. The inverse demand function
for the use of a road reflects the marginal benefits of consuming trips on that road, and the
area under the inverse demand function thus corresponds with the total benefits enjoyed from
using that road. The difference between these total benefits, and the total ‘generalized’ expenses,
gives the consumer surplus associated with the use of the road. A related measure is the social
surplus, defined as the difference between total benefits and total social costs; or, equivalently,
as the sum of consumer surplus and government surplus. In our analysis, the consumer surplus
is equal to the social surplus in absence of taxes. Otherwise, the two diverge, because a ¤of
taxes paid does not constitute a real social cost, but merely a transfer.

4
Note that if we rewrite the second definition of social surplus as: S − CS = GS, and substitute the first two
definitions, we find that N p − C = GS. The left-hand side of this latter equation gives total taxes paid.
5.3. An economic market approach to the analysis of traffic congestion 93

5.3.2 Supply side

5.3.2.1 A typopology of cost functions

The discussion of the generalized price above facilitates our exposition of the supply-side of the
market for trips on a road. The supply relation should tell us how many trips can be made
against which generalized price on the road. On an uncongested road, and with identical road
users driving identical cars, this generalized price would be constant: every trip requires the
same amount of travel time, fuel, etc. With congestion, however, things become different. As
more cars enter the road, drivers will slow down for safety reasons, implying that the travel
time increases.5 Because all road users have the same speed in equilibrium, this means that the
average cost of a trip, ac, will rise with the number of trips made. Without tolls, this average
cost ac equals the generalized price p. It is, however, useful to define ac separately from p,
because the two will diverge when tolls are introduced in a later stage of the analysis.
The type of relationship between the number of trips and the average cost has been subject
to much empirical analysis. It is typically found that for low use levels, the ac curve is nearly
flat: when there are few users on the road, an additional user will hardly, or not at all, affect
the speed on the road. However, if road use intensifies, the ac curve becomes increasingly steep,
and when the maximum capacity of the road is approached, it may become nearly vertical. To
keep our numerical example simple, we will ignore this non-linearity, as it does not affect the
qualitative conclusions of our analysis. We will assume in our example that the average cost
function has the following linear shape:

ac = c0 + c1 N (5.5)

where c0 and c1 are parameters denoting the intercept and slope of the average cost function,
respectively with current values for the numerical example of c0 = 10 and c1 = 1. This fuction is
shown in Figure 5.5 below:
It is easy to derive the total social cost, C, from the average cost: we simply multiply the
cost per trip, ac, by the number of rips, N . For our numerical example, this yields:

C = c0 N + c1 N 2 (5.6)

The final type of cost function of interest is the marginal cost: the increase in total cost following
from the addition of one user to the road. This can be found by taking the derivative of C with
respect to N :

mc = c0 + 2c1 N (5.7)
5
Fuel costs may of course also rise, because more time is needed to complete a trip. We will ignore this latter
aspect, and will focus on the effect on travel time in the remainder. Likewise, we will ignore other external costs
that may arise from road trips, such as caused by environmental pollution or noise annoyance.
94 Chapter 5. The Transport Sector: Roads

100

80
ac
60
p0
40 D = mb

20

N
10 20 30 40 50 60 70
N0

Figure 5.5: The free-market equilibrium

This function will appear in Figure 5.6 below.

Independent of the exact form of the cost functions, the following relation will hold between
the three cost measures just defined:
Z N
C(N ) = N × ac(N ) = mc(n)dn (5.8)
0

The total cost of having N users is of course equal to the number of users times the average cost,
and also to the integral of the marginal cost function from 0 to N . As an exercise, the reader
can check these equalities for equations (5.5)–(5.7). For the same reason that total benefits
can be determined as the area under the marginal benefit function, total costs can therefore be
determined as the area under the marginal cost function.
A comparison between ac in (5.5) and mc in (5.7) reveals an important result. And that
is that the marginal cost exceeds the average cost when the number of trips exceeds zero: the
difference is c1 N . This phenomenon can be explained in a number of ways.
A first, technical explanation would be that the average cost can only be rising (as it is,
according to equation (5.5)) if the marginal cost exceeds this average cost. Compare it with the
grades you get on your exams. There is only one way of raising your average grade, and that is
to obtain a higher score on your next exam (‘the marginal grade’) than your current average.
The same principle holds for average and marginal costs.
A more intuitive economic explanation is as follows. When a next user is added to a congested
road, two types of cost are created. The first concerns the cost borne by the new user. These
are simply equal to the prevailing average cost on the road. Secondly, however, the travel time
for all other users will increase. And this part of the marginal cost causes these to differ from
5.3. An economic market approach to the analysis of traffic congestion 95

the average cost. With our linear ac function, for each of the existing N users, the average
cost will rise by an amount c1 when adding a new user (this can be seen by differentiating ac
in equation (5.5) with respect to N ). This means that in total, the cost for the existing users
will go up by N × c1 (which is exactly the difference between the average and marginal cost).
Because these costs are not borne by the person who creates them – the last user added – these
costs constitute the marginal external cost (mec) of a trip:

mec = mc − ac = N × ac0 (5.9)

(where a prime (0 ) denotes a derivative), which for our numerical example means:

mec = mc − ac = N × c1 (5.10)

The mec is implicitly shown in Figure 5.6, as the vertical distance between ac and mc. We will
see below what this means for the efficiency of road use.

5.3.2.2 Summary

The average cost of trips on a road rises because speeds fall and travel times rise as more users
enter the road. The marginal cost of trips therefore exceeds the average cost. The difference is
given by the time losses that an additional user imposes upon the other users. These are referred
to as the marginal external costs.

5.3.3 Equilibrium and optimum

5.3.3.1 The free market equilibrium

Having specified the demand and supply side of our model, we are now ready to identify
its equilibrium. Above, Figure 5.5 shows the relevant functions for our numerical example.
Without government intervention, a free-market equilibrium will arise at the intersection of the
inverse demand function D and the average cost function ac. This occurs at N 0 = 45. For a
lower equilibrium number of trips, the willingness to pay for some users would still exceed the
generalized price as given by ac, and these users would find it attractive to enter the road. To
the right of N 0 = 45, the opposite would occur. At N 0 , the generalized price and the average
(per trip) cost are just equal to the willingness to pay for the last user added. As indicated, the
corresponding price, p0 , amounts to 55.

5.3.3.2 The efficient equilibrium

To assess the efficiency of the free-market outcome, we first need to identify the appropriate
indicator for efficiency. This indicator is the social surplus S, defined earlier as total benefits
minus total costs. The Pareto efficient outcome is then that particular level of road use for which
the social surplus is maximized. If it is not maximized (but only then), changes in road use are
96 Chapter 5. The Transport Sector: Roads

apparently still possible for which the induced changes in total costs and benefits are such that
the winners could theoretically more than compensate the losers.
The question is now whether the equilibrium outcome depicted in Figure 5.5 is economically
efficient, as one might expect on the basis of standard expositions of the efficiency of free markets.
The answer is ‘no’. The crucial difference with standard markets is that in the equilibrium in
Figure 5.5, it is not an equality between marginal benefits (mb) and marginal costs (mc) that
is obtained (as usual), but instead one between marginal benefits (mb) and average costs (ac).
And we saw earlier that the marginal cost exceeds the average cost by an amount equal to
the marginal external cost. The level of the marginal external cost (mec) in the free-market
equilibrium, mec(N 0 ), is given by the arrow in Figure 5.6. But this means that at the equilibrium
N 0 , a reduction in the number of trips would save more costs (indicated by the mc curve) than
the amount of benefits foregone (indicated by the mb curve). In other words: the social surplus,
S = B − C, would increase if the number of trips were reduced.

!
mc
100

80 mec (N 0)
G
ac
60

40 D = mb

20

N
10 20 30 40 50 60 70
N* N0

Figure 5.6: The optimal and free-market equilibrium compared

The efficient or optimal level of road use is defined as the point where no increases in social
surplus are possible through a change in the number of trips. This is the point where marginal
benefits equals marginal costs (mb = mc). In our example, this optimal number of trips N ∗
occurs at 30. A further decrease in the number of trips would mean that more benefits are given
up (according to the mb curve) than the total cost savings (according to the mc curve). Moving
rightwards from N ∗ , the extra cost would instead exceed the additional benefits.
The gain in social surplus that can be achieved by reducing the number of trips from N 0
to N ∗ is given by the shaded triangle G. This triangle is found as the difference between the
reduction in total costs (the area under the mc curve between N ∗ and N 0 ) and the reduction in
5.3. An economic market approach to the analysis of traffic congestion 97

total benefits (the area under the mb curve between N ∗ and N 0 ). Table 5.2 summarizes some

Table 5.2: Estimated values of time in road use in The Netherlands in ¤/hr (1997). Source AVV
(1998)

N ac mc mec mb B C S =B−C

Free Market 45 55 100 45 55 3487.5 2475 1012.5

Optimum 30 40 70 30 70 2250 1200 1350

key characteristics of the two equilibria for our example, and implies that the size of triangle G
is 337.5 (the difference between social surplus in the two equilibria). This figure can be verified
by determining the size of the triangle G as 12 × 15 × 45 = 337.5 ( 12 times breadth times height).
A useful exercise is to check all results in Table 5.2.

5.3.3.3 Summary

The free-market equilibrium on a congested road is not efficient. In a free market, an equilibrium
will arise in which marginal benefits are equal to average costs. However, efficiency requires
marginal benefits to be equal to marginal costs. There is a wedge between average costs and
marginal costs, caused by the marginal external costs. The social surplus can therefore be
increased by reducing road use to the point where marginal benefits and marginal costs are
equalized. This point is referred to as the optimal road use.

5.3.4 Summary

We presented an economic analysis of traffic congestion. From an economic perspective, we can

analyze the road as a market, with a demand side, a supply side, and a mechanism that will lead
to an equilibrium between these two. But we have seen that a distortion exists on this market,
that prevents the free market mechanism from leading to an efficient outcome.
As for any market, we can analyze the demand for trips on a road by the use of demand
functions. An important aspect is that ‘the price’ now also includes non-monetized components,
such as the travel time required. By multiplying the travel time by the value of time, we can
express this travel time in a monetary equivalent, and add it to other monetary costs such as
fuel costs, to obtain the generalized price for using the road.
The inverse demand function for the use of a road reflects the marginal benefits of consuming
trips on that road, and the area under the inverse demand function thus corresponds with the
total benefits enjoyed from using that road. The difference between these total benefits, and the
total ‘generalized’ expenses, gives the consumer surplus. A related measure is social surplus,
defined as the difference between total benefits and total social costs; or, equivalently, as the
sum of consumer surplus and government surplus. In our analysis, the consumer surplus equals
98 Chapter 5. The Transport Sector: Roads

social surplus in absence of taxes. Otherwise, the two diverge: a Euro of taxes paid does not
constitute a real social cost, but merely a transfer.
As far as the supply side is concerned, the average cost of trips on a road rises because speeds
fall and travel times rise as more users enter the road. The marginal cost of trips therefore
exceeds the average cost. The difference is given by the time losses that an additional user
imposes on other users. These are referred to as the marginal external costs.
As a result, the free-market equilibrium is not efficient. In a free market, marginal benefits
become equal to average costs. However, efficiency requires marginal benefits to be equal to
marginal costs. There is a wedge between average costs and marginal costs, caused by the
marginal external costs. Social surplus can therefore be increased by reducing road use to the
point where marginal benefits and marginal costs are equalized. This point is referred to as the
optimal road use.

5.4 Policy Implications

It is one thing to recognize the inefficiency of the free-market equilibrium and to identify the
efficient number of trips; it is of course quite another thing to answer the question of how to
actually achieve a move towards that optimum. In terms of Figure 5.5: how could a regulator
convince the users between N 0 and N ∗ not to use the road anymore? Clearly, asking them
politely is not likely to be very effective – we have seen that the benefits of these trips exceed
the generalized price at N 0 , so these users will find it attractive to continue using the road.
In this section, we will approach this question from an economic perspective. In Section 5.4.1,
we will discuss Pigou’s answer, which involves the use of congestion tolls. In Section 5.4.2, we
explain why many economists still think that this is an efficient way of dealing with congestion,
by contrasting optimal pricing to the effects a non-price policy measure.

5.4.1 The optimal congestion toll

Pigou’s answer to the economic problem of traffic congestion appears simple in hindsight, but
was path-breaking at that time. He recognized that by charging a toll equal to the marginal
external cost at the optimum, road users beyond the optimal level of road use would no longer
find it attractive to enter the road.
In our example, mec at N ∗ amounts to 30 (see Table 5.2). Figure 5.7 shows that with a toll
r∗ = 30, the users to the right of N ∗ indeed will find it no longer attractive to enter the road (in
contrast to the users to the left of N ∗ ). The toll adds to the generalized price p for road use:
the equilibrium generalized price p∗ becomes equal to ac + r∗ (= 40 + 30 = 70 in the example),
and N ∗ becomes the equilibrium with tolling for the same reason as why N 0 is the equilibrium
without tolling. This economic way of dealing with congestion is known as road pricing (which
is why we use r∗ to denote the optimal toll). Implementation of this policy indeed yields the
desired increase in social surplus, which is again denoted by the shaded triangle G.
5.4. Policy Implications 99

!
mc
100 ac + r*

80
G
p* ac
60
p0
40 D = mb
r* = mec (N*)
20

N
10 20 30 40 50 60 70
N* N0

Figure 5.7: Achieving the efficient equilibrium through optimal congestion pricing

A number of characteristics of this so-called toll-supported optimum are worth emphasizing.

The first is that the toll secures that in the optimum, the equality between marginal benefit and
marginal cost is secured. To achieve this, the toll is set equal to the marginal external cost in
the optimum. This means that each road user pays a toll exactly equal to the costs he or she
causes for all other road users. These costs remain unpaid in the free-market equilibrium. The
use of optimal tolls is therefore commonly referred to as the internalization of external costs:
the consumers are, through the tax, confronted with the costs that their actions (in this case:
their presence on the road) imposes on others.6
Secondly, the fact that in the free market equilibrium mec = 45, does not mean that the
optimal toll is equal to 45. The optimal toll is equal to the marginal external cost in the optimum
(so: 30, in the example). This means that an empirical estimate of the marginal external cost in
a free market equilibrium does not yet give us the optimal toll level.
Thirdly, the fact that an external cost constitutes a market failure does not mean that it
should be completely eliminated. It should be optimized. This typically implies a lower level of
congestion in the optimum, but not a zero level. Practically, it will typically not be optimal
to eliminate all traffic congestion, but it will be optimal to reduce it. A complete elimination
would mean that most probably, some users are priced off the road whose benefit from making a
trip exceeds the marginal social cost.
Fourthly, because the intensity of demand and as a result mec will typically vary over time and
place, so will the optimal toll. Figure 5.8 illustrates this for our example. Suppose that outside
6
Just to make sure: the mec is equal for all users on the road in any equilibrium. For example, the fact
that mec = 10 at N = 10 in the numerical example should not be misinterpreted to mean that the mec for this
particular user remains equal to 10 also if N = 30. It is only equal to 10 for that user if N = 10.
100 Chapter 5. The Transport Sector: Roads

!
mc
100

80
ac
60

40 D peak

20
D off - peak
N
10 20 30 40 50 60 70

Figure 5.8: Optimal tolls (the arrows) vary as demand varies

the peak period, the inverse demand function Dof f −peak applies, reflecting a lower demand at
each generalized price level. The optimal congestion toll (indicated by the small arrow) will
then accordingly be much lower than during the peak (the original, larger arrow). An important
advantage of (electronic) congestion pricing, over other instruments such as capacity expansion,
is that the instrument can be targeted to reflect actual traffic conditions.
Finally, the optimal toll achieves the efficient outcome for two reasons. The first is that
the optimal level of road use is achieved; N ∗ = 30 in the example. The second reason is more
subtle but at least as important. This second reason is that also the optimal composition of
road users is achieved. The 30 remaining trips in the example are those trips that represent
the highest benefits. This is realized because the toll secures optimal ‘self-selection’ of road
users: those with benefits below ¤70 will not find it attractive to enter the road. Alternative
policies, that also may lead to a level of road use of 30 but that do not discriminate according
to marginal benefits, will typically realize a gain in social surplus smaller than G, despite the
fact that a casual observation of the situation on the road—simply counting the number of
cars—may suggest that the optimum is reached. The trips may include some that are located to
the right of N ∗ according to the inverse demand function. The next section will illustrate this
point further, by considering the effects of a non-toll policy in the regulation of congestion.
In conclusion, the optimal road use in our model can be realized by setting an optimal
congestion charge, equal to the marginal external cost in the optimum:

r∗ = N ∗ × ac0 (N ∗ ) (5.11)

which for our numerical example means:

5.4. Policy Implications 101

r∗ = N ∗ × c1 (5.12)

5.4.2 The efficiency of non-price policies: an example of ‘government fail-

ures’

Despite the welfare improving impact of optimal tolling, and despite the widespread increase
in traffic congestion, economists have not been able to convince policy makers and the public
at large of the attractiveness of congestion pricing. Real applications of congestion pricing in
transport are scarce. This is closely related to the limited public acceptability of pricing: most
voters in a democracy simply do not like the concept of new taxes. As a result, alternative ways
of coping with congestion have been proposed and applied. One of these is the expansion of
road capacity. However, for our application we will assume that capacity is fixed. Other policy
instruments try to achieve reductions in road use through non-price measures. In this section,
we will analyze such non-price instruments from an economic perspective, with the eventual
purposes of (i ) showing how economic methods can be used to analyze non-price instruments,
and (ii ) to further clarify the final statement in the previous section, namely that optimal
pricing results in an efficient outcome because it simultaneously secures the optimal level, and
the optimal composition of road use. We will see that this is an important feature of pricing,
that distinguishes the efficiency of pricing from that of non-price measures.
As an illustration, we will consider the impacts of a policy that resembles a policy currently
in operation in Athens, Greece. This involves regulation by number plates: on odd days (the
1st, 3rd, etc. of a month), only cars with an odd final digit on the number plate are allowed on
the road; and the other way around on even days. At first glance, this may appear a good way
of dealing with congestion: road use can be reduced substantially, without having to charge an
unpopular toll. However, there are some serious drawbacks. One is that this policy has, over the
years, mainly led to increased car ownership. Many households now have two cars: one with
an odd and one with an even number plate. This is a rather inefficient outcome: if the policy
mainly leads to higher capital costs (more vehicles are owned), without substantially reducing
traffic congestion, a welfare loss rather than a gain may be created. In our example, we will
ignore this aspect, and will study the efficiency of non-price policies even in absence of this
particular disadvantage. So: we assume that car ownership is given.
In our example, the optimal level of road use is two-thirds of the free-market level (N ∗ = 30;
N 0 = 45). To avoid exaggeration of the policy’s inefficiency, we consider the case where road
users are divided in three groups (rather than two as in Athens) of equal size, for instance on
the basis of the final two digits of their number plates (i.e., 00–33 for group A, 34–66 for group
B, and 67–99 for group C; ignoring rounding errors). On each day, only two groups are allowed
on the road, and one is forbidden. The question we ask ourselves is: is this policy (nearly) as
efficient as optimal pricing?
102 Chapter 5. The Transport Sector: Roads

!
mc
100

80
n ac
k l
60 m
p AB
40 D

20 D AB

N
10 20 30 40 50 60 70
AB
N* N

Figure 5.9: The efficiency of a non-price policy

To answer this question, we first determine the new inverse demand function under this
policy. If the groups are indeed of equal size and similar in terms of composition, each group
has an inverse demand function with the same intercept, but a slope three times as steep as the
original inverse demand function shown in Figure 5.4 (and later figures): at each price level, each
group would demand one-thirds of the total number of trips demanded. The inverse demand
function with only two groups present, and one excluded, will therefore be 1.5 times as steep as
the one shown in Figure 5.4 (and hence twice as flat as that for one group): the group excluded
can no longer demand trips. On a day with only groups A and B allowed on the road, and C
forbidden, the prevailing inverse demand function DAB will therefore be as shown in Figure 5.9
above. The original inverse demand function, for all groups together, is shown as the dashed D.7
Because no toll is charged under the number plate policy, we can again find the equilibrium
on an ‘AB-day’ at the intersection of DAB and ac. This yields N AB = 36. A first surprise is
that this exceeds the target level of 30 (the original value of N ∗ , also indicated in the figure),
although exactly one third of the potential road users is excluded, and the target was to reduce
the number of trips by exactly one third. The explanation is that the absence of pricing, and
the exclusion of C-users, make the generalized price at N = 30 so low that some additional
AB-users will still find it attractive to enter the road.
This slight deviation from the truly optimal number of trips is, however, not the main source
of inefficiency due to the policy. With the aid of Figure 5.9, two such sources can be illustrated:

7
It is of course immaterial whether we consider a day with only AB-users, or any other day. Every day, one
group will be excluded, and because we assume that the three groups are identical, an identical analysis could be
made for AC-days and BC-days.
5.4. Policy Implications 103

1. The policy prevents the use of the road also by those C-users who attach relatively high
benefits to their trips, and who would make their trip under optimal pricing. The associated
loss in social surplus is given by the shaded areas k + l. These areas show the loss of
total benefits over the first N ∗ users under the number plate policy (according to DAB ),
compared to the total benefits for the same amount of drivers under optimal pricing
(according to D).

2. Even when ignoring that users C are not allowed on the road, a non-optimal use for the
remaining group of AB-users results. These users, too, impose externalities on each other.
For groups A and B alone, the optimum would be at the intersection of DAB and mc; not
DAB and ac as actually occurring. The additional welfare loss of having the AB-users
between N ∗ − N AB on the road amounts to shaded areas m + n.

Table 5.3 repeats the earlier results for the free-market outcome and the optimum to allow a
comparison with our number plate policy. As before, the reader is invited to check the results in
the table as an exercise.
Apparently, in the example, the two inefficiencies identified above are so important that the
policy even leads to a reduction in social welfare compared to the free-market outcome. In other
words: non-pricing policies may not only perform less well than pricing policies, but may even
perform worse than no policy at all. We would then have a very clear example of a government
failure: the government not only fails in steering the free market towards the optimum, but
even fails in improving efficiency. Instead, the opposite occurs. Consistent with our earlier

Table 5.3: Key characteristics of the free-market, optimal, and number plate equilibria in the
numerical example

N ac mc mec mb B C S =B−C

Free Market 45 55 100 45 55 3487.5 2475 1012.5

Optimum 30 40 70 30 70 2250 1200 1350
Number plate 36 46 82 36 46 2628 1656 972

explanation of why optimal pricing policies are so efficient, we can thus identify the primary
reason why non-pricing policies are typically less efficient than pricing. This is the inability
to discriminate among users according to their benefits. Even if the policy would succeed in
exactly replicating the optimal consumption level (the number of trips, in our example), it is
very unlikely that the optimal composition of consumers is realized. Such inefficiencies plague
non-pricing policies in general. Although the example given here refers to traffic congestion,
similar problems will typically arise in non-price regulation of other types of externalities; both
inside and outside the road transport sector.
104 Chapter 5. The Transport Sector: Roads

5.4.3 Summary

This section demonstrated the efficiency of optimal pricing as a means of dealing with external
costs, applied to the case of traffic congestion. A toll equal to the marginal external cost in the
optimum secures that the excess consumption of trips occurring in the free market equilibrium
is prevented. The generalized price, as faced by road users, equals the marginal cost of trips
under optimal pricing, as opposed to the average cost in a free market. This means that each
road user pays a toll exactly equal to the costs he or she imposes on all other road users. The
use of optimal tolls is therefore commonly referred to as the internalization of external costs.
The fact that an external cost constitutes a market failure does not mean that it should be
completely eliminated. It should be optimized. This typically implies a lower level of marginal
external cost in the optimum than in the free-market outcome, but not a zero level. Practically,
it will typically not be optimal to eliminate all traffic congestion, but it will be optimal to reduce
it. The required optimal toll may in reality vary over time and place, along with variations in
demand.
The optimal toll achieves the efficient outcome for two reasons. The first is that the optimal
level of road use is achieved. The second reason is more subtle. This is that also the optimal
composition of road users is achieved, through self-selection of users. Non-pricing policies, even
if successful in the first respect, will typically fail in the second, and the risk of government
failures is therefore to be considered seriously. As a result, the gain in social surplus that can be
achieved with such measures is typically below the maximum achievable gain, and it may even
be the case that a loss in social surplus is generated. This explains the popularity of pricing
instruments among economists for the regulation of externalities.

5.5 The social acceptability of congestion pricing

There appears to be a strong case on economic grounds for the use of pricing instruments in
the regulation of external costs. Nevertheless, there are only very few practical applications
of externality pricing in general, and traffic congestion pricing in particular, despite the fact
that reliable technologies for electronic toll collection are generally believed to be available and
affordable. Instead, most policy proposals to implement congestion charging appear to fail
somewhere during the political process, due to a very limited social acceptability. For instance, in
The Netherlands, we have seen no fewer than four different proposals for a wide-scale application
of pricing in road transport come and go in just more than a decade: peak hour permits
(spitsvignet), toll plazas (tolpleinen), cordon charging (rekeningrijden) and kilometre charging
(kilometerheffingen). It is important for economists to understand why such proposals fail. Does
this indicate that there is something wrong with the theory, which predicts substantial welfare
gains from road pricing that should even convince the greatest sceptics? Or are economists
simply unable to explain sufficiently clear the social benefits of internalizing external costs?
5.5. The social acceptability of congestion pricing 105

5.5.1 Some theoretical considerations

There is no single answer to the question just raised, and it would take too far to appraise here
all alternative answers that have been suggested in the literature. But it is useful to discuss one
possible answer, which is in fact already implicitly given in Figure 5.7 above.
Suppose that we would hold a referendum among the road users, asking them to give their
yes or no on a road pricing proposal. What would self-interested voters then choose? To answer
this question, it is helpful to divide the road users in two groups.

1. The first group of users are those between 0 and N ∗ along the horizontal axis of the
diagram. These are the users who will be on the road both with and without road pricing.
For these people, the generalized price of a trip will increase from p0 (= 55) to p∗ (= 70)
after implementation of the congestion charge. These people can therefore be expected to
vote ‘no’.

2. The second group of users are those between N ∗ and N 0 . These are the people who
change their behaviour, and therefore have to choose some alternative (abandoning the
trip, travelling outside the peak, using public transport, etc.). Clearly, also before the
implementation of the toll, this same option was already open to them, but they preferred
not to choose it. They prefer using the road against the initial price p0 . Therefore, also
these people must be worse off due to road pricing, and also these people will vote ‘no’.

This may appear a puzzling result. How can it be that a policy that achieves a gain in social
surplus would leave everybody involved worse off?
The answer is simple: we have so far forgotten the only ‘actor’ that does not vote: the
government. The government is the only party who initially gains from the policy, through
the collection of tax revenues. All others involved lose. And a policy that leaves everybody
except the government worse off simply does not stand a great chance of success in a democracy.
According to this explanation of the limited social feasibility of road pricing, the clue to the
problem should be sought in finding ways of using the revenues such that road users feel they
are (perhaps more than) compensated for their initial welfare losses.
The welfare gain created through the policy should at least in theory create an opportunity
to redistribute the revenues such that everybody is eventually better off due to the policy. A
direct reimbursement is of course not a very useful way of doing this, as tolls that are directly
refunded afterwards will not induce much—if any—behavioral response. A more sensible indirect
reimbursements could for instance be a reduction of annual vehicle taxes. And in Chapter 2 we
will consider the use of revenues for the financing of highway capacity.
There is an important lesson in this simple analysis. And that is that efficient policies, aimed
at maximizing social welfare, may nevertheless and somewhat paradoxically be extremely difficult
to implement in a democracy. Individuals’ voting behavior will often not so much depend on
the aggregate welfare effects of proposed policies, but much more on the distribution of these
 There is an important lesson in this simple analysis. And that is that efficient policies,
aimed at maximizing social welfare, may nevertheless and somewhat paradoxically be
extremely difficult to implement in a democracy. Individuals’ voting behaviour will often not
so much depend on the aggregate welfare effects of proposed policies, but much more on the
distribution of these welfare effects. And policies that aim to internalize external costs
106 Chapter 5. The Transport Sector: Roads
through taxation have the unattractive distributional consequence that the government will
certainly be among the main initial winners. And that reduces the democratic viability of such
welfare effects. And policies that aim to internalize external costs through taxation have the
policies.
unattractive distributional consequence that the government will certainly be among the main
initial winners. And that reduces the democratic viability of such policies.
1.5.2. Some empirical evidence
5.5.2 Some empirical evidence
A recent study in The Netherlands aimed to investigate the empirical relevance of the
A recent study
theoretical in The Netherlands
considerations aimed The
given above. to investigate the empirical
study involved relevance of
a questionnaire the theoretical
among morning
considerations given above. The study involved a questionnaire among morning
peak road users in the Randstad Area. Some 1300 responses were obtained. This sectionpeak road users
in the Randstad
briefly reports theArea. Some
answers 1300relevant
on two responses were obtained. This section briefly reports the
questions.
answers on two relevant questions.

Road users' opinions on road pricing

100

90
80
Percentage of respondents

70

60
50
40
30
20

10
0
Bad idea Moderate Don't feel like Good idea No opinion Opinion
idea paying, but dependent
still a good on use of
idea revenues

Figure 1.10 Road users’ opinions on road pricing

Figure 5.10: Road users’ opinions on road pricing

Figure 5.10 presents the answers to the direct question “What is your opinion on road pricing?”.
Nearly half of the respondents think that the implementation of road pricing is a bad idea.
About a quarter think it is only a ’moderately good idea’. Some 17% admit they do not feel like
paying for road use, but nevertheless think it may be a good idea. Less than 10% think it is a
good idea. This paints a rather pessimistic picture of the social acceptability of road pricing. At
the same time, the results are less pessimistic than what could be expected on the basis of the
theory, namely that all respondents would think it is a bad idea. One important explanation for
this divergence is that in reality, people with a higher value of time may expect to benefit from
the policy even before redistribution of tax revenues (compare Table 5.1). Such dispersion in
values of time is not accounted for in the model we have developed in this chapter.
A hypothesis that seems to be supported by the results concerns the importance that
respondents attach to the use of revenues for their overall evaluation of road pricing. Nearly
85% of the respondents indicated that their opinion partly depends on this matter.
1.1). Such dispersion in values of time is not accounted for in the model we have developed in
this chapter.
A hypothesis that seems to be supported by the results concerns the importance that
respondents attach to the use of revenues for their overall evaluation of road pricing. Nearly
85% of the respondents indicated that their opinion partly depends on this matter.
Figure
5.5. The 1.11
social shows theofresults
acceptability for the
congestion logical next question: then how should these
pricing 107
revenues be used? Various options were offered, which could be rated on a 5-point scale. A
very clear
Figuremessage arises:
5.11 shows the the mostforpopular
results usesnext
the logical are those that then
question: are closest to thethese
how should interests of
revenues
car drivers themselves. Next come revenue uses that favour other transport purposes, and least
be used? Various options were offered, which could be rated on a 5-point scale. A very clear
message
popular arearises: the most
allocations popularfunds
to public uses outside
are those thethat areofclosest
field to the
transport. Theinterests
chances of of
carpublic
drivers
themselves.
acceptance of Next come revenue
road pricing uses that
thus indeed appearfavor
to other
depend transport
on the usepurposes, and least
of revenues, and popular
particularare
onallocations to public
the question funds this
of whether outside the that
is such field car
of transport.
drivers feel The chances
they of primarily
benefit public acceptance
from theof
road pricing
revenues thus indeed appear to depend on the use of revenues, and particular on the question
generated.
of whether this is such that car drivers feel they benefit primarily from the revenues generated.

Road users' opinions on revenue use

5
4.5
Average score (out of 5)

4
3.5
3
2.5
2
1.5
1
s

l
s
rt
s

rt
s

ra

a
xe

e
ad

xe

ie
po

po

er
ne
ax
lit
ta

ta

n
ro

ns

ns

lt

ge

ge
ac
al

el

tra
in

tra

ra
nu

lf
fu

in

in
ne
ts

oo
ic

ic
an

ts
n

s
en

ge
bl

bl

nd
io

en
pu
m

pu

ar
n

t
uc

fu
n
tio

m
st

io
ts

es
ed

st

lic
ve

uc

ct
en

ve
di

b
u
R
In

ed

Pu
ed
i
m

in
bs
R

st

ic
Su
ve

bl
Pu
In

Figure 1.11 Road users’ opinions on the use of road pricing revenues
Figure 5.11: Road users’ opinions on the use of road pricing revenues

5.5.3 Summary
Efficient policies, aimed at maximizing social welfare, may somewhat paradoxically be extremely
difficult to implement in a democracy. Individuals’ voting behavior will often not so much
depend on the aggregate welfare effects of proposed policies, but much more on the distribution
of these welfare effects. And policies that internalize external costs through taxation have the
unattractive distributional consequence that the government will certainly be among the initial
winners. And that reduces the democratic viability of such policies.
This general mechanism is particularly relevant for congestion pricing. In the model developed
in this chapter, it even appeared that before the taxes are redistributed, everybody involved,
except the government, will be worse off due to the policy. It was hypothesized that one of the clues
for increasing the social acceptability of road pricing could therefore be the careful specification
of an appropriate tax revenue allocation scheme. Empirical results for The Netherlands are in
accordance with these theoretical considerations.
108 Chapter 5. The Transport Sector: Roads
Chapter 6

Macroeconomic implications of ICT

6.1 Introduction

In the last decade of the 1990s, expectations about Information and Communication Technology
(ICT) and its implications on the economy were enormously high.12 Many policy makers,
scientists and entrepreneurs considered that the ‘old’ economy has changed into a ‘new’ economy;
an economy that exhibits mostly favourable characteristics. Or in the words of Eric Bartelsman
and Jeroen Hindelopen in 2000:
“This development is now clearly discernible in the observing economy community. Both
flashy magazines for young entrepreneurs and respectable yearly reviews of experienced policy
makers are full of impressions of the ‘new economy’. From the multitude of opinions of a similar
multitude of economists, it is at the moment not yet clear what the economic implications will
be of this surge in ICT. According to some, a ‘new economy’ will be created that is redeemed
of economic scarcity. Others lower this optimism and emphasize that that the economic
consequences of the developments in ICT will be close to zero.”(Bartelsman and Hinloopen 2000)
After the Internet bubble collapsed roughly in 2001 and some time has passed, we can see
more clearly now how ICT is different from other technologies and how it is similar. Going back
to the years before 2001, the ‘new economy’ was considered to be about fundamental changes
in the economy, with ICT occupying the central stage. ICT itself (i.e, hard- and software,
automation services and telecommunications) is nowadays by some thought to constitute a
so-called general purpose technology which widely affects the economic process. According
to some analysts, ICT was actually changing the economic process so fundamentally that old
economic experiences and insights were no longer valid. The economy was even said to be stand
on the eve of a period of sustained high economic growth, underpinned by high productivity
growth, low inflation and no business cycles. A period which was to be governed by new economic

1
This chapter is primarily based on CPB (2000) and Gelauff and de Bijl (2000), with some slight modifications.
2
To be honest, current expectations about ICT, and especially big data and data science with its large
companies as AirBnb, Uber, Booking and Facebook, are perhaps even higher!

109
110 Chapter 6. Macroeconomic implications of ICT

laws. Because reproduction and copying became increasingly cheaper, some analysts even argued
that the relation between prices and marginal costs would disappear.
After the collapse of the Internet bubble, we now know that most of the above does not
hold true. However, ICT did and still does make an impact on the economy as a whole. This
chapter deals with that macroeconomic impact. Basically, it shows that ICT should be seen as
the next stage in an ongoing process of change. The next section shows that from a historical
perspective the characteristics of ICT corresponds to a certain extent to those of earlier general
purpose technologies. Those characteristics modernize and support the economic process. The
discussion then turns to the implications for productivity and to the question whether old
laws and experiences still apply. This last question can be analyzed from both the meso and
the macro perspective. The meso perspective looks at the functioning of markets, the macro
perspective at cyclical movements and inflation. The main lesson to be drawn from this chapter
is that substantial changes have occurred and are occuring in the economic process, but the
fundamental economics laws have remained the same.

6.2 ICT as a general purpose technology

ICT has been around now for about 40 years and shows some stunning economic features. For
example, average prices of computer have decreased annually with more than 15%. In general,
prices of ICT investments have declined on average with 8% during the last 40 years (Bartelsman
and Hinloopen 2000). In the Netherlands, on every desktop and in most households there is
nowadays a personal computer (PC). As Figure 6.1 shows, penetration of PCs with Internet
access have grown rapidly since the late 90s.
This fact alone, does not make ICT a general purpose technology. The general appearance of
light bulbs does not make light bulbs a general purpose technology (although electricity itself is,
see below). So what makes ICT a general purpose technology (from now on we use the acronym
GPT)?
In general, four characteristics can be distinguished which a GPT must exhibit. First, there is
scope for improvement. GPTs witness an evolutionary process, in which the technology improves
and costs of applications decrease. An example would be the invention of electricity, which is
nowadays widely used and the costs have declined steadily over the decades. Secondly, GPTs
exhibit a large variety in their applications. The technology can be found in a wide variety of
products and processes. Going back to our example; electricity can be found for example in
electric engines, radio’s, illumination, and production of aluminum (electrolysis). Thirdly, GPTs
have a wide range, i.e. they are evident in many areas of the economy. Note that this is different
than having a large variety; light bulbs can be found anywhere in the economy as well, however,
only electricity shows a large variety of applications. Finally, a GPT is complementary to other
existing or potential new technologies. This entails that the GPT not only creates technological
6.2. ICT as a general purpose technology 111

PC with internet possession

80

70

60

50
Percentage

40

30

20

10

0
1998 1999 2000 2001 2002 2003
Year

Figure 6.1: Penetration of PC’s with internet access in Dutch households. Source: Statline,
CPB.

adaptations and innovations in other areas, but that it also needs these to come fully into its
own.
As an example of this complementarity, we take an organizational innovation which is
complementary to the invention of electricity (see Figure 6.2).
At the time of the industrial revolution, water was the primary source of power. Basically, the
whole interior of an industrial plant (if any) was based on water power. All machines themselves
were connected with a central driving axle. Steam engines created some more flexibility by
driving various groups of machines. Only with the invention of the electric engine, came the
separate propulsion of engines into existence. Thus, the interior of plants were from now on
designed to resemble the production process instead of being dependent upon the main power
source. Moreover, the process of production was less vulnerable to disruptions. A failure in the
main power source did not cause the entire production to stop anymore.
ICT meets all requirements to act as a GPT. It has already undergone a significant evolution,
though there are certainly many more opportunities, especially in the area of networks. The
large variety is reflected in various applications, such as aircraft navigation, medical scanners,
DVD players, communications, ATMs, word processing, etc. The wide range is evident. There
are nowadays few spheres of activity in which ICT is not present in some form. Complementary
technological innovations have occurred in the above-mentioned applications. Moreover, ICT has
changed and still is changing the production processes, marketing, financing and organization of
businesses.
112 Chapter 6. Macroeconomic implications of ICT

Water Steam Electricity

100

90

80

70

60

50

40

30

20

10

0
1700 1800 1850 1900 1950 2000

Direct propulsion Group propulsion Seperate propulsion

Figure 6.2: Share primary energy source (percentage) and prevailing form of propulsion of
industrial machines. Source Freeman and Soete (1997).

If often takes a long time for a GPT to permeate the economy. Usually several decades
elapse between the invention of the technology and its large-scale application. For example,
as we can see in Figure 6.2 the principle of steam power had been known for seven decades
before steam power started to replace water power. The same accounts for the introduction
of electricity. These delays occur for a number of reasons. Complementary innovations and
structural changes in many areas take time. The same applies to learning processes, both within
businesses, households and in the government. Uncertainties, fixed and sunk costs other reasons.
The introduction of GPT’s often involves substantial outlays, which cannot be easily reversed.
The investment of staff retraining comes to mind here. Because of the uncertainties attached to
technological changes, a firm runs the risk of investing too early in the wrong technologies. It
will therefore tend to wait until a dominant technology has emerged. Adoption externalities
related to complementary technologies accentuate this: the sector producing the GPTs will
only have sufficient incentives to innovate when there are sufficient applications, and the sectors
applying the GPTs will only innovate when it is sufficiently well developed.
A gradual diffusion of the GPT ICT is therefore certainly imaginable. The now fully emergence
of the Internet and organizational applications in particular has lead to the breakthrough. The
Internet offers new means of communications and cooperations at a distance, and taps new
markets. Organizational changes have been necessary because computers increasingly replace
low-skilled administrative tasks. The remaining tasks become more service-oriented: advising
customers on products which are becoming increasingly complex and increasingly tailored
to individual preferences. This requires not only more education and training but above all
6.3. Productivity and economic growth 113

more skills in dealing with people and more autonomy for the staff. The gradual process of
organizational changes and the emergence of the Internet has therefore contributed to ICT
completely taking off in the 1990s.

6.3 Productivity and economic growth

If the historical analogy applies, and ICT did indeed became more significant, then does the
economy stand on the eve of a period of structurally high economic growth? This is definitely
not self-evident, for two main reasons.
First, the emergence of a GPT manifests itself in macro productivity only very gradually
and in a complex way. Hence, there is not necessarily a sharp acceleration at the macro level.
Diffusion of technological innovations often follow a S-shaped pattern: after a phase of gradual
penetration of the GPT, a period of more or less rapid growth follows, which eventually slows
down after a point of saturation. Productivity growth often follows a similar S-shaped pattern.
Macro productivity growth, however, is often constituted of a large variety of technologies, of
which the phase of development usually do not coincide. Thus, at a macro level, an S-shaped
pattern is usually hard to find.
Secondly, in the long term a GPT is more inclined more to prevent the saturation of
productivity growth than actually increase it. If no GPTs emerge, technological developments
eventually reach a saturation point in many areas and overall productivity will tend to slow
down. A GPT rejuvenates the growth process in the economy by creating a whole new range of
opportunities for further development. It is possible, though, that productivity growth could
come out higher over the long term if ICT makes the innovation process itself more productive.
In that case ICT would be different from previous GPTs.

6.3.1 The empirics: The United States

Empirical research should provide more evidence of the influence of ICT on productivity and
growth. The empirical debate has concentrated mostly on macroeconomic developments in the
United States. In the US average productivity growth dipped to 1.5% per year during 1972–1995
and has since 1995 accelerated to 2.5%, thus virtually returning to the golden age of the 1950s
and 1960s.
Table 6.1 analyses the components of the productivity growth in the United States based
on Gordon (1999) and CBO (2000). As Table 6.1 clearly shows, there is a cyclical effect of
0.3–0.4 percentage points. This occurs after a recession, when firms deploy previously idle
capacity, so that output grows without the number of hours worked rising signifantly. A better
measurement of inflation (lower inflation implies higher volume growth at the same valuation)
explains 0.1–0.2 percentage points of the higher productivity growth compared to 1972–1992.
According to Gordon, the remaining structurally higher growth is due entirely to the very high
productivity growth in the production of computers (35% per year) and software (10% per year).
114 Chapter 6. Macroeconomic implications of ICT

Table 6.1: Decomposition of productivity growth in the United States, 1995–1999. Source:
Gelauff and de Bijl (2000)

Gordon CBO

Actual 1995–1999 2.54 2.6

Cyclical effect 0.41−/− 0.3−/−
Structural 1995–2000 2.13 2.3
Structural 1972–1995 1.47−/− 1.5−/−
Acceleration 0.66 0.8
Inflation measurement 0.19 0.1
a
Computer industry 0.51 0.4b
Capital intensification −−−/− 0.4−/−
Unexplained −0.04 0.1
a
Production of computers and software.
b
Production of computers.

Hence, Gordon argues that there is no evidence that ICT has boosted productivity in the private
sector outside the computer industry. The CBO considers only computer manufacturing sector,
and assesses its contribution at merely 0.2% percentage points. Capital intensification explains
virtually all of the remainder (0.4% percentage points).

Now, this does not mean that ICT only affects the computer industry. The capital intensifi-
cation calculated by the CBO consists mostly of investments in ICT throughout the economy.
Gordon does not examine productivity growth in other sectors of the private sector. There may
well be sectors which under the influence of ICT are achieving structurally higher productivity
growth, and other sectors which for whatever reason are seeing structurally lower productivity
growth. Moreover, the question arises as to the appropriate frame of reference. Because historical
analogies indicate that without a GPT, productivity growth tends to slow down over the long
term, stable structural productivity growth over time may still points to the influence of ICT.

Research at the micro level in the United States suggests that firms achieve efficiency gains
with the application of ICT if this is accompanied by a larger proportion of high-skilled staff
and changes in organization. Firms which rely heavily on ICT and employ a large number
of high-skilled staff are more productive than firms where these two factors are relatively less
prominent. And the productivity of firms which have either invested in ICT or have a large
number of high-skilled staff is higher than the productivity of firms which have made neither
of these two types of investment. This striking result confirms the complementarity between
ICT investment and the education qualifications of staff. The results are analogous for the
combination of ICT and changes in corporate organization.
6.4. Market behaviour 115

6.3.2 The empirics: The Netherlands

Figures of the late 1990s for the Netherlands offer no evidence of higher productivity growth.
in 1995–1998 macro labor productivity increased by an average of 34 % per year. Since 1986
productivity growth in the service sector has been particularly modest. In the first half of the
1990s productivity trends in the service sector lag behind those in the United States. This was
particularly the case for trade and the banking and insurance industry. However, productivity
growth in the transport sector was higher than in the United States. Thus, in a number of
sectors where the influence of ICT might be expected, there is as yet not much evidence in the
productivity figures for the Netherlands. Compared to the rest of the economy, the growth of
labor productivity in the Dutch ICT sector is impressive, at 4% on an annual basis. But this
figure is still considerably lower than the US figure. Moreover, the share of the ICT sector in
the economy is nearly twice as large in the United States compared to the Netherlands.

6.3.3 Conclusion

In conclusion, developments in the United States point to a renewing economy. Productivity

growth through above-average investments in ICT, the outcomes of micro-level research and the
high productivity growth in ICT-producing sectors offers indications for such a conclusion. Yet,
for the US economy as a whole, it has not been unequivocally shown that productivity growth
since 1995 is structurally higher than in the past. But if ICT as a GPT prevents a slowing of
productivity growth in the long term, then a stable growth could still point to the influence
of ICT. Although the contribution by the ICT sector to the growth of labor productivity is
high because of high productivity growth in the sector itself, this has not yet brought about
an acceleration in the Netherlands, as has happened in the United States. In addition, Dutch
productivity growth in the late 1990s has been appreciable lower than that in the United States

6.4 Market behaviour

Is a rejuvenation of the growth process through ICT accompanied by a change in the functioning
of markets? When analyzing this question, one should distinguish between markets for ICT goods,
markets for information goods and markets in general. This, because ICT influences these markets
in different ways. ICT goods include hard- and software, automation devices, telecommunication
networks and voice and data transmission services. These markets accounted for around 4.5% of
GDP in 1998. Information goods, such as news, information and entertainment, can in principle
be stored, transmitted and reproduced by digital means. The markets for information goods
accounted for around 2.5% of GDP in 1998. Markets in general covered all the other sectors in
the remaining 93% of the economy in 1998.
An important aspect in the analysis of the relationship between ICT and the functioning of
markets is the response of businesses and government to latent market failures (see also Section
116 Chapter 6. Macroeconomic implications of ICT

4.5). ICT can lead to market failure in the ICT goods and ICT information goods markets in
particular, because ICT generates externalities of market power. But this market failure often
remains latent: that is to say, firms develop strategies that reduce or even prevent market failure
(this is explained as well to some extent in Shapiro and Varian 1999). In general, this restricts
the need for interventions through government policies.

6.4.1 Markets for ICT goods

ICT goods differ from other goods because of the significance of network externalities and
switching costs. To recall: network externalities increase the usefulness of a particular product
or technology to a user as the number of users of compatible products or technologies increases.
Switching costs are incurred by a user who switches to a new technology, for instance another
software program. Both network externalities and switching costs imply that users can be
locked into a particular product or technology. That is why producers with a large market share
(e.g., Microsoft) have greater market power than is common in other sectors. The strongest
counterforce is innovation by new entrants and competitors. But new products have to offer
very considerable advantages to induce users to change. This gives established suppliers greater
scope to demand high prices. Competition supervision therefore needs to be particularly alert to
abuse of market power. Government policy can play a role in facilitating standardization and
innovation.

6.4.2 Markets for information goods

Information goods markets entail several forms of latent market failure. High fixed costs for the
production of information goods go hand in hand with low marginal costs of reproduction, while
digitalization makes perfect copies possible. If competition were to result in prices that were
equal to marginal costs, no one would produce these goods, because the fixed costs could not be
recouped. This problem does not seem insoluble, however. Several existing institutional solutions
help in this respect, such as copyrights and patents. Corporate strategies also ensure that latent
market failures need not manifest themselves. Examples are issuing new versions, combining
information goods with advertising and services, and distributing the goods via decoders.
ICT creates growing opportunities for producers (e.g., artists, writers, and musicians) to
publish information goods themselves and perhaps distribute them as well. Some artists are
already distributing music themselves via the Internet (e.g. via Youtube). This innovation
promotes access to and dissemination of information goods, and is in principle good for producers
and consumers. Publishers, however, may not have an interests, from the point of view of profit
maximization, in a wide distribution and a pluriform, innovative supply at low prices. It is
uncertain whether they will be able to maintain their current positions, for instance by creating
new added value, engaging in further concentration and cooperation (e.g., with Internet service
6.4. Market behaviour 117

providers) and developing encryption and distribution standards (to counter copying and control
dissemination).

6.4.3 Markets in general

Of course, ICT results in new goods and services the rest of the economy as well, and in new
opportunities to produce, promote and sell them. But this does not mean that all markets and
sectors will operate differently or change to the same extent. ICT is likely to have favorable
effects in markets where intermediaries fulfill an actual or potential role, such as the property and
financial services markets. New technologies, such as the Internet, make it easier for outsiders to
enter the market and compels established intermediaries to revamp their services and contacts
with customers.
Broadly speaking, transaction costs decrease and market transparency increases. The Internet
adds to the range of available distribution channels. Consumers can use search engines and
other sites on the Internet to collect product information and compare prices, which means that
existing markets will operate more efficiently. And participants in user and news groups can
exchange information very quickly about experiences with suppliers. This means that businesses
can build up a reputation more quickly and that market entry becomes simpler.
Against this, there are a limited number of potential barriers. New information asymmetries
develop in experience goods in particular, because buyers cannot visit online shops and therefore
find it more difficult to come to a considered decision. Moreover, consumers and suppliers have
to learn how to handle the new technologies. Online shopping does not yet seem developed
enough to exert price pressure or reduce price dispersion.

6.4.4 Conclusion

Taking all this together, it is not possible to draw a firm conclusion about the functioning
of markets in the economy as a whole. It is true that transaction costs are falling in various
markets. The potential significance of this in terms of social welfare may be considerable. Several
markets may also become more transparent as a results of the rapid spread of information via
the Internet. Two groups of markets, which together account for around 7% of GDP in 1999
will undergo more profound changes. In the ICT goods markets the established suppliers have
greater market power in the other sectors. The policy focus here will be above an on innovation
and standardization. And in those information markets where physical appearance is not crucial,
the Internet offers itself as a sales and distribution medium and established publishers may not
be able to maintain their positions.
118 Chapter 6. Macroeconomic implications of ICT

6.5 The end of the business cycle?

Empirical evidence for the United States indicates that cyclical movements moderated in the
second half of the 20th century. If ICT is at the root of this, then how could this happen? The
main direct cause are formed by the dynamics of inventories. On a day-to-day basis firms cope
with fluctuations in demand by eating into inventory stocks or allowing these to rise if demand
is more sluggish. To prepare for these fluctuations, over the long term firms aim at a stock
of inventories that constitutes a fixed proportion of their production. This makes inventory
movements on a quarterly or annual basis pro-cyclical, however, that is, they reinforce the
business cycle. Developments related to ICT, such as just-in-time delivery and more efficient
distribution in the retail trade, reduce desired and hence actual inventory levels. This process
has been unfolding for some time. Hence, it is not an outcome of recent developments in ICT.
Nor does it mean that inventory replenishment or depletion no longer contributes to cyclical
movements.
Several other structural factors may also be moderating cyclical movements, but these are
either controversial, too limited to herald the end of the business cycle, or only partly related to
ICT. This is the case for the growth of the services sectors in the economy, deregulation and
internationalization.
Apart from these structural factors, current economic policy is more stable, both in the
United States and in Europe. Cautious fiscal policies have replaced interventionist policies,
which often had pro-cyclical effects. In Europe, the EMU criteria and the stability pact certainly
contribute to this. Central banks, with the Fed leading the way, are reacting effectively to
monetary developments, not least because they are more concerned about the situation on the
financial markets, which are very sensitive to monetary developments and policy measures.
Reviewing these arguments, it seems that economic policy and inventory movements are the
main reasons for the more moderate business cycles. But these are only partly related to ICT.
And in any case, the interaction between financial markets, business profits and investments
remains a key impetus for cyclical movements. The proposition that ICT will kill off the business
cycle is not persuasive. At most ICT has some weakening effects on cyclical movements.

6.6 Lower structural inflation?

Developments in the United States in the late 1990s provide the basis for another point of
debate, namely the view that ICT leads to structurally lower inflation. How else can one explain
the weak inflationary pressures in the United States despite a prolonged period of economic
growth? There is of course another explanation, namely that the dollar’s appreciation and
the low commodity prices in the wake of the Asian crisis eased inflationary pressures. Or that
the high investment levels, which led to a forceful expansion of production capacity, and the
absorption of part of demand by a sharp rise in imports meant that capacity limits were not
reached, so that no inflationary pressures developed.
6.7. Policy implications 119

The main impact of ICT on inflation should then make itself felt through the labor market.
There has been a fall in what is called the ‘non-accelerating inflation rate of unemployment’
(NAIRU). For 1995, the NAIRU in the United States was estimated at 5 34 %. According to
estimates for 2000 the NAIRU has fallen to 4–4 12 %. This would mean that the current low
employment rate still does not exert any upward pressure on pay levels.
But why would NAIRU have fallen in the United States? And to what extent is this due to
ICT? A first effect of ICT concerns the role of the Internet in creating a more efficient match
between supply and demand on the labor market. Both job-seekers and employers are using the
Internet increasingly more to find each other, at considerably lower costs than the traditional
methods of advertising. If, consequently, the number of successful matches between supply and
demand increases per unit of time, the NAIRU will fall. But an opposite effect also plays a role
here. If ICT reinforces the process of creative destruction, then the number of job changes will
increase. So every individual match may become more efficient, but at an aggregate level this is
less noticeable because the number of required matches is rising.
The second reason why NAIRU is falling is the acceleration in productivity growth in the
United States since 1995, which according to the analysis above is related to ICT. Higher
productivity in principle translates into higher wages. However, wage levels adjust only gradually,
and it can take some time before higher productivity growth manifests itself in the wage
bargaining process. In the meantime the NAIRU will be lower. This effect is only temporary,
however. Once expectations about productivity have adjusted, the NAIRU will return to its
original level.
Taking all this into account, the main permanent effect of ICT on inflation concerns more
efficient matching of supply and demand on the labor market. However, this effect can at most
explain several tenths of percentage points of the fall in NAIRU. So, the structural impact of
ICT on the inflation rate is very modest.

6.7 Policy implications

What is the role of public policy in the ‘new’ economy? In general, enabling new players to
enter the market place. Besides, public institutions usually adopt GPTs first. A GPT is often
accompanied by changes in institutions, which only emerges after a learning process. Major
uncertainties and the government’s information deficit in a rapidly changing environment limit
the effectiveness of government intervention, regulation and public provision.3 Coordination via
the market, on the other hand, offers opportunities for experimentation and selection. Moreover,
corporate strategies are often quite successful in resolving latent market failures. Therefore,
markets for goods and services have to be sufficiently flexible to react innovatively and creatively
to the challenges posed by ICT.

3
Recall Section 4.5 that deals with government failures in general and principal agent problems in specific.
120 Chapter 6. Macroeconomic implications of ICT

In some specific spheres the rise of ICT does require closer attention from policy makers.
The complementarity of ICT and high-skilled staff, together with the required learning processes,
underline the importance of education and lifelong learning.

ICT points to organizational changes. The complementarity of ICT and organisational

changes raises the question whether all segments of business and industry, in particular small
and medium-sized businesses, have sufficient information and incentives to carry out these
changes. The follow-up question is whether there is a role here for government as initiator and
intermediary (for instance, by sponsoring model projects) or whether private consultants can
fulfill this role just as well or maybe even better. ICT of course also influences the operation of
the public sector itself. And not only with regard to organization. ICT also offer considerable
opportunities for improving the efficiency and effectiveness of government policy.

ICT touches upon various aspects of the relationship between government and business. ICT
sometimes obstructs the government’s supervisory role, for instance in the financial markets,
where ICT makes innovations possible and at the same time makes financial flows less manageable.
It also complicates taxation on electronic commerce. More detailed analysis is necessary on the
demarcation of the roles of the government and the private sector with regard to investments in
the ICT infrastructure. Public investment made the Internet possible, but in these days it is by
no means self-evident that the public sector should take the lead in a further expansion of the
network, especially now that it is increasingly used for commercial purposes.

The market for ICT goods and information goods require additional attention from the
perspective of competition policy. Without common standards or strong impulses through
innovation from the market, established suppliers in the ICT goods markets have more market
power than in other sectors. And in the information goods markets, a point of attention is
whether powerful market players take control of technological developments at the expense of
entry and innovation.

The combination of a sober and open vision with regard to the impact of ICT on public
policy is also appropriate for the ‘new’ economy itself. It was and is certainly not just a
hype. ICT brings and will bring about major changes in a wide range of fields. But it is
also a pity to turn ICT into a hype, since one hype will only last until the next one comes
along. At the interface of ICT, knowledge, and the economy, many fascinating developments
are unfolding which are already radiating out to society, economic research and public policy.
6.7. Policy implications 121

6.8 Summary
This chapter argues whether ICT is a form of a General Purpose Technology (GPT).
To be a GPT a technology has to possess four features: scope for improvement, wide
variety of applications, permeated many areas of the economy, and complementary to
other technologies. It can be argued that ICT possesses all of those four characteristics,
although the productivity and economic growth effects of ICT are difficult to measure
and do not seem to be significantly different that other era’s.
Effects of ICT on market structure, structural inflation, business cycles and labor
markets are difficult to measure as well, however, it does seem that there is a
significant effect and that this effect is persistent. Especially changes in transaction
costs and coordination mechanisms seem to be relevant.
122 Chapter 6. Macroeconomic implications of ICT
Bibliography

Aalders, R. F. T., Dijkgraaf, E., Varkevisser, M. and Vollebergh, H. R. J.: 2002, Welvaart en de
regulering van netwerksectoren, Ministerie van Economische Zaken, Den Haag.

Albert, R., Jeong, H. and Barabási, A.-L.: 1999, Diameter of the World -Wide Web, Nature
401, 130–131.

AVV: 1998, Advies inzake Reistijdwaardering van Personen, Adviesdienst Verkeer en Vervoer,
Directoraat-Generaal Rijkswaterstaat, Ministerie van Verkeer en Waterstaat, Rotterdam.

AVV: 2002, Verkeersgegevens Jaarrapport 2001, Adviesdienst Verkeer en Vervoer, Directoraat-

Generaal Rijkswaterstaat, Ministerie van Verkeer en Waterstaat, Rotterdam.

AVV: 2003, Prestaties Nederlands Wegennet: De Ontwikkeling van het Wegverkeer, de Wegca-
paciteit en Congestie in Verleden en Toekomst, Adviesdienst Verkeer en Vervoer, Directoraat-
Generaal Rijkswaterstaat, Ministerie van Verkeer en Waterstaat, Rotterdam.

Ball, P.: 2004, Critical mass: How one thing leads to another, Arrow Books, London.

Bartelsman, E. and Hinloopen, J.: 2000, De verzilvering van een groeibelofte, ICT en de Nieuwe
Economie: Preadviezen van de Koninklijke Vereniging voor de Staathuishoudkunde 2000,
Lemma, chapter Soete, L., pp. 61–81.

Braeutigam, R. R.: 1989, Optimal policies for natural monopolies, in R. Schmalensee and R. D.
Willig (eds), Handbook of Industrial Organization , Volume II, Elsevier Science Publishers
B.V., chapter 23, pp. 1289–1346.

CBO: 2000, The Budget and Economic Outlook: Fiscal Years 2001–2010, Congressional Budget
Office, Washington, D.C.

CPB: 2000, Centraal Economisch Plan (CEP), CPB, Den Haag.

Freeman, C. and Soete, L.: 1997, The Economics of Industrial Innocation, MIT Press, Cambridge.

Gelauff, G. and de Bijl, P.: 2000, The Renewing Economy, CPB report 2000, 19–24.

Gordon, R. J.: 1999, Has the ‘new’ economy rendered the productivity slowdown obsolete?
Unpublished paper, Northwestern University.

123
124 BIBLIOGRAPHY

Jackson, M. O.: 2008, Social and Economic Networks, Princeton University Press, Princeton.

Laffont, J. J. and Tirole, J.: 2000, Competition in Telecommunications, MIT Press, Cambridge.

Lipsey, R. G., Bekar, C. and Carlaw, K.: 1998, What requires explanation?, General Purpose
Technologies and Economic Growth, MIT Press, chapter Helpman, E., pp. 15–54.

McCarthy, P. S.: 2001, Transportation Economics: Theory and Practice – A Case Study Approach,
Blackwell, Malden MA.

Newbery, D. M. G.: 1999, Privitization, Restructuring, and Regulating of Network Utilities, MIT
Press, Cambridge.

Noll, R. G.: 1989, Ecomic perspectives on the politics of regulation, in R. Schalensee and R. D.
Willig (eds), Handbook of Industrial Organisation, Volume I, North-Holland, p. 22.

Perry, M. K.: 1989, Vertical integration: Determinants and effects, Handbook of Industrial
Organization, Volume I, North-Holland, chapter Schalensee, R. and Willig, R. D., pp. 183–255.

Prieger, J. E.: 2001, Telecommunications regulation and new services: A case study at the state
level, Journal of Regulatory Economics 20, 285–306.

Shapiro, C. and Varian, H. R.: 1999, Information Rules, Harvard Business School Press, Boston.

Veraart, M. D. L.: 2004, Elektriciteitsnetwerken Afsplitsen, ESB 89, 140–141.

Watts, D. J. and Strogatz, S. H.: 1998, Collective dynamics of “small-world” networks, Nature
393, 440–442.