POLICY RESEARCH WORKING PAPER   2446
Forecasting the Demand                                                    This overview of issues that
regulators should be aware of
for Privatized Transport                                                  in demand forecasting
discusses challenges that
What Economic Regulators Should                                            come with the decision to
WHLAt Econornic Regulators SI1ould        privatize transport, the
Kno;w, and Why                                                            perverse incentives
introduced when privatization
Lourdes Trujillo                                                          teams use strategic demand
Emile Quinet                                                              forecasts to evaluate assets,
the most common problems
Antonio Estachge
with demand forecasting, the
reasons that demand
forecasting matters, and how
to think about demand
forecasting in the context of
regulation.
The World Bank
World Bank Institute
Governance, Regulation, and Finance Division                                        a
September 2000



| POLICY RESEARCH WORKING PAPER 2446
Summary findings
Forecasting has long been a challenge and will remain so        Many governments and regulators avoid good demand
for the foreseeable future. But the analytical instrumetits   modeling out of lack of conviction that theory and
and data processing capabilities available through the        models can do better than the "old hands" of the sector.
latest technology and software should allow much better       This is dangerous when privatization changes the nature
forecasting than transport ministries or regulatory           of business.
agencies typically observe.                                     For projects amounting to investments of $100-200
Privatization brings new needs for demand forecasting.    million, a cost of $100,000-200,000 is not a reason to
More attention is paid to risk under privatization thaln      reject a reasonable modeling effort. And some private
when investments are publicly financed. And regulators        forecasting firms are willing to sell guarantees or
must be able to judge traffic studies done by operators       insurance with their forecasts to cover significant gaps
and to learn what strategic behavior influenced these         between forecasts and reality.
studies.
This paper-a product of the Governance, Regulation, and Finance Division, World Bank Institute-is part of a larger effort
in the institute to increase understanding of infrastructure regulation. Copies of the paper are available free from the World
Bank, 1818 H Street NW, Washington, DC 20433. Pl¢ase contact Gabriela Chenet-Smith, room J3-147, telephone 202-
473-6370, fax 202-676-9874, email address gchenet@worldbank.org. Policy Research Working Papers are also posted on
the Web atwww.worldbank.org/research/workingpapers. Antonio Estache may be contacted at aestachecLworldbank.org.
September 2000. (35 pages)
The Policy Research Working Paper Series disseminates the findings of work itt progress to encourage the exchange of ideas about
development issues. An objective of the series is to get the fihdings out quickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Policy Research Dissemination Center



EARLY DRAFT
COMMENTS WELCOME
Forecasting the Demand for Privatized Transport:
what economic regulators should know and why.
LOURDES TRUJILLO
(Universidad de Las Palmas de Gran Canaria, Spain)
EMILE QUINET
(Ecole Nationale des Ponts et Chaussees, France; and
Ministere de l'Equipement, France)
ANTONIO ESTACHE
(World Bank and
European Center for Applied Research in Economics and Statistics)
*We would like to thank Javier Campos, Gustavo Nombela, John Strong and many colleagues at the World Bank working on
tranpsort privatization projects ,for useful discussions, suggestions and comments. The usual disclaimers apply. Moreover, the
views expressed are own and do not represent in any way the views of the institutions we are affiliated with. Please send
comments to aestache@worldbank.org.






1
1. Introduction:
While public-private partnerships in the delivery of transport infrastructures and
services is expanding, there is also growing evidence of the lack of appreciation of the
importance of demand forecasting in preparing and monitoring these partnerships. This often
gives an opportunity to the private operators of transport services to complain, soon after taking
over a business, about overestimates or underestimates of demand based on the initial
information provided by governments. It tends to result in an excuse for the private operators to
try to renegotiate the contract to improve its terms.' The most common case finds a
concessionaire arguing that the "overselling' of the potential business by a government eager to
maximize fiscal gains, generated unjustified high expectations. And nevertheless it is quite
common that both regulators and concessionaires or bidders devote much more money to the
construction costs studies than to the demand analysis--the average ratio quoted among expetrs is
one to five.
The lack of focus on good demand forecasting in the context of an increased role of private
operators and investors in the transport sector may be somewhat counterintuitive. Transport
planners have a long tradition of concern for demand. The analysis of demand has been at the
core of the assessment of national or sectoral policy options, including the introduction of new
transport modes. But these concerns have generally been addressed through more "macro"
oriented modeling exercises focusing on identifying broad support for, or rejection of, policy
alternatives.
In the context of more specific evaluations of projects--such as high speed trains, subways
or new airports-or in the context of important policy decision regarding modal choices--, many
planners have also developed disaggregated demand models capable of generating the kind of
information much closer to what is needed to meet the forecasting requirements of a project
financier, a privatization commission or a transport economic regulator. There are a few very
good recent surveys of these models (see for instance Small and Winston (1999) in English,
Quinet (1998) in French) and several comprehensive textbooks on the subject (Ortuzar and
Willumsen in (1994) in English or Ortuzar (2000) in Spanish). They reveal impressive
improvements over the years to model demand and increased accuracy in identifying and
assessing the relative importance of the various factors explaining demand, including modal
shifts resulting from policy changes. Although there is a lot we still don't know, many countries
can now count on useful results on the value of time and the willingness to pay for various types
of transport service users. These are relevant indications when pricing new services to be
provided by private operators.2 It makes sense to compare the tolls or tariffs calculated from the
cost side with these rough estimates of the willingness to pay for some services or with the value
of time revealed by the post-mortem analysis of comparable projects increasingly found in the
literature (see Quinet on Marseille (1998), Small and Winston (1999) on the Dutch data base).
Unfortunately, this improved knowledge does not seem to be spreading to the toolkits used
by transport planners or privatization teams in developing countries--and for that matter... in
many developed countries-because few are willing to invest in the data needed to make the
IThis "overselling" and the common resulting renegotiation of the terms of the contract typically allows th eprivate operator's to obtain
extensions of contract duration, new or larger subsidies and/or lower investment requirements. All these changes alter the fundamental financial
terms of the contracts which often end up being significantly less attractive than anticipated from the viewpoint of the government--and often also
from the view point of the users. On the other hand, new fact sometime do arise and renegotitiation is not necessarily a bad thing, as long as it
follows fair and transparent rules.
2For a useful overview of how to value time and to relate this to willingness to pay, see Jara-Diaz, S.R. (2000) or Galvez, T.E. and S.R. Jara-
Diaz (1998)



2
most of these new tools. In many cases, "privatization" has a tight time table which does not
allow enough time to get the data and sufficiently work the model.3 In fact, many academics
would argue that few transport planners have effectively used the most recent developments in
demand modeling techniques. In France, for instance, one of the top transport research centers,
recently published a book taking stock of developments in the field outside of France, arguing
explicitly that it is intended to meet the need for France to catch-up in the area of demand
modeling by urban transport planners. 4
Luckily for regulators or project financiers, things are changing. The best evidence of
recent progress is that demand forecasting is a profitable business and that many of the firms
offering these forecasting services are willing to stand by their forecast offering guarantees or
insurances to their clients. The US has some good experience but so do Mexico and Brazil who
are relying on serious modeling to assess the future demand for its commuter rail and assess the
government financial support requirements for various levels of tariffs. 5 This market is likely to
grow and there is already a plethora of software to pick from, just like there is a plethora of
methodologies to pick from. The increased availability of private firm willing to offer advise on
demand forecasting will however not address well the second main type of error a regulator
needs to be concerned with: the traffic modelers often are not referees, but advocates for the
initial actors' interests as discussed later.
The main purpose of this paper is to guide regulators though this very technical and
somewhat foggy field to allow them to make the most of new opportunities to collaborate
effectively with specialists. The goal is not to provide a detailed technical introduction but rather
to provide a "light" overview of the main issues a regulator needs to be aware in demand
forecasting. To achieve this objective the paper is organized as follows. First, it discusses the
changes and challenges brought about by the privatization decision. Privatization introduces a
number of perverse incentives in the process that leads to strategic demand forecast used in the
evaluation of assets by privatization teams. Next it provides a checklist of the most common
problems with demand forecasting, highlighting the main reasons why demand forecasting
matters in practice. Section 4 then provides a brief overview of the main techniques. Section 5
provides some sector specific illustrations. Section 6 provides some practical tips as how to start
thinking about demand in the context of regulation. Section 7 concludes.
2. The changes brought by the privatization context
This section reviews the main changes resulting from public sector restructuring brought
along privatization. The increased role granted to the private sector does not mean that the public
sector disappears from the sector. In many ways this role is strengthened by the recognition of
the importance of an independent regulator. But the implementation of a restructuring that
recognizes the importance of independent regulation requires an increase in the number of
players since the policy and regulatory functions typically enforced by an integrated public
sector are now assigned to an unbundled public sector. The regulatory dimension of this
unbundling is quite complex since it provides a set options which influence the relative
importance of demand forecasting in the context of privatization. A particularly important aspect
3Privatization is here defined as any type of public-private partnership in which the private sector takes on the main responsibility for the
operation and the investment in the sector. It includes sales/divestitures, concessions/franchises, management contracts and greefield projects. In
developing countries, roughly 66% of the public private partnerships are done through concessions/franchises. For more details see Estache
(2000)
Dossier du Centre d'Etude sur les Reseaux, les Transports, l'Urbanisme et les Constructions Publ iques (1998), "Comportement de deplacement
en milieu urbain: les modeles de choix discret"
5This is that it is not cheap. Depending on the specifics of a projects at good demand study can cost as much as $250,000 dollars.



3
of this choice is the trade-off between the incentives given to the operator to perform well and
the risks that this operator is expected to take on. The specific assignment of responsibility and
of the choice of regulatory instruments are the main reasons for the strategic use of demand
forecasting by the key players involved in the privatization process as discussed next.
2.1. The diversity of objectives across actors
In practice, at least four groups of actors are involved: consumers, operators (in a large
sense, that is including sponsors and financiers), the government (which represents the taxpayers
and the voters) and the regulator and it is important to understand how their concerns differ. Users
will worry about prices, service quality and reliability. All influence demand. The operators
typically  worry  about profits, risks and  market power.6  All are influenced  by  demand.
Governments, who are often the dominating players in the context of the reform of the sectors
covered here, are generally interested in reducing the fiscal burden imposed by the public
enterprises of the sector and often also try to generate a flow of resources through the reform
process. They generally want to please tax payers by cuttting taxes and respond to some of the
constituants with environmental and distributional concerns.7 These concerns can both influence
demand and be influenced by demand. In discussing demand forecast key actors are often more
advocates of their agenda than scientifics. Regulators which happen to be the Government officials
or political appointees in many cases, want to balance everybody's concerns fairly accounting for
many aspects of demand typically ignored in the privatization process, including a reasonable
comparison of willingness and ability to pay to avoid, in particular in the case of passenger
transport, an unfair exclusion of some segments of the population.
2.2. The regulatory regime and the balancing of risks and incentive
Regulatory regimes come in multiple sizes and shapes (rate of return, price cap, hybrids,...)
and have very different effects on the behavior of operators in terms of pricing strategies, quality
or investment decisions. The real importance of demand forecasting in the context of privatization
of local monopolies cannot be appreciated without a proper recognition of its relevance for the
effectiveness the regulatory regime chosen. In turn, it is almost impossible to understand the
desirable design of the regulatory regime without recognizing that it depends on the ranking of the
various objectives (cost minimization, investment level and speed, social concerns, fiscal goals)
8
reflecting the many concerns of the various actors influenced by the privatization decision. .
From a regulatory viewpoint, the regulator must chose who bears the technical and
commercial risks. This means balancing essentially between "fixed price" (e.g. toll price caps
based and lumpsum construction cost contracts which pushes the operators to minimize costs and
contracts, in which both risk types are passed on to the operator) and "cost-plus" type contracts
(which guarantees that the operator recovers at least cost and make a minimum rate of return on
the business). If the concern for efficiency dominates and business risks are minor, fixed price
type contracts would dominate all other forms of regulation. Under this regime, demand would
mostly be the operator's problem since it is supposed to decide autonomously the risks it is willing
to take. It turns out that fixed price contracts are most likely to attract efficient private operators
6Most regulators would argue that the concems of the financial advisors and of the operators do not always coincide The operators mostly care
about getting the deal signed and this requires very good demand prospects . The operator wants to have a longer term view and is typically more
cautious about the prospects.
7They are sometimes as well concerned with the need to deal with unions as the restructuring of these sectors often result in labor redundancies.
8For a longer discussion of this point, see Crampes and Estache (1998)



4
when risk levels are low. This means that the more precise demand is, the lower the risk, and thus
the easier it is to rely on fixed price contracts. In addition, the reliability of demand forecast allows
the regulator to better assess the actual efforts of the operator to improve efficiency and to lower
the information rent potentially available to the operator (i.e. the pure profit that the operator can
obtain from its superior knowledge of costs and traffic type when compared to what the regulator
knows) and the incentive rent (i.e. the gains that may be obtained from saving on the cost of the
efforts needed to meet the efficiency requirements built in the concession contracts).
An alternative strategy for the regulator is to come up with a regime that requires a minimum
of information on demand. For toll roads and airports runways, Engel et alt (1997a & b, 1998)
suggest a new type of auction which consists in selling the toll or fare and the investment
obligations and to allow for variable duration so that the effects of demand uncertainty are
eliminated. The contract ends when to the operator has recovered its investment at the set toll
level. It protects against over and underestimates of demand. The mechanism is quite creative and
useful although it faces its own share of problems and should reduce the need assess demand from
the view point of a regulator. (see Tirole, DeRus and Nombela). In practice, it is very unlikely that
a regulator will ever have to ignore the interactions between its regulatory choices, the economic
environment of a project, including the determinants of costs and demand and the outcomes. 9
The problem is the most acute when risks are significant and there is a problem in attracting
potential bidders. In this context, some type of cost-plus form of regulation which allows better
risk sharing should generally be preferred. The main concern under this regulatory regime is that
operators will cheerfully endorse the overshooting of actual demand since it justifies investments
to be included in their asset base. If tariffs are set and demand is overshot, operators are simply
likely to get more time to recover the costs. This last problem is compounded in cases in which the
government structures the privatization around some payment to the public sector by the operator
and is thus concerned with maximizing the fiscal payoffs of this privatization. In this context, the
payment to the government will be highest when the potential value of the business is highest. This
is why the calculation of the minimum price a government is willing to accept for a "privatized"
asset is such a strategic variable. For a given asset value, the valuation of the business is essentially
driven by a demand forecast which both the operator and the government have an incentive to
overestimate. The better the demand prospect, the highest the minimum price the government can
hope for.
2.3 The relevance of the actor's strategic behavior
Whatever the regulatory regime selected, many of the players have a strong incentive to play
strategically. Politicians will want to look good during their tenure and support policies that
maximize short run fiscal payoffs and/or minimizes tariffs. They can do so quite consciously and
knowing perfectly well that requiring high payments and expenses from the operators while
imposing low tariffs are generally not consistent and sustainable policies. Willingness or ability to
pay and hence the real potential value of a business are seldom analyzed very analytically in this
context. The political gain for them to announce a new infrastructure is much higher than the
political loss of having to increase taxes; furthermore these concerns and the eventual renegotiation
of the deal is left to their successors since they generally imply political costs. But it is clear that
private operators happily play in this game. For many of the best deals, their main concern is to get
The regulator can be a little bit less concemed with the risk that the operator undershoots demand and that users do not get what they want
initially. If willingness to pay is demonstrated, operators are quite likely to make the additional investment requirements ad regulators will simply
have to ensure that there is no abuse in the pricing of the new service.



5
the contract signed by the government, knowing quite well that there is generally significant room
for renegotiation. Patience in this field is often rewarded once the contract is won.
In sum, there are enough reasons and there is enough evidence to argue that in the context of
privatization, it is not easy to achieve convergence on the views of what a good demand forecast
should be because both firms and government have some interest in playing strategically with the
demand forecast. This should make a convincing case to ensure that regulators do their best to
come up quite early on in their tenure with independent assessments of demand. This assessment
will be useful at almost every stage of a regulator's activity. Demand is important in most types of
conflicts that have to be resolved through tariffs or quality adjustments. Demand is important when
assessing financial support requirements for projects requiring subsidies.  Demand is also
important in understanding the distributional consequences of any regulatory decision. Demand is
finally important every time there is a renegotiation and this means it will often matter because
most contracts end up being subject to some degree of renegotiation.
3. How to deal with a wrong demand forecast?
As most policymakers will acknowledge, there are always many ways of getting any
policy decision wrong. In the case of demand forecasting, there are some clear patterns. It may
be useful to briefly review them since a better understanding of these recurring mistakes may be
a good place to start improving demand forecast. One way of approaching the problem and to
track the reasons for the importance of demand forecasting is to assess the consequences from
the viewpoint of the main characteristics of a transport system and in particular the changes in
this system to meet the expected demand.
Investment for significant infrastructures are quite slow and costly and at some point it is
necessary to try to take bets on growth prospects in demand. If these bets are unfounded, it can
be quite costly. Forecast can either be too optimistic or too conservative. Figure 1 summarizes
the main problems resulting from the mismatches between demand and supply for capacity and
the main suggestions discussed in this section.
3.1. How to deal with an overoptimistic demandforecast
Before getting into demand overestimation, it may be worth to point out that many
casual observers may easily be misled into seeing an overoptimistic forecast where there is
nothing but prudent long term planning! Consider the case of airports, roads or ports. The
investments needed to achieve the right medium to long run capacity are huge but mostly slow
and bulky (indivisibilities result in lumpy investments!). It will generally be impossible for most
of these types of infrastructures to design a flexible piecemal investment strategy to adjust
quickly to increases in demand for capacity)-justifying the presence of a single operator. In
many instances,. beltways or urban access roads have been build with assumptions
underestimating growth patterns. Because of the nature of the investment process, most of these
projects require some short run excess capacity to be able to absorb the longer run demand.
These are not mistakes. The main policy question here is whether to allow the tariff today to
cover the needs of tomorrow or to have the government pre-finance these future capacity needs
through subsidies. This is not a demand forecasting problem-since it is foreseen-- but it is a
pricing problem very similar, although different from the one resulting from a sheer forecasting



6
mistake in the context of a fixed duration contract build around specific demand and related
return assumptions. 10
Figure 1: Problems with demand forecasting
u   Peakpdemand            c
System                     Dieqilbr
Capacity                                     q        strategsupply  i  Congestionly
,andd     t as demand                    w   f  longems r
I tapacityex¢eeds  jDemand/
But xmistakes d p     a      e     s       g     p    nb             breexceeds
Problems to recover p          Can increase capacity    f    Cannot increase
ineestments   6                 in short run  H  a     capacth   in s fort run
Adjust prices or introduce     Cannot changes
foranal queuing strategies    pricesd.o.hcan only
to addresshows that demand    wait for long run!
But mistakes  do happen and there  is enough experience to benve opto  argue that
overoptimistic demand forecasting is common. For instance, traffic forecast for some of the most
publicized toll roads projects overshot actual traffic  from   25%   (Cuet n avaca-Acapulco  in
Mexico) tos an  mucrs 60% (M1   -MI 5          Highway in Hungary or the average for the Mexican toll
road progrna). Most of Asia's BOTs projects for toll roads were based on very optimistic growth
assumptions pre-dating the fallout at the end of the 1 990s.
A wonderfully detailed review  of the mavelopin  countries a          n 8 US cities by Don
Pickrell (1992) provides similar evidence. It shows that the forecast that led local goveorments
to advocate  10 rail transit projects over competing less capital intensive options grossly
overestimated rail transit ridership construction and underestimated rail construction costs and
operating expenses.   In 7 out of 8 cities, actual ridership was less than half of its forecast level.
Skamris and Flyvberg (1997) provide a useful survey of the few publications documenting cost
overruns and overoptimistic forecasts. Their conclusion is that traffic forecasts that are incorrect
by 20-60% compared with actual developments are common in large transport infrastructure
projects in a sample of countries including developing countries and the UK, Denimark and
10 Of course, the most cynical among our readers will try to argue that overoptimistic forecasts will eventually become realistic forecasts if
demand is given enough time to grow to be consistent with the size of the project. Once more, whenever possible, this is what makes variable
duration contracts attractive since they require a less precise forecast of demand.
IIPickrell, D.H. (I1992)



7
Sweden. 12Excess capacity is indeed resulting in a renegotiation of many road contracts in Latin
America and East Asia13 the users or the providers. It is for instance, difficult to re-use a railway
or an airport runway and yet the builders have to be compensated. This is why to the extend
possible, when demand is hard to forecast, it makes sense to try to follow a progressive or
piecemall approach to investment. But this advise can be hard to follow (indivisibility of
investment).
3.1.1. Why does overinvestment arise?
The first reason in the context of privatization is that one of the 'changes often brought by
the private operators is the introduction of cost reflecting prices and a switch from taxpayers to
users for the responsibility to pay for the service and an increase in prices.'4 If users are sensitive
to prices, a regulator would want to know how much. Users shop around. If the use of toll roads
is deemed too expensive, users look for free alternative itineraries-this is why the existence of
substitutes must be reflected in the demand forecast. But these changes are also important
because they influence available income and this in turn appears as the income effect in the
demand estimate. Income elasticity also matters and this is too often ignored by governments
and regulators. In sum, users are often more sensitive to prices and to "income effects" than
expected. This is why when there is no or little tradition of payment of fair prices, the
introduction of efficient pricing policies can result in significant trend changes and the past is a
poor indicator of the future. Since many planners rely on trends to forecast demand, this can lead
to significant overestimates.
3.1.2 What can a regulator do to deal with an overoptimistic forecast?
When the contract duration is fixed-i.e. cannot be changed without penalties or
renegotiation--, the main option is to cut tariff or and subsidize the operator. From an efficiency
viewpoint-i.e. to ensure that resources are allocated to their best use and that costs are
minimized--, tariffs should be set at least to the short run marginal cost. From a financial
viewpoint, this is not good enough since it does not allow the recovery of the costs of capital and
hence results in a deficit. On the other hand, trying to include investment amortization in the
tariff results in a demand inconsistent with the investment made-i.e. a lower demand than what
the government may want to cover. This means that the optimal tariff reduction or subsidy will
depend on the specific objective of the government. This objective will have to be spelled out
for a regulator to be able to act in this kind of situation in a manner consistent with the political
preferences without threatening the commitments made to the private operator.
One option is to allow the use of two part tariffs designed to allow the recovery of both
operational and capital costs. The idea is to set a unit price equal to the short run marginal cost
and at the same time to levy a fixed charge to recover the capital cost.15 In practice, this is done
12 Skamris and Flyvbjerg (1997)
13 .In most cases, the outcome is a longer contract and/or some explicit subsidy. Changes in tariffs are not that politically acceptable and hence
more seldom used. But it may be useful to remind the reader that the renegotiation is not necessarily a proof of failure. In many instances, it
means a better use of new information. In a nutshell, when considering the winners and losers of the renegotiation, the experience suggests that
the users are often winners when they can get a bener service from the private operator. The govemment ends up somewhat of a loser as they end
up having to continue financing the sector through subsidies more than they had expected. In net terms however, the sector represents a lower
burden on public funds than when the government was operating the sector.
14 This is why prices often increase initially, but they will often increase as a result of clauses in contracts that allow the adjustments to changing
macroeconomic conditions. Many private operators for instance will claim the right to be able to pass through variations in exchange rates or
increases in inflation through tariff increases.
15 From a more conceptual viewpoint, this allows a recovery of the cost from the consumer surplus.



8
by thinking of the infrastructure as a "club" and have frequent users pay an annual fee and then
have them pay the marginal cost every time they use the facility. Occasional users who are not
members of the "club" end up paying a higher usage fee every time they use the facility. This is
in fact how many governments finance their toll roads. What is labelled a subsidy may
sometimes be simply seen as a pass through of the revenue from the vehicle owners licence fee
that finances some of the fixed costs of facilities. This device is mainly used for rail
infrastructures and seaport traffics.
What is interesting from the viewpoint of the government concerned with the risks of
having to pay subsidies, is that this pricing strategy can allow the operator to explore alternative
forms of price discrimination between its users. It can recover the fixed part of the tariffs by
differentiating it enough and adjusting it to the willingness to pay of all potential users. This
tariff discrimination allows the recovery of a larger share of the investment than a flat fixed
charge would allow. The rule would be as follows. First, identify the various user groups. Second
set the fixed charge in relation to the degree of "dependency" the users have on the service. The
groups most dependent on the service provided-i.e. the groups with the demand least elastic to
prices-- will be charged the most-e.g. truckers who need to minimize transport time and cannot
use alternative longer and slower roads. The groups most likely to drop out of the market when
tariffs are too high-those with a demand most elastic to prices-will be charged the least-e.g.
tourist who can also enjoy the scenery on the slower road. This is not as simple as it seems. For
instance, the trucker working for an exporter of time sensitive merchandise may seem to be a
"captive" client of the road, but if the fixed charge becomes too high, it may payoff to
subcontract with a rail company. Similarly, it may be useful to be able to differentiate between a
tourist and a commuter to charge more from the commuter but few politicians would take such a
decision with impunity.
3.1.3 Politics vs. economics
The main problem with these solutions is in fact not technical but political. The
policitican will not hurt commuters who are likely to represent its electorate and make life easier
on tourists. For most politicians, a good tariff design is often like a good tax design: it exports the
burden to non-voters. So what can be done in the case of an overoptimistic forecast? The answer
is a bit cynical. First, if the earnest goal of the government is to minimize subsidies, it should not
give up considering price discrimination. It "simply" requires a little bit of creativity and a strong
commitment. Second, if this does not work, subsidies are the most likely outcome. In this
context, the two part tariff idea is useful because it allows a better targetting of the subsidies.
The government should focus its subsidy on part or all of the fixed part of the tariff. Often, in
this context, to limit the costs of subsidies, the government will have to replace the willingness to
pay concept by an ability to pay concept. This allows a focus of the subsidies on the share of the
fixed part to that would have accrued to the poorest users of the infrastructure to ensure the need
to achieve a financial balance does not result in an exclusion through prices of the poorest users.
Finally, if this does not work or is not enough, an extension of the contract duration, or a slower
investment requirement will generally do the trick. In fact, the Engel et alt (1997, 98) mechanism
which avoid setting the term of the concession contract granted to an operator would avoid many
of these problems... .although not necessarily all of them.



9
3.2. How to deal with an overpessimistic demandforecast?
Undershooting demand is less common in the general context of transport privatization
but it does happen and it is hence worth discussing briefly. In case of an overpessimistic forecast,
the main outcome is a lack of transport capacity and hence congestion.This can be quite dramatic
in the short run when it is impossible to revise investment plans to adjust quickly to the larger
than expected demand. This is a common problem in urban transport modes, metro rides are
often underestimated at peak time, urban access roads in many of the largest urban centers
witness permanent traffic jams at rush hours and in some cases throughout the day as in
Bangkok.
3.2.1. Why does underinvestment arise?
Underestimation arises for two main reasons: (i) privatization changes the perceived
transport services, changing the nature of the demand and (ii) the reformers have failed to
recognize the network aspects of the demand. The first effect is quite well documented in several
of the suburban rail concessions in Latin America where demand was underestimated quite
significantly. The main reason for the underestimation was that most policymakers were
concerned that the potential passengers of the privatized services would be reluctant to pay for
what they use to get for close to nothing. The source of the underestimation was the failure to
recognize the existence of a rationed demand for quality. The passengers are actually buying a
different transport product because they now get a safer, timelier and more reliable service.
There are many instances in which the demand for a specific mode is also suprisingly
stronger because the supply of another mode is deteriorating unexpectedly or faster than
expected.16 More generally, when a forecaster at the project level does not take in account the
interactions with a much wider transport network, it is easy to underestimate the derived demand
stemming from the network characteristics of a transport system. Since the network externalities
can go either way the sign of the correlation between the various modes is not always the same.
An improvement in a bus system may improve the demand for train or for subways if it fucntions
as a feeder as in many of the Brazilian Northeastern large cities. If these characteristics are not
accounted for, demand will generally be underestimated and underinvestment or underpricing are
the likely outcome. Implicit subsidies are created which can be addressed through changes in
pricing strategies which account for the complementarity of transport services and
infrastructures.
3.2.2 What can a regulator do to deal with an overpessimistic forecast?
If for whatever reasons, demand is temporary or occasionally stronger than expected but
a long term adjustment in capacity is not needed, the short run solution generally recommended
by economists is a temporary rationing through prices. This means that prices have to increase to
ensure that demand falls to meet the available supply of infrastructure. The problem is more
complex when recurrent or lasting congestion appears. One solution then is to set the toll or tariff
equal to the long run marginal cost plus a markup to reduce demand. The revenue from the
markup over and above what should be the optimal price can be signficant if it is not dissuasive
enough and could be allocated to finance new infrastructures. Part of the demand may disappear
as a result of this pricing strategy and never return if the prices are not adjusted down again.
16 The underestimation of demand for a mode can mean the overestimation for a competing mode....



10
This also illustrates the importance of serious cost-benefit analysis in projects which detailed
analysis of willingness to pay under various environments.
The problem with the pricing solution is a political one in the context of privatization.
There are many episodes in which toll or fare increased have lead to riots and regulators or
politicians are thus reluctant to engage into pricing changes that are too sensitive politically. A
more general solution that can address this problem is to avoid fixed term contracts with varibale
fares and to award variable term contracts with set fares or fares subject to minimal changes.
Under a flexible duration contract, if the demand is stronger than expected and capacity increase
is needed, the operator keeps the business for a duration that is consistent with the investment
recovery time at a given fare or toll level. This approach is increasingly popular in the transport
sector and is the de-facto outcome of many renegotiations in Latin America as discussed in
Estache (2000).
3.3 Summing up: the needfor accuracy
This brief overview of the various ways of getting demand forecast wrong clearly shows
that there is room for improvement. But it also shows that regulators have at their disposal
various instruments to fix things when they are wrong and better yet to it get it right the first time
around. This is fact often the best solution. Indeed, once the government has decided to rely on
private operators to provide transport services and infrastructures, it is getting into a permanent
relationship with these private operators. Their interactions create a history and both sides build
up a reputation during the development of this history. A government prone to mistakes will
expose itself to tougher negotiation with the private operators and increase the incentive
operators have in contesting regulatory decision on the basis of the doubtful value of their
supporting analysis. A strong regulator with a record of sound, well founded decisions will not
have to give in-complex and tense renegotiations. Moreover, a reliable predictable regulator also
implies a lower regulatory risk. This is turns means that private capital requires a lower rate of
return and is hence easier to attract. 17 In nutshell, even if regulators have instruments that allow
the correction of forecasting mistakes, these corrections are generally not challenge-free. This is
why investing in getting it right to from the start may be the best strategy to begin with for a new
regulator. Since this means considering the construction of demand forecasting models early on
in their regulatory tenure, it is be useful to briefly review the theory and practice of demand
modeling and forecasting.
4. The main forecasting methods
As any basic microeconomics textbook would remind the non-specialized reader, the
demand for a transport service or infrastructure is the economist's jargon to describe the measure
of the willingness to pay for this service or infrastructure. A well documented study of demand
should allow the regulator to track down how the use of these services or infrastructures will
change with changes in the fares or tolls charged by the operator but also with the income level
of its users and with the environment in which these users function. It is quite important indeed
to recognize that the demand for transport is a demand derived from the demand for other
activities because it is an input into these other activities. Roads, railways or ports are used by
productive sectors to distribute products or send staff to represent them. Individuals travel to
work or to visit places. Few travel for the sake of travelling.
17 Alexander, Estache and Oliveri (1999)



11
All this means that a regulator interested in forecasting demand must understand the
various sources of the demand for a specific service or infrastructure and their interactions-i.e.
the demand for metro services is influenced by changes in the bus fare; or a recession reduces
product shipments but may also employment and hence the demand for workers' transportation
as well as the average income of the potential users of the transport services or facilities. This is
quite important when transport privatization takes place in the context of a restructuring
economy in which severe industrial and employment adjustments result in dramatic shifts in the
nature, composition and level of the demand of transport.
While the demand for transport can be studied at the very aggregated level and the level of
expenditures allocated to the transport sector in the macroeconomic accounts provide useful
information on overall trends, it will generally not be sufficient to generate the information
needed by a transport privatization commission or regulator. The analysis of the demand for
transport required to simulate the effects of various combinations of quality of service (travel
time, frequency, coverage, reliability,...) and fare or toll levels or to assess the impact of
regulatory decisions requires a much more "microscopic" study of a specific geographic
network. It will have to be designed to the allow the assessment of the network effects of the
various combinations of prices and quality considered by the government on traffic levels for
each mode for each segment of the network while accounting for a large number of specific
socio-economic characteristics or similar determining factors. In general terms, these factors are
all picked up in the various stages of the classic transport modeling efforts.
The "classical" theory supporting the modeling of transport demand for a specific transport
infrastructure or service has a strong microeconomic foundation, that is a strong concern for an
understanding of the specific factors that influence users of transport infrastructure or services. It
is however framed into a more general framework which reflects the strong influence of more
traditionally engeneering approach to the processing of data. It starts with a clear assessment of
the area to be covered by the transport service under study. The next stage is to divide this area
into various zones and to collect detailed information on each one of these zones. This data
includes population levels, activity types and levels, employment, commercial areas, educational
and recreational institutions,.. . and a detailed analysis of the travel patterns and preferences of the
users. The information is then processed in 4 stages to determine transport demand: generation of
trips, distribution of trips, modal choice-which is where most of the economic underpinning of
demand modeling is introduced--and route selection. 18
Although in theory all types of transport demand can be conceptualized this way, in
practice, however, the approach has been essentially followed for urban transport, and to a lesser
extent for intercity rail passenger travel. Moreover, the analysis of demand for other transport
infrastructures or services has generally focused more on the analysis of modal choices and in
some cases of route allocation, the stages at which the key variable relevant to a regulator such as
quality are explicitly adressed and the stage at which actual demand models are estimated. This
classical model is slowly being adjusted by practitioners to better reflect the activism of users.
When congestion is an issue, a research field is to also model explicitly the preference for the
choice of departure or arrival time of the users. In addition, there is a recognition that the
generation and distributional choices are increasingly complex in many instances as the options
are increasing (probably most importantly the recreational options).
Table I synthetizes the classical steps of demand modelling and the main regulatory each
stage can adddress. Annex 1 provides some insight into each of these steps. The more analytical
18 This approach is most typically associated with the forecast of passenger transport demand but conceptually it can easily be adopted to
describe the various stages of the determination of freight transport demand as well even if there are very few studies which have done so
explicitly.



12
aspects of the various stages of the classical transport model are presented in the Annex inspired
by Ortuzar and Willumsen (1994) . While these stages are still the norm in the sector and many
extensions are still finding their way in the applied litearature, they have serious limitations
which cannot be ignored. For instance, the idea of showing the transport problem as a sequence
of 4 sub-models implies an order which is not necessarily realistic.
Table 1. The "classical"stages of the idenfication of the demand for transport
Stage               Transport decisions         Policy and Regulatory Issues   Modeling
in the context of privatization
Trip generation    How many trips does the user based in Is there an obvious unmet  Land planning
some specified location wants to take  willingness to pay for  and zoning
in day/week/month?                improvements in services
which could be met by a new
project or a concession to
improve existing services?
Trip distribution    Where is the user going with each trip  What would be the optimal size Origin-
among all possible destinations of  of the project to be packaged    Destination
interest to the transport service  for private sector participation?  matrix
provider?
Modal distribution  Which transport mode does the user   What price-quality combination Demand models
adopt for each trip? What are the  should the privatization    -Agreggated
factors influencing that decision and   commission aim at and how  -Disaggregated
to what extent                    much margin should the
regulator give to the private
operator to adjust price and
quality given the overall
objectives of the
"privatization"
Also, how much coordination
is needed between different
modal regulators (if these are at
different government levels for
instance)
Route allocation    6Which route between the origin and   How do pricing (including  Network
the destination does the users pick  access pricing) and quality  simulation models
under various typcs of service    rules influence the efficient use
packages? (most important for roads   of the transport infrastructure?
and rail)
A more modem vision is now increasingly common and recognizes that the behavior of
users needs to be modeled as economist would tend to recommend. This vision must allow for
changes in opinion with changes in transport characteristics (e.g. congestions) and it must allow
for flexibility in the face of very different time and data constraint. As illustrated by the next
section, the most effective practioners seem willing to learn from theory and improve their odds
of successful demand forecasting. This is clear in the modeling of demand in the railways sector
and in urban transport. Some of the sectors seem to be more resistant to make the most of
opportunuties offered by theory focusing instead on increased sophistication in data analysis
through econometrics and expert opinion as is the case in the ports sector and to some extent in
the airport sector (see section A.3. 1. of the annex).



13
5. Demand forecasting in practice
This section describes some case studies to illustrate the ways in which regulators,
privatizers or planners have used the various demand forecasting described in the annex. It
covers examples in which demand forecasting is used in activities which were privatizes as is the
case for the Mexican suburban rail and the French toll roads. It also presents cases in which the
public sector is still very much a key players as for the Spanish airports and the Belgian ports. In
some instances, the experience with the use of modem demand forecatsing techniques is too
recent to be able to do a post-mortem comparing forecast and realized but since to the extent that
they signal a trend characteristic of their specific sector, the exemples are quite informative. To
the extent possible, the process and the degree of use of the information generating by the
forecasting teams is discussed to illustrate the importance of the political decision to use the
information for its relevance in the regulatory process.
5.1 Mexico City Commuter Train Project'9
This case is a textbook approach to demand forecasting.20 It involves a decision in 1998 to
invest in a commuter rail line to reduce pollution and congestion in Mexico city. Over 70 percent
of the Mexico city transportation market traveled exclusively by road in 1994 when out of 20.5
million total weekday trips, buses accounted for 48%, cars 24% and metro 2%. Two or more
public modes were used by 22% of the travelers. The main concern was to forecast how effective
the commuter rail would be in capturing a sufficient share of this market. The demand
projections would be used to increases the transparency of the prospects to any potential private
investor but also to develop the capital and operating program for the commuter service, the
financial analysis which would eventually be used by the regulator to estimate the requirements
of financial support by the government.
The data available was reasonable although not perfect. The forecasters could rely on a
1994 survey of trips origin and destinations in Mexico city conducted by the Mexican National
Institute of Statistics. They also has estimated trip growth rates to allow an update of the matrix.
These growth rates (1.2% average annual from 1994-2030) is based on population growth
estimates and land use trends. Data on the characteristics (trip time, costs, frequency,
reliability...) of bus, commuter rail and metro travel in Mexico collected by a local consulting
firm and a couple of traffic counts at key locations in the corridor. The forecaster then conducted
a stated preference survey of potential commuter rail users to estimate users preferences for
alternative public transportation modes. The demand is based on diversion from other modes of
public transportation only (buses and metro). It excludes diversion from cars (to get a lower
bound) and ignores induced demand as well. The results are then introduced into a model
generating trip volumes for the commuter rail. It includes up to 2030 estimates of annual
passenger demand, annual passenger-kms, demand on weekdays and weekends and boarding and
Based on: Mercer Management Consulting (1998), "Commuter Rail Demand Study for the Buenavista-Huehuetoca Line", prepared for the
Mexican Communications and Transport Secretariat
20 Another, less 'textbooks", set of very relevant illustrations is provided by D. Pearmain (2000), "The measurement of users' willingness to pay
for improved rail facilities", in Ortuzar (2000), op. Cit.. It includes a brief survey of the UK experience with WTP methods and a couple of
examples illustrating some of the implementation difficulties of the stated preference method in the context of a variety of decisions potentially
important to regulators such as the extent which investment decisions in facilities impact on revenues.



14
lightings by station. The consultant also identified the main sources of uncertainty which could
invalidate the forecast.
The stated preference survey had a crucial role. It generated data on the preferences, recent
trips and demographic characteristics of 800 potential commuter rail users in the corridor-
including quotas for time-of-day, age and gender of users. The respondents were recruited at 16
bus stops in or near the corridor from 5 am to 8.30 pm and were offered a fee of up to 200 pesos
to participate in the interview. They were offered the option of being interviewed at their home
or and an interview site next to the bus station. On the interview day, they were guided though a
30 minutes computer-based interview.
The questionnaire was designed to estimate the trade-off between time, cost ad other
service factors. It essentially asked the user to pick among of alternatives differentiated
according to walking time, waiting time, time in mode (bus, rail, subway), total trip time, total tip
cost and total number of transfers. The experiments were customized based on the attributes of
the intercepted tip to ensure realism. The choice experiments clearly show that potential users
place different values on walking time, waiting time, in-vehicle time and transfers between
modes. They value their time at 3 to 9 pesos per hour (roughly UScts25 to 75) depending on the
use of time. In contrast to other locations where in-vehicle time seems to be preferred to walking
time, Mexican seem to generally perceive time spent walking comparable to time spent traveling.
Mexicans prefer to spend time walking to travelling than waiting (as anywhere else). They see
transfers as equivalent to 13-15 minutes of traveling. This means that the speed advantage of
subway or commuter rail transfers is offset by 15 to 30 minutes for 1 to 2 transfers. Overall, they
prefer the commuter rail to bus or metro (but this may be biased by the perception of additional
security).
The network data is used to assign each trip in the total market to a specific mode or
combination of modes based on the user trade-offs between travel time, cost and modes obtained
from the survey and on the characteristics of each model. This is done for the proportion of trips
for each Origin-Destination pair that will travel by each mode of transportation. The
specification of the modal choice model is a nested logit structure with the following mode
combinations: (1) bus only, (2) metro only, (3) metro and bus, (4) commuter rail only, (5)
Commuter rail and bus) and (6) a commuter rail, metro and bus option. The choice of mode is
estimated separately for each trip, taking into account the distinct preferences of the person
making the trip with respect to age, gender and other factor.
The model is calibrated to allow the reproduction of time and cost in 1998 for buses and
metro and generates a commuter rail demand for each average week day assuming that
commuter rail fares are set equal to the bus fares of 0.28 pesos/km. More specifically, the
demand model estimates the probability that a person will choose a particular mode for their trip
for each of more than 90,000 public transportation trips. This information is then used to
estimate demand for each transportation mode for weekday demand. So for any pair of origin
and destination, it computes the total trip time, the cost and the number of transfers and on that
basis it estimated that the user will pick any one of the 6 travel options identified earlier. This
week day demand is discounted for weekends. For the subway, this is done by applying a factor
of 0.58 based on traffic counts. Annual demand is equal to weekday demand multiplied by a
conversion factor for each mode as well based on a traffic count. Prospects are quite good and
are likely to grow even better with an annual weekday demand for commuter rail services of
about 150 million passenger in 2005 getting to over 200 in 2030.
There are some serious caveats also quite important to assess from the viewpoint of a
regulator. There may be a certain degree of diversion from expected traffic due to increased use
of vehicle due to changes in travel habits. On the other hand, congestion may get worse and



15
diversion in favor of public transportation may result in an underestimate of demand. The
behavior of bus services may also have a strong influence since they are quite an important
feeder service. Increases in waiting time for bus services could hurt the demand for the
commuter rail. All these assumptions can be quantified through relatively simple simulations and
allow the regulator to better assess the risk and ... compensations if needed.
5.2 Willingness for urban toll roads in France
Toll roads have been extensively developed in France, since the 60's. Their total length
amounts to roughly 7000 km. But these toll roads are mainly intercity motorways for which
traffic forecasting is a rather easy task as network effects are rather small in intercity roads;
generally speaking there is just one competitive road and the main problem is to assess the split
of the total traffic between the toll motorway and the competitive road. Furthermore, a long
experience exists and traffic models for these situations are well calibrated and provide good fits
with the reality. This is not the case for urban toll roads, which begin to appear in the largest
agglomerations. There are few lessons from the past, and network effects are so large that it is
not at all straightforward to estimate the sensitivity to competition or complementarity between
routes. This is why the effects of the level of tolls on the traffic of the motorways have to be
much more carefully studied than for other types of roads. Currently this is done in France by
relying on rather large econometric models. One representative exemple is now in operation in
Ile-de-France, the region which surrounds Paris and is almost full urbanized with about 11
Millions inhabitants.
The models used to forecast traffic on new infrastructures in this area are classically built
on a four steps basis. Demand data are drawn from household surveys (16 000 households)
updated every 5 years, and provide trip matrixes for several purposes of travel: job, non job, and
for duty vehicles. The network supply is defined by nodes and arcs linking these nodes. Arcs
represents the links between each modes; they are defined by technical characteristics such as
their length, the cost and speed, safety, comfort ... While this is all quite reasonable and standard,
the details of the modeling are not always relying on the frontiers of the applier theory. For
instance, modal choice is taken into account through a very crude model. Also, for traffic
generation and distribution, the O.D matrix is drawn from current data, and induced traffic is
determined as a fraction of cost decrease of each O.D relation.
The traffic model is run first to reproduce the current situation, and its parameters are
calibrated so as to minimize the sum of the square differences between present and reckoned
traffic on each link. Then the model is run with the new planned infrastructure, and it is run
several times, for different values of toll.
More precisely, traffic modelling is developped following the following steps21
Step 0, Before the study : Setting up the traffic modelling framework (the "Modele global
de la Region Ile de France"). This step is done by the regional branch of the Department of
Transport, which usually runs the model ; the underlying work includes gathering data on
supply (network description and encoding), demand (socio-economic characteristics of the
zones into which the total area of the study is divided) and traffic countings. The result is a
21 Drawn from "Etudes de trafic et de sensibilite au peage en Ile de France", internal document, Ministere de
l'Equipement, France (1999)



16
general calibration of the parameters of the model, providing Origin-Destination matrices,
and traffics calculated by the model on each link of the network.
Step ] . Calibration of the modelfor the year of implementation of the planned investment.
This calibration is done by the franchiser, usually the political body in charge of the
Region Ile de France or the Department of Transport. It corresponds essentially to an
updating and a fine-tuning of the previous step.
Step 2: Analysis of how the network works. This step is a zoom of the previous step over
the area around the planned infrastructure. It is implemented by the public authority and
and is designed to highlight the potential problems: where is congestion, what are the
areas with low accessibility indexes... This step helps to determine the precise scheme to
be achieved, for instance the number of lanes, the design of the interchanges...
Step 3 . Definition of the reference scenario. This step is also implemented by the public
authority . It defines the network and the socio-economic characteristics for the various key
years of the study if the planned infrastructure were not achieved. It also identifies the
alternative investments which should be implemented,the regulatory measures needed and
their effects on the growth and structure of traffic.
Step 4: Calculation of the traffic in the reference scenario. The outcome of this step , run
by the public authority or its traffic consultant, is to apply the "modele global" to the
reference scenario. Traffics on each link and for each mode are calculated, along with
travel times, monetary costs...
Step 5 : Policy simulations. In this step the network includes the planned investment to be
evaluated ; the model is run for several key years for which there is a change either in the
network or in the demand. Comparison between steps 4 and 5 allows to set up the main
changes between the reference scenario and the planned scenario. The conclusions are both
quantitative (the rentability of the investment, cost -benefit indexes, accessibility
indexes...) and qualitative (where does congestion remain, what are the possible effects on
the urban development...). Usually the model is run several times, each time with a
different value for tolls. It helps at setting the proper toll.
Steps 6 : Interacting with the bidders. These results are provided to the bidders, who
achieve their own traffic study on those basis. They are allowed to suggest modifications to
the planned investment ; they achieve their own profitability study, and deduce the toll
structure and levels that they include in their bids, as well as the possible subsidies or
securities they ask to the franchizer.
It is clear from this process that the public authority has in Ile de France a good knowledge
of traffics and traffic modelling: it is at the source of the data used in the traffic modelling of the
bidders; it operates his own model which allows to establish a diagnosis of the transport
problems to be solved, and to set up the possible solutions. Its expertise in this matter allows him
to audit the bids, to screen them and to rank them, to discuss with the bidders about the traffics,
the subsidies and the tolls. As far as toll are concerned, the results generated by the model are
quite typical results: as long as toll increase, traffic on the new infrastructure decrease. Revenues
from the toll, which are the produce of the toll and the traffic, follow a Laffer curve--they first
increase up to a maximum and then decrease. The toll which maximizes revenues for this
specific toll road system is rather high, about 1$ per kilometer.
It is possible to calculate a profitability index associated with each toll. The profitability
for the operator is maximal for the toll which maximizes revenues i.e. 1 $ per kilometer. The
collective rentability is different from the former; it takes into account, not only the revenues of
the infrastructure operator, but also the consumer's surplus and the consequences on



17
environment. Collective rentability is maximal for a toll equal to zero a situation where the
highest number of users benefit from the new infrastructure. In practice the toll regulation tries to
reach a toll a bit below the toll which maximize the revenue, in order to get a revenue close
enough to the maximum level , in order to make the motorway just profitable for the
concessionnaire, without driving too much subsidies from the government, while managing the
political sensitivity of the introduction of tolls. Increasingly also France is learning from some of
the conflicts it has gone through between users unwilling to pay and local governments
constrained fiscally. The decisions on the tolls are now set to levels consistent with the ability to
pay of the majority of the potential users.
5.3. Forecasting Port Trafc in Antwerp22
Forecasting techniques used in ports tend to fit into strategic planning programs. This is
important in the context of restructuring (how many ports do you really need in your country?)
but more generally for most common evaluations of port expansion projects (e.g. new terminals,
dock, hinterland projects,...). Most casual observers would recognize that macroeconomic trends
and reforms (trade liberalization, labor market reforms,...) and the specific characteristics of
each location matter but recent progress in logistics techniques and technology are also
revolutionizing the demand for port activities and this needs to be taken into account by port
regulators.
Consider the port of Antwerp. Located about 70km from the North sea, it stretches for
some 20km along the banks of the Scheldt river. It enjoys an international maritime traffic over
1 15 million tons/year, sees over 16,000 ships every year and is one of the five largest ports in the
world. It loads (about 1.6 million TEU), unloads (about 1.6 million TEU), stores, repackages,
distributes and forwards goods. It has dedicated terminals for many commodities and handles dry
and liquid bulk containers. The container traffic is about 35 million metric tons, 61% to roads
and 25% to waterways. General cargo represents about 60% of the traffic. The port employs
almost 70,000 people. It is subject to a strong competition from various ports in the Hamburg-Le
Havre, a critical fact which makes forecasting all the more important.
Antwerp has been forecasting traffic formnally since the late 60s. The Transport Economics
Department of the University of Antwerp has been very active in helping the port authorities
with the management of the port in general but in particular with these forecasting efforts. This
is coordinated in fact by a special unit of the port, the Study centre for the Expansion of Antwerp
(SEA). SEA has generally had a very clear preference to prepare a base forecast to be
complemented by more detailed information and insights to be provided by experts. This
approach is quite common in OECD ports and this is why it may be worth to describe it in some
detail.
The whole exercise is based on a base model used to forecast for two main commodity
groups, bulk traffic and general cargo. An optimistic and a pessimistic variant differentiated by
the forecast of macroeconomic conditions in the OECD countries are worked on for a series of
specific traffic group. Bulk traffic is divided into 10 standard categories (agriculture, food and
food stuffs, solid fuels, crude oil and products, minerals and ores, metal products, raw materials,
fertilizers, chemicals and machinery and vehicles not included elsewhere). Initially, for general
cargo, loadings and unloadings, were pooled together and an aggregate figure was forecasted.
22 This section is based on , Coeck, C, A. Verbeke and W. Winkelmans (1996), " Back to the Futute: an Evaluation of Traffic Forecasting
Techniques used in Antwerp and Rotterdam", pp. 3-26 and Verbeke, A, E. Declercq and D. Teurelincx (1996), "The future growth of contained
traffic in the Port of Antwerp", pp 51-70, both in Winkelmans, W. and T. Notteboom (eds), The issue of prognoses in maritime transport and
seaport production in Belgium and Poland: a conceptual and empirical study, Garant, Leuven, Belgium



18
Since 1985 they have been disaggregated and analyzed for both containerized and non-
containerized traffic. The analysis of the disaggregation of cargo looked into imports and exports
figures for the country, transshipments and transit figures and origin/destination relations.
Forecast consists essentially of extrapolations provided by an econometric modeling of the
historical traffic. The main explanatory variables for the future of traffic are taken to be: (i) the
general business cycle and trends variable in Belgium (the evolution of the share of exports and
imports in GDP and the evolution of industrial production) and a similar variable in a large range
of trading partners (an index of GDP in the EEC, the USA and Japan), as well as international
trends in containerization, the specific variable depends on the traffic to forecast; (ii) the
competition (remember that there are many competing ports close by); (iii) the microeconomic
features that make the port interesting for specific traffic (it is the second petrochemical port in
the world for instance).
The 1996 Masterplan for the port shows how the quantitative part of the forecast was made
for every major traffic groups. For general cargo, the variable used to pick up the business cycles
and trends is an index of GDP in the developed world (EEC, USA and Japan). For bulk cargo,
the driving force is assumed to be industrial production in the EEC. For both cases, the
importance of competing ports in determining prospects is taken in the Hamburg-Le Havre range
which covers all the major competing ports. It is summarized into Antwerp's market share in this
corridor. The plan also presents a qualitative forecast. It covers an assessment of general cargo
traffic disaggregated into specific commodity groups (iron and steel, non-ferro metals, fertilizers,
wood products, vegetable and fruits, grains and animal fodders, rolling material, flowers, sugar,
food and other. For bulk traffic, the experts are asked to consider ores, solid fuels, grains and
animal fodder, fertilizers, sand, crude oil and chemical products (including petro-chemical).
Overtime the quantitative models' performance disappointed and it became increasingly
used only to supplement a more qualitative approach. Indeed, the quantitative model was
overestimating quite significantly (between 16.5% for general cargo and 37.3% for total
seaborne cargo). The qualitative method generated an underestimate of 1.6% for 1985 when it
was first used (with a deviation of +9.1 % to -14,6% for different targeted activities). This is why
the qualitative approach is in fact used to generate information which is then built-in the more
aggregate quantitative forecast of the demand for port services (in terrns of tons) .
But forecast modeling is an evolving science also in ports and the amazing growth in
containerized traffic is generating interesting progress in the field as seen in the case of Antwerp
as well. The initial step is to estimate how the overall macroeconomic prospects influence the
overall general cargo traffic (defined as conventional cargo traffic + container traffic + roll-on-
roll-off traffic) in the competitive region (i.e. the Hamburg-Le Havre market segment). This is
taken to follow a simple trend in which an increase in 1 point in an index of the world GDP
results in an increase of 2.307 in general cargo traffic in the corridor (but this effect is reassessed
every year using the latest information available on the linkages between the two variables). The
future containerization rate is the next aspect forecasted. It is also an extrapolation but it
recognizes that the degree of containerization is limited and the growth will eventually start to
decrease. This drives the choice of the functional form to a semi-logarithmic function (i.e. the
degree of containerization (Y) is explained by the logarithm of a time trend (T): Y=.61+.04 InT)
but the specific parameters of this equation are picked by experts rather than simply estimated
analytically from past data . Jointly with the first step, it yields an overall regional
containerization forecast which is an average figure since the experts recognize that its evolution
will vary for the various departure/destination combinations. This is then used to generate a
traffic forecast for containers in the corridor. In 1996, it was expected to increase from 140.6
million tons in 1995 to 184.5 in 2000 and 227.5 in 2005.



19
The assessment of the market share of the port of Antwerp is then needed before coming
up with its container traffic forecast measured in Twenty Feet Equivalent Units (TEU), the
standard measurement unit for container traffic. For the 1996 Master plan, the working
assumption was that the market share of the port of Antwerp would remain the same at 18.4%.
This results in a forecast of traffic increase from 25.8 million in 1995 to 33.9 in 2000 and 41.9 in
2005.23
To most purist, this is a very subjective approach to forecasting. But it is worth recognizing
that this is how it works in many places. Most port projects which these authors have had to
review followed the line described here. It is obvious that the analytical part of the project is
weak. The quantitative methods used did not work but it is very likely that they could be
improved with recent development in the analysis of time series. In the short run however, the
proof is in the success of the forecast. The forecast reported here were made in 1996. They
resulted in an underestimation since actual total container traffic had reached 35.4 million tons at
the end of 1998 already.
5.4. The forecasting of Traffic in Spain's Airports 24
In Spain, the forecasting of air traffic is done by a public entity, Aeropuertos Espafioles y
Navegaci6n Aerea (AENA). The process is itself revealing and involves many of the approaches
discussed earlier. Demand is forecasted quite technically at the airport level but the aggregate
demand is the result of meetings of all the airport managers who decide collectively on the future
investment needs and allocation. In general, the demand forecast distinguish between short run,
medium run and peak traffic. Peak traffic forecast is needed to know when and what to do with
prices in the short run to adjust demand to capacity. These estimates are compared with estimates
made by users of the airport services to test for their robusteness.
Short run forecast is based on time-series techniques-moving averages, exponential
smoothing and ARIMA models. They are designed to generate forecasts learning from past
forecasting mistakes. They generate monthly estimates for a 12 months period for national and
international traffic for all airports. Medium and long run forecasts are based on econometric
models which reflect the complex interactions between mutliple variables. These are used to
make forecast to 5, 10 and 15 years horizons. Figure 2 shows how AENA proceeds to come up
with these estimates. This is done every year during the first quarter of the year.
Figure 2 shows that the forecast is based on past traffic trends but also on regional, national
and international determinants of business cycles but also on the tourist traffic, the possibilities
of substitutes offered by other transport modes and airport specific characteristics including the
number of airlines present in each airport, their capacity, the size of the planes reaching each
airport and their degree of occupancy. It also shows that the specific forecasting process is
actually done in two stages.
23 This is equivalent to about 3.1 million containers measured in TEUs (divide the volume expressed in 1000 tons by the average tonnage of
TEUS (twenty feet equivalent unit) that were loanded and unloaded)
24 This section is based on Trujillo and Jacob (1999)



20
Figure 2: The process of forecasting demand for the medium and long run at AENA
Indicators reflecting the
SoRegional, national and                                     Previous year's
inrstfrecasntl nrtn onrt a      e          passenger             ta/ traffic
\       ~~~traffic I /
Indicators reflecting the supply  | _ ix model  iA/|Impact of transport substitutes           l
of tourists (number. hotel  <
* nauionnav nvllrnare ztsv  |
mt pacOMtofa PBepth      regnumber of airlines
PAXINTt   kPAXINTt I t   PlBes nt PIBinteach aiUport
SuXDandpplXINT sar norrectea   Passenger               f  Impactofairport characteristics
repectuve ly and TURt whc efet the~ nm         er ftuit,tedge   focpnyo   oesh
Average plane size              /
Average occupancy rate 
Ina eon  sae,th   eglao  cnsrctV  model   tofrcsJh   ipaetafcfrec
Source: Aena (I1999
The first forecasts national and interational passenger traffic independently as follows:
A national
PAXDOMt = kpAXDOMt _ I cc P_IBespt /3 PIBregt 9
• International
PAXINTt = kPAXINTt - I T PIBespt 0 PIBint " (TURt Lv
where,
PAXD and PAXINT are national and international passenger traffic respectively; k is a constant
factor and cc 8 d z, 0, 2 ,u are the elasticitis to the key explanatory variables which include
PIBesp, PIBreg, PIBint the variables reflecting national, regional and international GDP
repectively and TUR which reflects the number of tourists, the degree of occupancy of hotels, the
average stay duration, .......)
In a second stage, the regulator constructs a model to forecast the airplane traffic for each
destination, relying on the passenger forecasting models just described:
A4ER, f,= PAXj  ,IVTA4j, ,FOi, 1)



21
where AER is the volume of airplane traffic, PAX  is the volume of passengers traffic estimated
in stage 1, TMA is the average size of airplanes, FO is the occupacy rate, j is the traffic
segment-national or international. The model is quite effective. Forecasts done in 1997 for
1998 was only 0.13% off.
For the longer run (over 15 years), AENA relies on the Delphi technique, . the subjective
method described earlier. First, AENA selects a representative sample of experts to consult. Next
it prepares an initial questionaire submited to these experts for comments. The corrected
questionaire is then sent to the experts to obtain their first individual estimates.These experts are
then fed the results and the process is reiterated as long as needed to obtain a convergence of
views.
AENA also makes forecasts for peak demand. This is important because AENA tries to
minimize the number of peridos during which capacity is idle. In order to design its pricing
policy to meet this objective, it needs to have a reasonable forecast of the distribution of peak
demand over time and space. Both for passengers and airplace peak demand estimates, the model
estimate related peak demand to annual traffic in the following way:
(PHP IPT)%  = K(PT-)
where, PHP is the number of passengers at peak time and PT is the number of total passengers.
This aspect is unfortunately not very successful because the airlines are not too concerned
about it. Indeed, the share of airports costs in total costs is quite small for most airlines.
Moreover, for airports with a strong tourist clientele, the demand is quite insensitive to price and
this makes it very difficult to rely on prices to adjust to the "use of the capacity" to demand.
Excess demand and excess capacity continue to be a problem in practice (AENA, 1999).
6. Practical tips to come up with a useful forecast
Generally, forecasting in transport is based on a combination of various methods. Table 2
summarizes the main steps to follow in picking a method. This is of course a very rough set of
tips but they are based on common sense for situations in which most regulators are faced with in
developing countries.
Table 2 . Tips on getting rough forecasting right
Before study
* Past trend analysis
* National and international economic developments
* Competition from close substitutes
* Forecast made by transport service companies
* Expectations of traffic published by complementary industries
* Forecast of peak periods
* Technical development of transport means
• Check out the budget constraints for the demand study
* Check on data availability to decide how much to invest in data
After the study
* Comparison with traffic experienced in similar situations
. Elasticities of traffic to prices
. Achieve sensitivity and risk analysis



22
6.1 Before the study
To begin with, it is generally useful to have look at the past trend of the type of trip the
analyst is expected to forecast. Quite suprisingly, this can be a challenge as the quality of data
available from transport ministries has tended to deteriorate over the years. Assuming that
enough data can be collected, it it is generally not a good idea to stop there.It is worth finding out
more about the economic context in which the transport project under regulatory scrutiny is
operating. The national and international environment often have an impact of these projects. So
do the development of alternative transport options which can be a complement or a substitute.
This can help the regulator decide if a quick decision is necessary. In West Africa, Senegal, Mali,
Ivory Coast and Guinee currently have a portfolio of competing rail and road projects for which
the rate of return and hence the possibility of attracting private financing is directly related to the
speed with which the governments willmake the decision. Demand will go and mostly be locked
in the hands of the first movers. Many of the key decision makers have a hard time appreciating
the importance of speed for the demand for their transport services because they are not paying
enough attention to what their neighbourghs are doing. The surprising fact is that this
information is relatively easily available to a regulator or a policymaker since most of these
projects are generally government initiative and are quite widely publicized. Moreover, many of
the transport service companies interested in these deals have at least preliminary forecast which
can provide good insight on the real prospects of a proposed investment project. It may be worth
for regulators to find out more of these and to have them audited independently. Finally,
complementary industries such as aeronautic or car builders also make their own forecasts for
obvious reasons. All these are useful in educating a regulator in an initial stage.
Once these indicators of trend or prospects are internalized, it is often useful to try to
identify potential peak demand resulting from exceptional circumstances which may justify fast
increases in capacity than these trends would require. A city trying to sell itself as an great host
for conferences must have the supporting transport infrastructure to deal with associated peal
demands. Tourists airports also face a strong seasonality in peak demand. How much of these
peaks the government is willing to address through increases in capacity and how much by
allowing occasional congestion is a local choice--and it may make sense for a government to live
with occasional or even seasonal congestion problems rather than overinvesting under tight
budget constraints. The fact is that the regulator must be aware of the relevance of these
occasional demand shocks in its assessments of the performance of the operators. They will often
have the option of considering the introduction congestion pricing. If demand is in fact very little
sensitive to prices in peak periods, it may pay off not to adjust capacity and to rely on prices to
deal with peak demand.
Next, the regulator must ensure that technology developments are not interfering with
demand forecast. For instance, the appearance offastferrys reduced the demand for conventional
ferrys and this could result in unexpected shifts in demand as well as in financial drama for the
operators of conventional ferrys. In the crossing of the Channel bewteen France and the UK, the
introduction of new fast boats has resulted in a loss of traffic for trains riding through the tunnel
and contributed to the well publicized financial troubles encountered by the tunnel operators.
Similarly, expansions in capacity can also influence demand. For instance, the introduction of
post panamax ships with a larger capacity to carry containers changed many investment
decisions in ports since new technological constraints were appearing.
Finally, the regulator must look at its budget envelope. Most modern forecasting methods
are quite expensive and to do a decent job, it is not uncommon to have to spend over $100,000



23
on a demand study. How to pick a method also depends on this, as much as on the data
availability and the type of mode the regulators is focusing on.
Assuming that money is not an issue, in addition to the specific traffic type to be
forecasted and the specific type of nfrastructure concerned, the ideal model will depend on a data
availability, on how much detail is needed and on the specific goals of the forecast. Data
provided by the historical operators of the service will help. For the most aggregated data this is
generally possible but for more detailed data this can be a problem. Indeed, many operators will
consider a lot of the relevant information to be confidential and would see as a threat to their
commercial viability the obligation to disclose some data on their markets and clients. But on the
other hand, precisely in the context of privatization, demand studies should be undertaken with
the historical operator in preparation for the valuation of the assets to be "privatized"; this will
also constitute a good initial data base for a regulator. In most countries, a useful compelemtn is
data published by public (statistical offices) or private entities monitoring the sector. It is
surprising to causal observers the number of think tanks or private survey firms collect and sell
relevant information collected through regular surveys. Finally, all this can be compelemented by
data produced through field surveys. The enthousiams for revealed and stated preferences is also
generating vasts databases which often end up under a pile of dust after their initial use. A better
organization of data bases should allow these surveys to become good sources of information to
any regulator. Ultimately, the choice of methods will be driven by the data availability and for
many of the most promising methods they new surveys are the way to go and do not require
more than a couple of months to collect field data.
While data matters, the amount needed depends often on how much details are really
needed from a regulatory viewpoint The devil can often be in the details but the cost of a survey
is directly related to the level of detail that is needed. Of course, this depends on the questions at
stake . If the concerns of the regulator is the potential payoff from the addition of service
requirements such as a toilet or a business center in a train station, there are not too many
alternatives to targetted field survey. But in many more general issues regarding prospects, fairly
aggregate data can be sufficient as seen in the case of the port of Antwerp. The fact is that often
the specific concerns are not well defined and this can results in wastes of time and resources.
Targetting the issues also allow a better targetting of the data requirements and hence the design
of demand studies and the choice of a methodolohy that meet their purpose in the most cost
effective way.
6.2. After the study
Once the study is done, the job is not over. First, it makes sense to ensure that the validity
of the results of the study can be assessed. A first benchmark consist in a comparison with th
results experienced in similar situations elsewhere. It is of course difficult to find exact
comparability but the "distance" between the situations can be evaluated and provide an idea of
the "distance" between the results. It is often useful to request from the consulting team doing the
analysis to provide a survey of their own studies or similar studies. This information should
include a review of the elasticities of traffic to prices and cross-price elasticities when substitutes
or complements are involved. In this context, conventional wisdom can also help and surveys can
be consulted to provide first order estimates.25 Last but not least, the study should provide a point
estimate of course, or at least a narrow range. But it should also indicate how is the traffic
modified under different assumptions for the exogeneous parameters, such as value of time,
25 For instance Oum,Water and Yong (1992) and Small and Winston (1999)



24
economic growth, petrol price, rate of exchange...The specific number of parameters should be
picked by looking at the number of likely regulatory issues that are likely to emerge and for which
regulator is likely to have to set conflicts. Safety, service quality and environmental issues are
likely to arise and hence making sure that the appropriate elasticities are provided as part of the
study for future uses.
7. Conclusions
Forecasting has long been a challenge. It will continue to be one for the foreseeable future.
But the analytical instruments and the data processing capabilities provided by the latest
technologies and softwares should allow much better forecasting than is observed in transport
ministries or regulatory agencies. Privatization induces new needs for demand forecasting; it
leads to pay more attention to risk than it is the case when investment is publicly financed.
Furthermore the regulator has to be able to judge the traffic study made by the operators, and to
find out the strategic behavior which influence these studies. For many governments and
regulators the decision to avoid good demand modeling is driven by a lack of conviction that
theory and models can do better than the "old-hands" of the sector. But this is particularly
dangerous in situations in which the nature of the business is changing as a result of
privatization. Another argument is that it is too costly. The fact is that for projects adding up to
over US$ 100-200 million, a cost of US$ 100-200,000 should not be a reason to reject some effort
to do reasonable modeling. In some cases, the forecasting firms are willing to sell their forecast
with an insurance in case of significant gaps between forecasts and occurrences! These recent
developments have to be balanced against the fact that bad forecast can result in rationing of or
excess capacity which are not costless either. Subsidies are often a necessary outcome.
Moreover, they can ruin the credibility of a whole concession program: a much costlier but often
underestimated consequences of demand forecasting failures.
References
AENA (1999), "Metodologia para el Calculo de la Prevision de la Demanda del Trafico Aero en
los Aeropuertos Espanoles", Direccion Corporativa de Planificacion, Division de Planes
Directores, Madrid, Spain
Alexander, Estache and Oliveri (2000), "A few things transport regulators should know about
risk and the cost of capital", Utilities Policy, forthcoming
Ben Akiva, M. (1985), Discrete Choice Analysis: Theory and Application to Travel Demand,
The MIT Press, Cambridge, MA..
Centre d'Etude sur les Reseaux, les Transports, l'Urbanisme et les Constructions Publiques
(1998), Comportement de deplacement en milieu urbain. les modeles de choix discret
Coeck, C, A. Verbeke and W. Winkelmans (1996), " Back to the Future: an Evaluation of Traffic
Forecasting Techniques used in Antwerp and Rotterdam", pp. 3-26, in Winkelmans, W. and T.
Notteboom (eds),
Crampes and Estache (1998), "Regulatory Trade-offs in Designing Concession Contracts for
Infrastructure Networks", Utilities Policy, 7, pp.1-13



25
De Rus, G. and G. Nombela (1999), "Variable Term Concessions for Highways: Are the Most
Efficient Concessionaires Always Being Selected?", mimeo, The World Bank Institute.
Washington D.C.
Engel, E., R. Fisher and A. Galetovic (1 997a), "Highway Franchising: Pitfalls and
Opportunities". American Economic Review Papers and Proceedings. 87:2. pp.68-72.
Engel, E., R. Fisher and A. Galetovic (1997b), 'Privitizing Roads: A New Method for
Auctioning Highways'. Viewpoint Note, N. 112, The World Bank: Washington D.C. Reprinted in
S. Smith (ed.). The Private Sector in Infrastructure. The World Bank, Washington D.C. pp.109-
112.
Engel, E., R. Fisher and A. Galetovic (1998), "Least-Present-Value-of Revenue Auctions and
Highway Franchising". NBER Working Paper 6689. Cambridge
Estache, A. (2000), "Privatization and regulation of Transport Infrastructure in the 1990s:
Successes... and Bugs to Fix for the Next Millenium", World Bank Policy Research Working
Paper No.2248, forthcoming in the World Bank Research Observer
Galvez, T.E. and S.R. Jara-Diaz (1998), "On the social valuation of travel time savings",
International Journal of Transport Economics, Vol. XXV, No. 2, June
Gomez-Ibanez, J., W.B. Tye and C. Winston, ed. (1999), Essays in Transportation Economics
and Policy, Brookings, Washington, D.C.
Jara-Diaz, S.R. (2000), "Allocation and valuation of travel time savings", mimeo, Universidad de
Chile
Lefevre, C. and J.M. Offner (1990), Les Transports Urbains en Question, Edition CELSE, Paris.
Mercer Management Consulting (1998), "Commuter Rail Demand Studyfor the Buenavista-
Huehuetoca Line ", prepared for the Mexican Communications and Transport Secretariat
Ministere de l'Equipement, France (1999), "Etudes de trafic et de sensibilite au peage en Ile de
France", internal document
Ortuzar, J de D. (2000), Modelos de Demanda de Transporte", Segunda Edicion, Alfaomega
Grupo Editor S.A., Mexico
Ortuzar, J.de.D. and L. G. Willumsen (1994), Modelling Transport, Wiley, New York
Oum, T.H., W.G. Waters, II and J.S. Yong (1992), "Concepts of Price Elasticities of Transport
Demand and Recent Empirical Estimates", Journal of Transport Economics and Policy 26
(May), 139-54
Pickrell (1992), "A Desire Named Streetcar- Fantasy and Fact in Rail Transit Planning", Journal
ofAmerican Planning Association, Spring



26
Piron, V. (1999). "Urbain-Interurbain: La Problematique est devenue globale". Transport. No.
393. Jan., Feb.
Quinet, E. (I1998), Principes d'Economie des Transports, Economica, Paris.
Skamris, M.K. and B. Flyberg (1997), "Inacurracy of traffic forecasts and costs estimates on
large transport projects", Transport Policy, Vol. 4, No.3, pp 141 -46
Small, K.A. and Winston, C. (1999),  "The Demand for Transportation: Models and
A4pplications ", in Gomez-Ibanez, Tye and Winston
Tirole, J. (1997), "Commentarios a la propuesta de Engel, Fisher and Galetovic para las
concesiones de careteras", Estudios Publicos, No. 65, pp201-14
Trujillo, L. and Jacob, M. (1999), "Prediccion de demanda para infraestructura de transporte en
Espana", documento de trabajo, Departamento de Analisis Economico Aplicado, Universidad de
La Palmas de Gran Canaria
Verbeke, A, E. Declercq and D. Teurelincx (1996), "The future growth of contained traffic in the
Port of Antwerp", pp 51-70, in Winkelmans, W. and T. Notteboom (eds)
Ville, J. (1970), "Le paradoxe de la congestion", mimeo, Ministere de l'Equipement, Paris,
France
Winkelmans, W. and T. Notteboom, eds. (1996), The issue of prognoses in maritime transport
and seaport production in Belgium and Poland, a conceptual and empirical study, Garant,
Leuven, Belgium



27
ANNEX 1: The four stages of the classical model and their meaning for regulators
A.1 Trip generation models.
There are two main types of trip generation models. The first is the aggregated one which
is essentially what the first generation of demand modelers focused on at a very aggregated level,
the second is the most disaggregated approach which tries to model more explicitly the
individual preferences. The aggregate model often provides a useful first order approach which
eventually needs to be confirmed by more detailed studies.
A. 1.] The Aggregate Approach.
The aggregate approach determines the average number of trips for the population
concerned by the project and some type of extrapolation of that information is used to
forecasting medium to long run demand. This extrapolation has taken increasingly sophisticated
forms and accounted for an increasingly large number of factors to explain future trip generation
and hence demand. In general, the number of trips on any corridor between a specific origin and
a specific destination (i.e. a road project or a rail project) will be explained by variables such as
the level or type of activity or any other relevant general characteristic of the population.
Consider the case of a railway company.
The general analytical form is expressed as follow:
T, = k,rjuaf,
in the case of a project concerning mostly passengers, these variables would mean
Ti   the average number of trips taken the potential transport users during a given unit of
time (day/week/month) is the average number of trips per unit of time of the
population of the city or neighborhood i.
ri  is the average per capita income of the potential transport users
fJ   can be any additional relevant factor such as the ownership of a own vehicle in the
cases of urban transport models,
k. a et b   are parameters
In practice, this provides a useful first order idea of get an idea of the future demand.
Extrapolating the values of r and f gives a first order of magnitude. It easy to come up with
because a lot of the required data is relatively easy to come by. In the case of trains for instance,



28
most public enterprises record a significant volume of data on volume carried, number of trips
and the main characteristics of the largest clients for freight passengers or reasonable surveys of
their client base. It is generally not sufficient to do much more than a valuable input into a more
sophisticated analysis of prospects.
A. 1.2 Disaggregated models
The second type of model is disaggregated and focuses on the individuals trip generation
characteristics. The users' population and its movements are divided into homogeneous
categories within which there is some degree of forseeability in the trip types and numbers. For
instance, residence-office, work-related trips, private trips for private business, week end
leisure trips and holidays leisure trips. The users of the transport service can then be classified
according to more specific criteria: income, household composition and size, age, ownership or
not of a vehicle, employment type and location characteristics.
The distinction into these various categories is important since it determines the various
elements a private operator may consider in discriminating prices across users and usage types,
explicitly through the design of tolls or other prices, or implicitly, through targeted discounts to
frequent users. From a mode political view point, it is also important to use this kind of
information to ensure that the poor are not penalized as users. 26
From a regulator's viewpoint, their value is quite significant in particular in an urban
transport context since it provides a snapshot of the users' constituency of this regulator. This in
turn is likely to be critical information in most instances of conflicts between service providers
and users. Indeed, the valuation of the service will significantly be determined by their individual
characteristics and differences across users are important to understand for any referee or
arbitrator which the regulator often is.
A.2. Trip distribution models
The distribution is the stage allocating traffic from a center to the various possible
destinations and its use in the context of specific projects considered in the context of
privatization is mostly concentrated in urban issues. An intercity road or rail project have
In practice, these models suggest great constancy of total daily urban or suburban trips (on average 3.5 per day per person for the last 20 years
around Paris and between 3-2 and 2.4 in the country side). The main change is taking place in the modal choice with a significant increase in the
use of cars. Motorized trip vary between 1.7 and 2.1 per day-increasing with the distance from the center (LEFEVRE et OFFNER (1990)).
Also, while these studies in France also show an increase in the average length of the trip, the speed also increases.



29
essentially one destination, even if it possible to consider the inclusion of feeder services for
other regions or connections opportunities. In many developing countries, the value of a road
considered for private tolling depends on the existent of a supporting public network to enhance
the commercial value of the private road. In many contracts in fact, concessionaire will demand
the construction of this feeder road and the failure to do so is a reason for contract renegotiation
(see Malaysia example). This is why the overall vision of an origin distribution important is
generally important, even if it is true that it is significantly more important in an urban context.
Conceptually, an origin-destination matrix is almost always derived from a gravity model
which can be justified in many ways and specified in almost as many forms. The oldest and best
known is:
T~k(Pi pi
and: - Tij is traffic from i toj
- Pi et Pj are the potential users in i and j (e.g. population, number of firms)
- dij is the distance between i and j
- a et b are parameters
Some of the problems of this model is that its does not address differences in socio-economic
profiles or differences in the economic nature of the firms covered (industrial, service,...?).
In practice, since most estimates suggested that the value of a is I and for b around 2, a
more general formula is:
7, = kO, Djf (co)
and: - O is an "emission factor with origin i
- Dj is an attraction factor by destination j
-f is a decreasing function
- Cij is the generalized cost of trips between i and j
An unconstrained version of this model allows a good estimate of the elasticity of demand
to transport costs and deals with both trip generation and distribution. It is quite common for
inter-urban traffic studies. But the model can be subject to various types of constraints (e.g.
limiting the total traffic and some aspects of the traffic exogeneously or limiting total costs). The
challenge is to specify a model that is well adapted to the situation to be analyzed. A wrong
specification of this part of the model can be quite dramatic not only in terms of quality of



30
service to be provided to users but also in terns of the possibilities to recover investment costs or
imposing additional congestion costs to users. Models subjects to double constraints (both traffic
exiting each zone and entering each zone are set exogeneously) are well adapted to urban traffic
and suburban residence-work trips. The single constraints models (the sum of traffic exiting
each zone is set) is used for all other trip purposes.
A.3. The modal choice
This is the key stage from the viewpoint of demand modeling. The underlying idea is that
in many cases, demand for a specific mode is increased by reducing the demand for another
mode. The modeling of modal choice is based on the assumption that this choice depends on the
difference-or sometimes the ratio-between transport costs. The general form in the case of the
choice between two modes is a-logistic-function which gives the probability of relying on one
mode as a function of the difference between these two costs. 27:The intuition is thus quite
simple, the users goes for the most cost effective way of getting the kind of service needed.
While other formula can generate the same type of outcome, this one has the advantage that it
can be derived from an explicit modeling of the economic behavior of an individual-a discrete
choice logit model. The real challenge is to decide whether to implement the model at the
aggregate or disaggregated level.
A. 3.1 The aggregate models
The aggregate approach uses very aggregated information to describe the behavior of large
groups of users (passengers, tons of products transported between two cities in a given transport
mode), generally over time (dynamic concerns) and sometimes across regions (geographic
concerns). The behavior described is thus the average behavior without consideration to the
relevance of the diversity of behavior. The analyst focuses on data made of an origin-destination
pair and on the share of users relying on a specific mode in the total population. The demand is
eventually assessed by recognizing that it depends on the usual relatively standard factors such as
prices, the price of substitutes, quality and relevant socio-economic variables taken from
macroeconomic accounts. This kind of model is for instance used to assess the distribution of
Pr(l) =    I      ,where:Pr(l) is the share of use of mode I, Cl et C2 are the generalized costs of each one of the two modes-the
I+ expu(C, - C, )
generalized cost isthe sum of the cost of using the mode and of the value of the time needed to get from origin to destination and p is a positive
parameter which drives the sensitivity of the modal choice to the differences in costs.When the value of p is infinity, the choice is one of all or
nothing. When it is 0, the modal choice is insensitive to costs.



31
traffic between day and night by railways companies or to model allocation between public and
private vehicles in urban settings. These models are generally too aggregate to meet the needs of
a regulator but they are used anyway because they can provide useful initial insights. What
matters from the viewpoint of a regulator is that the cost account for many of the variables that
will be subject to regulatory supervision such as tariff/tolls or government control such a fuels
but these are seldom take into account in a sensitive enough way to be useful because they do not
account well enough for the different characteristics of the various types of users.
The main aggregate methods can be classified in 2 groups: qualitative or intuitive methods
and "mathematical or statistical" methods. Among the qualitative methods, it is worth pointing to
a series of techniques based on expert assessment or field surveys of expectations designed to
complement the quantitative estimates. The expert assessment is least refined since it assumes an
expert can assess the impact of mutliple factors on a potential new business environment-since
in the context of privatization, new business opportunities are typically being created.
Field surveys of expectation are targeted to anyone who can be affected by the project and
is in a priviledged position to assess the prospects. A refined version of these surveys is provided
by the techniques called Delphi, where the experts forecast and get a feed-back on the results
from the full sample of people surveyed allowing a series of iterations until some degree of
convergence is achieved. This approach is used by IATA or EUROCONTROL for instance for
airline traffic forecast. These techniques can be useful at the national or regional level but are
unlikely to be useful to forecast traffic and traffic risk on a given corridor. Finally, the technique
of brain-storming, quite common in socioliogy, consists in putting together a group, for instance
users, and through in a series of question to be discussed by the group. This technique is used by
Puertos del Estado in Spain complement the more analytical forecasts of the spanish ports.
The more quantitative methods to forecast aggregate demand can be classified into: trends
extrapolations-assuming a linear, exponential, quadratic or some other function that seems to fit
past observations well--, fancier time series analysis and more sophisticated econometric work.
Extrapolation is the weakest but also the most commonly used around the world. The future is
assumed to be determined by what happened in the past. It is mostly useful for very short run
needs and has a better chance of working on monthly data than on annual data. But this is



32
limiting since for many sectors it is very likely that at some point changes in trend will occur.28
The main problem is that with monthly data, seasonal effects are likely to arise and better
techniques are needed. Some marginal statistical improvements (moving averages over several
months to smooth the series from its seasonally or a better analysis of the statistical model
explaining the growth rate, e.g. exponential vs. linear) are common for short term forecasting. At
best these methods can be used to inform or complement more sophisticated work.
Time series statistical methods (e.g. Box-Jenkins) are a major improvement over these
approaches because they use the information more effectively and can be used to explain trends
through a much more detailed assessment of the behavior of time series over time. They have a
weak economic theory foundation and hence are more difficult to use in the context of structural
changes such as a privatization. They do not leave much room either to model incentives or
regulatory influences on patterns and trends.
Statistical methods can be divided into models of growth factors and econometric
modeling. The growth factor models assume that the demand for transport is quite simply and
directly related to specific economic indicators. Statistical correlations provide the basis of the
identification of the indicators that need to be picked. Assumptions on the growth rate of these
indicators drive the forecast of the demand for transport. In general, the best inputs are
indications of the growth in physical outputs that will need to be transported. When these are not
available, income, employment or population provide approximations of the driving forces. Once
more, this mode is much too aggregated to be useful to most regulators or privatization
commission. It does not do much more than provide a glorified trend.
The aggregate econometric models are probably the most useful tools in this category. They
allow the recognition of the fundamental relations between the demand for transport and a large
number of economic, social and political factors. They can be based on studies of historical time
series, cross sections (i.e. a number of cities, a number of users at any point in time) or panels of
data (which tracks the evolution over time of a cross section). For readers unfamiliar with
econometric techniques,
28 This can be picked up by a modified exponential (traffic = k. + a.b i...e with a<O and b<l), a logistic (I/traffic = k. + a.blif'e ) or



33
A.3.2 The disaggregated models
While the aggregated model focuses on the average choice for a given origin-destination
pair, the disaggregated models focus on the specific choice of each users. The data observed is
the choice of an individual user and the "pleasure" ("utility" in economics jargon) generated by
each mode is reflected in a (logit) function which related this pleasure to the price of the mode,
time spent and any other relevant characteristic of the individual or mode.
u, E pt XI  E 29
and: i: is the individual,j: is the mode, k: represents each of the k relevant characteristics of
the pair (i, j) for each individual (while for aggregated models, the average of the population is
enough), A, : are parameters and E is simply an operator saying that the "utility" is determined
by the sum of all the relevant variable weighted by the parameters estimated from the samples
surveyed and e, :are random shocks influencing utility which need to be modeled explicitly.
The specification of these random shocks is what differentiates models and determines the
specific econometric technique that must be used to derive demand. They often depend on the
nature and the degree of precision with which the data is collected. The lack of realism of the
specification can be a problem in practice and generate counter-intuitive results as discussed in
Quinet (1998). This is why the way the information is collected is something the regulators must
under as well as possible. It will drive their ability to be a fair referee.
The basic information can be collected in two main types of ways: revealed preferences
and declared/stated preferences. Revealed preferences are provided by a direct observation of
the behavior of transport service users. They do what they prefer to do. This means that the
regulator only needs to get data on the current use of transport services and/or infrastructure. The
main problem is that it may not provide enough variability in the information to do good
statistical work. Moreover , it is often difficult to isolate all variable relevant to a regulator.
Gompertz function (In traffic = In k. + .b"". In a).
For instance, the choice between bus and cars can be modeled in a logit as follows
Pb' =      , ,    ,where PJ is the probability of choice of bus, Ca, Cb are the monetary costs of cars and buses for all users,
i is the individual and 0 are constants; if bus fares increase by dc . the probablity of taking the bus changes by  b = Pb (1-Ph )P1;
dc
Individual i picks modefover a if: C' < C' that is, if hi <
ti - t



34
Declared/stated preferences are provided by responses given by a sample of potential or
actual users provided when they are placed in hypothetical situations. The potential users are
asked to rank, rate or pick among several options. This is generally organized through field
surveys conducted around the area in which the benefits of a project are supposed to be
observable. Asking users to pick or rate is faster and generally more reliable than asking for a
ranking and is hence more common. T he interview can be conducted in 10- 1 5 minutes-it takes
at least twice as much to ask for a ranking of options. The specific design has of course to be
tailored to the type of mode in use. You have more time to interview a passenger on an intercity
train than to ask the driver of a car waiting for the light to turn green. In some instances, the
survey requires the users to be paid to come to a specific location to do the questionnaire.
Overall, the declared preferences approach is to some extent less demanding initially. It is
however very sensitive to the quality of the questionnaire design and to the quality of the team
conducting this field survey. It needs to be tested carefully as errors are easy-and common. For
urban projects, it is recommended to have sample sizes of the population of about 10% when
passenger flows are larger than 900 pax/hour-and assuming a response rate of 20-30%. It
generally ends up being more expensive than a revealed preference study. The list should make
it clear that this is quite time consuming and requires a strong commitment to success.
While the economists have all the required tools to derive individual demand forecast from
these models, they are not cheap to implement. They are indeed quite time consuming and
demand strong technical skills. These models require detailed surveys through questionnaires
and econometric analysis-which quickly drive costs up. Their main attraction is that they can
be quite effective at allowing an estimate of the probabilities that an individual with specific
characteristics will prefer one mode over another. To get the demand for each mode, this
individual information are aggregated. This also can be done in various ways. each introducing
its own biases which must be understood if the results are to be used in any regulatory decision
(Ben Akiva and Lerman 1985).
The quality of the policy advise generated from these models have made become a
standard in the industry. They are being used in assessing projects in various Latin American
countries at reasonable costs and some of these examples will be discussed below. One of their



35
useful adaptations is the inclusion of a constraints that recognizes quite explicitly that ability to
pay matters in making choices. 30
A.4. The choice of itineraries
These models are mostly used for the road sector. Without saturation, it is very similar to
what was discussed earlier with respect to modal choices. With saturation-i.e. when cost
depend on traffic levels--, the modelers need to implement the two Wardrop principles. The first
principle states that in equilibrium, the generalized costs of every itinerary used between an
origin and a destination given are equal or lower than those of any unused itinerary. The problem
can now easily be modeled thanks to major progress in computer technologies. The main
problem is that this is an optimum only from the view point of the individual user and in fact the
second Wardrop principle suggests that the collective optimum is unlikely to be consistent with
this individual optimum. What this means is that regulators must understand the sources of
indifferences before deciding how to intervene. This is particularly important because one f the
main policy conclusion derived from the analytical developments surrounding these principles is
that the best solutions are not always what they appear to be. More specifically, the best solution
to saturation is not always more investment but can include various types of price, rationing or
queuing arrangements.
In comparing alternatives between a short saturated itinerary and a long unsaturated
itinerary, it is important to consider that displacement effects resulting from an improvement in
one route may simply maintain travel costs constant. (Ville 1970). Indeed, users will shift to the
short way until congestion increase travel time while the users of the long itinerary see not
change. This can be exacerbated by a situation in which a regulator accepts to reduce the
frequency of public transport on the long itinerary to pay for the investment made to expand the
short itinerary or as part of a negotiation with the concessionaire of this short itinerary.
30 Price time models can also be used to pick itineraries or to model air-rail competition and hence can be useful to any regulatory or competition
agency. They assume that a user i decides as a function ofthe generalized cost for each mode: C' =P, +hi t,    = Pa, + hita
where: C', C' are the generalized costs for modes a andf; pa et pf are the prices of each mode;.t. et tf are the travel time, hi is the value of
time for individual i.






Policy Research Working Paper Series
Contact
Title                             Aulhor                   Date               for paper
WPS2429 Corruption, Composition of Capital    Shang-Jin Wei         August 2000        H. Sladovich
Flows, and Currency Crises                                                    37698
WPS2430 Financial Structure and Bank       Asli DemirgCsq-Kunt      August 2000         K. Labrie
Profitability                     Harry Huizinga                              31001
WPS2431 Inside the Crisis: An Empirical    Asli Demirguc-Kunt       August 2000        K. Labrie
Analysis of Banking Systems in    Enrica Detragiache                          31001
Distress                          Poonam Gupta
WPS2432 Funding Growth in Bank-Based and   Asli Demirgui-Kunt       August 2000        K. Labrie
Market-Based Financial Systems:    Vojislav Maksimovic                        31001
Evidence from Firm-Level Data
WPS2433 External Interventions and the      Ibrahim A. Elbadawi     September 2000      H. Sladovich
Duration of Civil Wars            Nicholas Sambanis                           37698
WPS2434 Socioeconornic inequalities in Child   Adam Wagstaff        September 2000     A. Maranon
Malnutrition in the Developing World   Naoko Watanabe                         38009
WPS2435 The Evolution of Poverty during the   Asep Suryahadi        September 2000      P. Sader
Crisis in Indonesia, 1996-99      Sudarno Sumarto                             33902
Yusuf Suharso
Lant Pritchett
WPS2436 Safety Nets and Safety Ropes:      Sudarno Sumarto          September 2000     P. Sader
Who Benefited from Two Indonesian   Asep Suryahadi                            33902
Crisis Programs-the "Poor" or the    Lant Pritchett
"Shocked?"
WPS2437 Quantifying Vulnerability to Poverty:  Lant Pritchett       September 2000      P. Sader
A Proposed Measure, Applied to    Asep Suryahadi                              33902
Indonesia                         Sudarno Sumarto
WPS2438 Measurements of Poverty in         Menno Pradhan            September 2000      P. Sader
Indonesia: 1996, 1999, and Beyond   Asep Suryahadi                            33902
Sudarno Sumarto
Lant Pritchett
WPS2439 State-Community Synergies in       Monica Das Gupta         September 2000      M. Das Gupta
Development: Laying the Basis for   Helene Grandvoinnet                       31983
Collective Action                 Mattia Romani
WPS2440 Lessons from Uganda on Strategies  John Mackinnon           September 2000     H. Sladovich
to Fight Poverty                  Ritva Reinikka                              37698
WPS2441 Controlling the Fiscal Costs of    Patrick Honohan          September 2000     A. Yaptenco
Banking Crises                    Daniela Klingebiel                          38526



Policy Research Working Paper Series
Contact
Title                            Author                   Date              for paper
WPS2442 A Firms's-Eye View of Policy and   Bernard Gauthier        September 2000     L. Tabada
Fiscal Reforms in Cameroon       Isidro Soloaga                             36896
James Tybout
WPS2443 The Politics of Economic Policy    Richard H. Adams Jr.    September 2000     M. Coleridge-Taylor
Reform in Developing Countries                                              33704
WPS2444 Seize the State, Seize the Day:    Joel S. Hellman         September 2000     D. Billups
State Capture, Corruption, and   Geraint Jones                              35818
Influence in Transition          Daniel Kaufmann
WPS2445 Subsidies in Chilean Public Utilities   Pablo Serra        September 2000     G. Chenet-Smith
36370