Research in Transportation Economics: Road Pricing, Vol. 9 (2004)

ROAD PRICING: THEORY AND EVIDENCE

RESEARCH IN TRANSPORTATION ECONOMICS Series Editor: Martin Dresner Recently Published Volumes: Volumes 1–6:

Research in Transportation Economics – Edited by B. Starr McMullen

Volume 7:

Railroad Bankruptcies and Mergers from Chicago West 1975–2001: Financial Analysis and Regulatory Critique – Michael Conant

Volume 8:

Economic Impacts of Intelligent Transportation Systems: Innovations and Case Studies – Evangelos Bekiaris & Yuko Nakanishi

RESEARCH IN TRANSPORTATION ECONOMICS VOLUME 9

ROAD PRICING: THEORY AND EVIDENCE EDITED BY

GEORGINA SANTOS University of Cambridge, Cambridge, UK

2004

Amsterdam – Boston – Heidelberg – London – New York – Oxford Paris – San Diego – San Francisco – Singapore – Sydney – Tokyo

ELSEVIER B.V. Sara Burgerhartstraat 25 P.O. Box 211 1000 AE Amsterdam The Netherlands

ELSEVIER Inc. 525 B Street, Suite 1900 San Diego CA 92101-4495 USA

ELSEVIER Ltd The Boulevard, Langford Lane, Kidlington Oxford OX5 1GB UK

ELSEVIER Ltd 84 Theobalds Road London WC1X 8RR UK

© 2004 Elsevier Ltd. All rights reserved. This work is protected under copyright by Elsevier Ltd, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier’s Rights Department in Oxford, UK; phone: (+44) 1865 843830, fax: (+44) 1865 853333, e-mail: [email protected]. Requests may also be completed on-line via the Elsevier homepage (http://www.elsevier.com/locate/permissions). In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 20 7631 5555; fax: (+44) 20 7631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of the Publisher is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier’s Rights Department, at the fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made. First edition 2004 British Library Cataloguing in Publication Data A catalogue record is available from the British Library. ISBN: 0-7623-0968-7 ISSN: 0739-8859 (Series) ∞ The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of  Paper). Printed in The Netherlands.

CONTENTS LIST OF CONTRIBUTORS

vii

LIST OF REFEREES

ix

PREFACE

xi PART I: THEORY

1.

2.

3.

4.

5.

THE RATIONALE FOR ROAD PRICING: STANDARD THEORY AND LATEST ADVANCES Kenneth Button

3

SECOND-BEST PRICING FOR IMPERFECT SUBSTITUTES IN URBAN NETWORKS Jan Rouwendal and Erik T. Verhoef

27

THE IMPACT ON NETWORK PERFORMANCE OF DRIVERS’ RESPONSE TO ALTERNATIVE ROAD PRICING SCHEMES Anthony May and David Milne

61

OPTIMAL LOCATIONS AND CHARGES FOR CORDON SCHEMES Anthony May, Simon Shepherd and Agachai Sumalee

87

TIME-VARYING ROAD PRICING AND CHOICE OF TOLL LOCATIONS André de Palma, Robin Lindsey and Emile Quinet

v

107

vi

6.

7.

8.

ROAD PRICING AND PUBLIC TRANSPORT Kenneth A. Small

133

MARGINAL SOCIAL COST PRICING FOR ALL TRANSPORT MODES AND THE EFFECTS OF MODAL BUDGET CONSTRAINTS Stef Proost and Kurt Van Dender

159

WELFARE AND DISTRIBUTIONAL EFFECTS OF ROAD PRICING SCHEMES FOR METROPOLITAN WASHINGTON DC Elena Safirova, Kenneth Gillingham, Ian Parry, Peter Nelson, Winston Harrington and David Mason

179

PART II: EVIDENCE 9.

10.

11.

12.

TRANSPORT POLICIES IN SINGAPORE Georgina Santos, Wai Wing Li and Winston T. H. Koh

209

NORWEGIAN URBAN TOLLS Farideh Ramjerdi, Harald Minken and Knut Østmoe

237

URBAN ROAD PRICING IN THE U.K. Georgina Santos

251

RECENT U.S. EXPERIENCE: PILOT PROJECTS Patrick DeCorla-Souza

283

LIST OF CONTRIBUTORS Kenneth Button

George Mason University, USA

Patrick DeCorla-Souza

U.S. Federal Highway Administration, USA

Andr´e de Palma

Universit´e de Cergy-Pontoise, France

Kenneth Gillingham

Resources for the Future, USA

Winston Harrington

Resources for the Future, USA

Winston T. H. Koh

Singapore Management University, Singapore

Wai Wing Li

University of Cambridge, UK

Robin Lindsey

University of Alberta, Canada

David Mason

Resources for the Future, USA

Anthony May

University of Leeds, UK

David Milne

University of Leeds, UK

Harald Minken

Institute of Transport Economics, Norway

Peter Nelson

Resources for the Future, USA

Knut Østmoe

Institute of Transport Economics, Norway

Ian Parry

Resources for the Future, USA

Stef Proost

Katholieke Universiteit Leuven, Belgium

Emile Quinet

Ecole Nationale des Ponts et Chauss´ees, France

Farideh Ramjerdi

Institute of Transport Economics, Norway

Jan Rouwendal

Free University Amsterdam, The Netherlands

Elena Safirova

Resources for the Future, USA

Georgina Santos

University of Cambridge, UK vii

viii

Simon Shepherd

University of Leeds, UK

Kenneth A. Small

University of California at Irvine, USA

Agachai Sumalee

University of Leeds, UK

Kurt Van Dender

University of California at Irvine, USA

Erik T. Verhoef

Free University Amsterdam, The Netherlands

LIST OF REFEREES Richard Arnott

Boston College, USA

Svein Br˚athen

Molde University College, Norway

Mark Burris

Texas A&M University, USA

Mark Delucchi

University of California at Davis, USA

Donald Hearn

University of Florida, USA

Stephen Ison

Loughborough University, UK

Peter Jones

University of Westminster, UK

David Levinson

University of Minnesota, USA

Todd Litman

Victoria Transport Policy Institute, Canada

Peter Mackie

University of Leeds, UK

David Maddison

Syddansk Universitet, Denmark

Inge Mayeres

Katholieke Universiteit Leuven, Belgium

Herbert Mohring

University of Minnesota, USA

Se-il Mun

Kyoto University, Japan

David Newbery

University of Cambridge, UK

James Odeck

Norwegian Public Roads Administration, Norway

Piotr Olszewski

Nanyang Technological University, Singapore

Juan de Dios Ort´uzar

Pontificia Universidad Cat´olica de Chile, Chile

Don Pickrell

U.S. Department of Transportation, USA

David Reams

Department for Transport, UK

Martin Richards

Policy Advisor, UK ix

x

Georgina Santos

University of Cambridge, UK

Stephen Smith

University College London, UK

Terje Tretvik

SINTEF, Norway

Martin Wachs

University of California at Berkeley, USA

Clifford Winston

The Brookings Institution, USA

Hai Yang

The Hong Kong University of Science and Technology, Hong Kong

Xiaoning Zhang

The Hong Kong University of Science and Technology, Hong Kong

DISCLAIMER The views expressed in each chapter are those of the authors and not necessarily those of the institutions or organisations they work for, nor of the editor or referees.

PREFACE Traffic congestion is a relatively new challenge for mankind. Although even in Roman times, congestion was a problem in city centres, specially with all the administrative activities and the forum, with traders and merchants, travellers and residents passing by or doing some business, it was not until the invention of the motor-vehicle that conditions started to deteriorate rapidly. One of the key facts used by the Mayor and by Transport for London to try to persuade Londoners about the need for congestion charging was that average speeds in Central London in 2000 were the same as in 1900. Although London might have been at the end of the scale, with very high levels of congestion, many towns and cities around the world are inundated by cars, and also suffer traffic congestion, at least to some extent. Congestion affects both developed and developing countries alike, and people’s productivity, leisure time and quality of life in general. Whilst a Roman soldier acting as traffic controller may have been sufficient to solve congestion in 400AD it is clear that more sophisticated measures are needed today. The solution of managing demand is so obvious that almost no one would dare say it is ill-conceived. Indeed it rests on sound economic theory, and the comparison with other markets that experience peak demand has become a text-book resource to exemplify how rational and sensible it would be to introduce road pricing to reduce congestion. Yet the cases of urban road pricing implementation can be counted on the fingers of one hand. The aim of this volume is ambitious. It contains papers that will be useful to a range of readers, including researchers and students, civil servants, policy makers and consultants. The book has two parts. The first part is mainly theoretical. It starts with a chapter on the rationale for road pricing, which describes the basic concepts and briefly reviews some of the pioneering work and recent extensions. The chapters that follow report results of recent research on the subject. Chapter 2 concentrates on second-best congestion pricing in urban areas, where it is usually very difficult to charge differentiated tolls on every link of the network and for every mode, and where alternative routes and modes are often imperfect substitutes. This has practical consequences, as in real life situations, charges cannot be optimally tuned on a link-by-link and mode-by-mode basis. Chapter 3 reports results of simulations of different types of road pricing and their impact on the performance of the road network. Clearly, the original idea xi

xii

of charging road users for the congestion externality they impose on others can be expanded in different directions and adapted in different ways. The outcomes will be different and so knowing the potential impacts in advance can help decide which pricing system to choose. Chapter 4 develops methods to determine optimal toll locations and charge levels on urban networks and demonstrates that considerable welfare gains can be obtained by adjusting location and level. Choosing the cordon and deciding on the toll to charge are thus a matter of major importance. Chapter 5 explores the problem of what links to price on a network and how to phase the system. A flat toll constant throughout the day, a step toll with a base and a peak level, and a fine toll that changes smoothly during the arrival of vehicles to a queue are considered. Each option leads to different recommendations as whether to concentrate tolls in some area or disperse them all over the network. The policy implications of these findings are obvious, as the degree of time variation chosen may influence the charge locations. Chapter 6 investigates the potential impacts of road pricing on costs and service quality of public transport buses, and the second-round effects of these changes on the behaviour of public transport operators and potential users, and the contribution of these impacts to the overall benefits from road pricing. Using some preliminary numbers from the London Congestion Charging Scheme implemented in February 2003, it appears that the positive impacts on bus services, and the associated contribution to the net benefits of road pricing, are substantial. Chapter 7 estimates the efficiency costs and transport sector effects of simple average cost pricing and Ramsey pricing, when they guarantee a balanced budget per mode, for a set of European cities and non-urban areas, with the help of an aggregate optimal pricing model. It is found that Ramsey-type pricing rules perform better in terms of welfare than average cost based rules, and that the absence of a budget constraint allows a better approximation of prices to marginal social cost. This has practical implications as the pricing rule chosen will affect social welfare, which is something policy makers may wish to take into consideration. Chapter 8 deals with the distributional impacts, one of the most common objections to road pricing. The paper is devoted to the analysis of the potential distributional effects of road pricing policies in Metropolitan Washington DC. Interestingly, it is found that high occupancy/toll lanes, a system in which drivers only pay to use the express lanes if the occupancy of their vehicle is below the minimum required number, very much supported by the Value Pricing Pilot Program, yields important social welfare gains with little perverse distributional effects across different income groups, and across local jurisdictions, even without any compensation schemes. Although this result rests on the assumption of preexisting high occupancy vehicle lanes, and does not apply to the imposition of new

xiii

single or multiple lane tolls on unregulated freeways, it is nonetheless important. The qualitative results for converting high occupancy vehicle lanes into high occupancy/toll lanes are general and will apply to other regions as they are not specific to Washington DC. It is clear that concerns over equity may have been exaggerated in the past, and that a thorough assessment should be carried out before rejecting a road pricing proposal for a town or city on the basis of regressivity alone. The second part of the book concentrates on the practice of road pricing. Although the idea of road pricing has been discussed for many decades and research has been carried out from a range of disciplines including maths, physics, engineering, geography and economics, road pricing in urban areas has only been implemented in very few places. Chapter 9 is a comprehensive review of the transport policies in place in Singapore. The famous area licensing introduced in Singapore in 1975 was the first application ever of charging to manage traffic demand. It was not however the only measure implemented but only part of a package that constitutes an example of integrated transport policy. Chapter 10 describes the toll systems in Norway. Although they were not conceived to manage demand but to raise revenues to fund infrastructure, they might be modified to manage traffic demand at peak times in the near future. The authorities in Bergen and Oslo are currently considering the possibility of managing congestion with some kind of time-varying toll. Chapter 11 describes the London Congestion Charging Scheme and the Durham Scheme, recently implemented in England, and reports some preliminary results. It also reviews the possibilities and potential impacts of implementing road pricing in Edinburgh. Finally, Chapter 12 describes the different projects in operation in the US, and some of the projects under consideration, with special attention to those funded by the Value Pricing Pilot Program, a program by which a federal grant is provided to states, local governments, or other public entities, matching 80% of the costs to implement, operate and monitor pricing projects. I would like to thank all the contributors to this volume and the referees that commented on the papers. Almost everybody who helped is senior to me, in age and experience, and so my gratitude is even greater, for having trusted me on this project and spared precious time from their very tight schedules. Needless to say this volume would have not been possible without the authors’ and the referees’ contributions. Georgina Santos Editor Cambridge, October 2003

This Page Intentionally Left Blank

PART I: THEORY

This Page Intentionally Left Blank

1.

THE RATIONALE FOR ROAD PRICING: STANDARD THEORY AND LATEST ADVANCES

Kenneth Button 1. INTRODUCTION Road Pricing is a simple concept that extends the common practice that is virtually ubiquitous in every other sector of a market economy whereby prices are used to reflect scarcity, and to allocate resources to those that can best use them. In most places road space, even in such supposedly market orientated societies as the U.S., is in actuality allocated in a manner more akin to the general practices employed in pre-1989 communist Russia, namely by waiting in queues and lines. While there are some that see merit in using waiting as an allocation device (Barzel, 1974), by and large society finds congestion inefficient and wasteful. After many years of adding capacity in an attempt to reduce congestion, a number of transport and road authorities at national, regional and local level are seeking to move away from the centralist approach to roads policy to one that has at least a veneer of economic rationality underpinning it. The introduction of Area Licensing to Singapore in 1975 is the classic case study of a pioneering application, but despite the success of the measure it is often seen as not being particularly relevant for other cities, and especially those in Western Europe and North America. It has taken over a quarter of a century for another scheme to essentially replicate it. Certainly there have been toll rings

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 3–25 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09001-8

3

4

KENNETH BUTTON

introduced in a number of Norwegian cities. High occupancy/toll lanes have emerged with the opening of Californian’s FASTRACK north of San Diego where new capacity is priced on a real time basis in lanes that run parallel to free lanes. Although a number of other initiatives have also taken place, they have all generally been driven primarily by the need to find revenues for engineering works rather than as a policy to allocate traffic on extant networks. The basic concepts of the costs of congestion and congestion pricing are explained in Sections 2 and 3. A brief summary of some pioneering work on the subject is presented in Section 4, followed by a description of the renewed interest in road pricing of the 1960s in Section 5. Section 6 moves on to look at some of the more recent advances in the theory of road pricing. Section 7 briefly links road pricing with capacity expansion, and Section 8 warns on the importance of the value of time. Some of the problems with implementation are described in Section 9, and Section 10 gives some brief thoughts on the applications to date. Section 11 concludes.1

2. CONGESTION COSTS Travel demand is mainly a derived demand. Travel is usually demanded not for its own sake but as a means of consuming some other good or service. Because the activities with which transport is associated vary over time, the demand for transport is not constant over time. For example, many towns and cities experience traffic congestion during commuting times (morning and evening), and holiday routes experience seasonal congestion. Transport infrastructure in the short run has a finite capacity. When users of a particular road begin to interfere with other users because the capacity of the road is limited, then congestion externalities arise. Although some degree of congestion is desirable, or capacity would be under-utilised most of the time, excessive congestion is not. The question then is, what is the optimal level of congestion? The economic costs of traffic congestion can be calculated using the engineering concept of the speed-flow relationship.2 If a straight one-way street is assumed, the relationship between speed and flow over time typically looks like that depicted on Fig. 1. Flow (the number of vehicles that pass a certain point3 ) is dependent upon the number of vehicles entering a road and the speed of traffic. Hence, at low volumes of traffic, high speeds are possible, constrained only by the capability of the vehicle and the legal speed limits. As the number of vehicles trying to enter the road increases, vehicles affect each other’s speeds and slow one another down. As more traffic enters the road, speeds fall but, up to a point, flow will continue to rise because the effect of additional vehicles outweighs

The Rationale for Road Pricing

5

Fig. 1. Speed-Flow Relationship for a Link.

the reduction in average speed. This is the congested branch of the speed-flow curve. At the point where increased traffic volume ceases to offset the reduced speed the road’s capacity is reached at the maximum flow or capacity of the link, indicated k on Fig. 1. At that point the flow turns unstable, with the characteristic stop-start conditions, typical of a traffic jam. If vehicles keep on entering the road further drops in speed and flow will result. This is known as forced flow or hyper-congestion. As flow keeps decreasing, the average speed will eventually increase and jump back to some point along the normal flow portion of the speed-flow relationship. The form of the speed-flow relationship depends on a number of factors including width of the lanes, grade, road curvature, speed limit, weather, mix of vehicle types, etc. (Button, 1993; Lindsey & Verhoef, 2002). Walters (1961) was the first one to translate the backward bending speed-flow function into a cost function. Today’s standard transport economic analysis still uses that framework to compute congestion costs and optimal pricing in static models. Thus, the speed-flow curve can be converted into a time-flow curve. Multiplying time per unit of distance (mile or kilometre) by the value of travel time and adding vehicle operating costs (monetary units per unit of distance) gives the average social cost-flow curve, ASC (Morrison, 1986). This is shown on Fig. 2. The ASC is a reverse of the speed-flow curve seen on Fig. 1, with the positively sloped portion corresponding to the negatively sloped section of the speed-flow

6

KENNETH BUTTON

Fig. 2. Derivation of the Cost-Flow Relationship. Note: ASC: Average Social Cost, MSC: Marginal Social Cost. Source: Morrison (1986).

curve. As speeds tend to zero, time and therefore costs, tend to infinity (Verhoef, 1999). The MSC curve is also shown. It represents the extra cost the additional user places on the existing traffic flow. The MSC approaches infinity as flow approaches capacity (Morrison, 1986). Although the hyper-congested portion of the speed-flow function and its associated backward bending ASC curve with multiple equilibria has recently been the subject of much study (Small & Chu, 1997; Verhoef, 1999, 2003), for economic analysis it is common to ignore it (Button, 1993) and this is what shall be done here.4 When a demand curve is added on to the upward sloping portion of the ASC, Fig. 3 is obtained. Road users are assumed to be identical except for their marginal willingness to pay to traverse the link, represented by the demand curve, D, equal to the marginal private benefit (MPB) which, for simplicity shall be assumed to be equal to the marginal social benefit (MSB). According to standard economic theory, the optimal flow is qe , where marginal social cost equals marginal social benefit. The actual flow however will typically be q0 , because road users ignore the congestion they impose on others. The ASC and MSC curves reflect the average and marginal generalised costs associated with different flows; they show time and vehicle operating costs borne by road users when making trips. They can be seen as representing social costs, in the limited sense that they are costs to the society of road users. Any individual driver entering the road will only consider his time and vehicle operating costs, including the congestion costs he will have to bear, which with many users, will be equal to the average cost prevailing at that moment or ASC. Thus the ASC is often referred to as the marginal private cost (MPC), or cost the new user will

The Rationale for Road Pricing

7

Fig. 3. The Simple Diagram of Congestion Pricing.

bear. He will not however take into account the costs that he will impose on other vehicles already on the road. The difference between the ASC and MSC curves at any flow level reflects the marginal congestion cost. From a social point of view the actual flow, q0 , is excessive because the q0 th motorist is only enjoying a benefit of q0 C but imposing costs of q0 M. The additional traffic beyond the optimal level qe can be seen as generating costs q0 MDqe but only enjoying a benefit of q0 CDqe . This yields what in transport economics is called the social cost of congestion or deadweight welfare loss, given by the quasi-triangle DMC. It should be noted that at the optimal flow qe there is still congestion, given by segment DE. The difference is that at this point the congestion externality is internalised, and drivers are paying for the full social costs of their trips.

3. CONGESTION PRICING One idea to optimise the level of congestion and ensure that qe is achieved is to use the price mechanism to make travellers more fully aware of the congestion externality they impose on others. The idea is that motorists should pay for the additional congestion they create when entering a congested road.5

8

KENNETH BUTTON

The optimal congestion charge reflects the difference between the MSC and the ASC, as defined in Section 2. On Fig. 3 this optimal congestion charge is equal to DE. That is the charge that equates MSB and MSC at the efficient level of traffic. Congestion pricing generates a welfare gain of DMC. Traffic flow is reduced from q0 to qe , resulting in some motorists not using the road any longer and thus loosing consumers’ surplus of BCD. At the same time however, the road authority collects revenues FDEG. From these revenues, FDBA are a transfer of consumers’ surplus enjoyed by road users to the road authority in the form of revenue. These revenues can (and perhaps should) be returned to road users.6 In mathematical terms the following model can be used to compute the efficient congestion charge. If the average social cost per km of a representative vehicle is b (1) v where a is the cost per PCU-km (pence per PCU-km), and b is the value of time (pence per PCU-h) and the total social cost of a flow of q vehicles is C = cq, then when an additional vehicle is added to the flow, the total social cost will be increased by c=a+

MSC =

dc dC =c+q dq dq

(2)

The congestion externality or marginal congestion cost is given by q(dc/dq). This externality measured at the efficient level of traffic qe gives the optimal congestion charge DE on Fig. 3. It can be numerically estimated by specifying a speed-flow relationship, v, and making the relevant substitutions in Eqs (1) and (2).

4. PIONEERING WORK The standard analysis presented in Sections 2 and 3 has its roots in work that was carried out in the 19th and early 20th centuries. This section honours that work, as without it, today’s understanding of the problem of traffic congestion and congestion charging would probably be different.

4.1. The French Engineers The French engineers of the 1840s–1850s were as concerned with how their structures should be financed and used as with their physical construction. Dupuit’s (1844) work on the provision of public goods, their pricing and investment

The Rationale for Road Pricing

9

assessment is seminal.7 Dupuit’s well-known analysis of bridge pricing (actually he has six examples scattered throughout his work and they are not entirely consistent), argues for a pricing structure that maximises utility whilst at the same time covering up-keep and capital costs. With an economically optimal capacity this is essentially what a congestion charge does. Minard (1850) also recognised this and, although he thought primarily in terms of physical wear-and-tear, his work also had an implicit congestion component in it. He pioneered the fact that travel time-savings have economic value and suggested simple, revealed preference based, ways of measuring it. He did not, however, possibly because of the context of his work, which mainly concerned uncongested facilities, explicitly link time savings into his infrastructure pricing concepts. At the turn of the 20th century, the French engineering concepts were not widely known. The originators were outside of mainstream economics and the fact that their material was not widely available in English did not help. Perhaps more importantly, the railway age removed much of the interest in charging for road infrastructure. The track costs issue debate switched to considering the appropriate way for railways to recover their fixed costs without a regulated environment designed to limit their quasi-monopoly powers. Matters akin to the later work on Ramsey pricing and the like came to the fore.

4.2. The Pigou Debate Perhaps the most frequently cited name in the development of Road Pricing is that of Arthur Pigou. His book The Economics of Welfare (Pigou, 1920) laid the foundation for much of the subsequent academic literature in the field.8 Pigou’s overriding objective in The Economics of Welfare was to systematically try to bring positive analysis to an area of study that had previously been dominated by normative arguments. Indeed, economics was still often treated as moral philosophy at the time. He made the case that roads were not being utilised efficiently because users were not being charged for the congestion costs they imposed on others. Many would not use the facility if they had to bear this cost and their resources and time would be more gainfully employed doing something else. His analysis looks at two competing roads, one wide but with a poor track and hence slow, and the other narrow but with a good track. Traffic will disperse itself between these alternatives up to the point where travel time to the destination is the same irrespective of the road used. He argues that imposing a differential charge on the narrower road will redistribute traffic that does not value travel time

10

KENNETH BUTTON

savings very highly to the wider, slower road. The result is that aggregate travel time is reduced and society as a whole gains.9 Pigou’s style of analysis however relies upon the argument that roads are public goods in the sense that they are publicly owned and allocated, if not in the more technical sense of them being non-rival and non-excludable. Knight (1924) pointed this out, and illustrated that a privately owned road in a competitive situation would give the provider the incentive to limit the road’s use to the optimal level by fixing tolls to reflect congestion costs. Pigou clearly accepted this argument and that is why he cut Road Pricing from later editions of The Economics of Welfare. Basically, the argument is only relevant when for institutional reasons roads fall outside of the private market. Knight’s argument has stood the test of time within its context of a competitive road system. But others (Buchanan, 1956; Mills, 1981) have correctly pointed out that an assumption of competing private road suppliers is a strong one. Private road owners would in practice enjoy a degree of monopoly power and may have an incentive to over-price roads. However, while with a uniform road price, a rent seeking monopolist limits traffic excessively (Edelson, 1971), a monopolist, with the ability to perfectly price discriminate between users would have the incentive to ensure that the road is used optimally. The consumer surplus in this case would be transferred as economic rent to the road owner, and although this may not be liked, it would be efficient. In a sense, one could see the analysis of Knight and Dupuit as providing benchmarks for road pricing, the former taking a competitive market where there are numerous alternative routes and congestion, and the latter looking more at an optimal monopoly facility without congestion.

5. THE RESURECTION OF THE 1960s The 1960s saw a sudden interest in road pricing. Two seminal papers were published by leading academics of the day (Vickrey, 1963; Walters, 1961) and in the U.K. a government study provided the first major policy analysis of road pricing (Ministry of Transport, 1964). The interest grew largely from practical considerations. Automobile ownership was expanding and cities were getting congested. Land-use planning, urban design and infrastructure expansion were in vogue as mechanisms for combating this but were seen as long term and costly. Subsidising public transport as an alternative to the automobile was under review but had not been widely adopted. In this context, road pricing was seen as a potential means for containing commuter traffic whilst leaving road users options about when, how and where they could travel.

The Rationale for Road Pricing

11

The Vickrey/Walters approach very much followed the lines of Pigou’s analysis. There would be some decision on what the optimal traffic flow should be, essentially determined by estimating the traffic speed-flow relationship underlying the cost curves on Fig. 3 together with the demand function. The congestion charge would then be an efficient way of attaining the target flow of traffic. It was seen as a fiscal instrument allowing those who gained the most utility from road use to use the facility. The model presented on Fig. 3 was developed in the 1960s and is still in use today. The basis for that model are of course much older and as stated above, they go all the way back to the work by Dupuit in the 1840s and Pigou in the 1920s and the engineers and physicists that had been studying the relationship between speed and flow since the 1930s. Vickrey (1969) combined his analysis with a comparable one for urban public transport. While the analysis was technically more rigorous than that of Pigou’s in that congestion was carefully defined, the argument was essentially the same – roads are being publicly provided and in the absence of a market they should be used to maximise social welfare. A slightly different conceptual approach within this vein that emerged at the same time as the Vickrey/Walters analysis is to treat roads as club goods (Buchanan, 1965). This approach has the intellectual attraction of isolating the allocation of road space debate from debates centered on more conventional Pigouvian externalities (that are retained in all of the editions of the The Economics of Welfare) involving the impacts of traffic on non-road users. This latter group embodies such things as environmental effects that are outside the market. A club good approach strictly assumes that a group of people derive sufficient benefit from a facility to provide it, exclude non-members whilst allocating out its use amongst members according to incremental costs (which would include a scarcity cost).10 Since road users are often willing to pay above their incremental costs to use a road, debates arose about ways in which commercial road builders would operate such a system. It may also be tied into a road charging regime that would involve a “membership fee” such as an annual license, and then user fees that capture wear-and-tear costs plus an element to reflect scarcity and congestion at various times.

6. RECENT REFINEMENTS IN SHORT-RUN PRICING The upsurge of interest amongst academics, as well as practitioners since the 1960s has led to refinements in assessing how the congestion charge should be calculated. As Vickrey (1968) pointed out, there is in ideal circumstances a need to vary the

12

KENNETH BUTTON

price according to traffic demand and costs. The time of day, the traffic mix, the physical features of the network and local road conditions (such as the weather and accidents) may influence these. The simple static model presented in Sections 2 and 3 has been extended in different directions. Introducing more realism into this framework adds to its complexity and ipso facto to the ultimate difficulty in calculation of the congestion charge. Research on the demand function, the time dependency and scheduling of trips, second-best pricing, heterogeneity of trip makers, transaction costs, technologies, and the shape of the speed-flow relationship and cost curves for links and areas, are only some of the issues on which attention has focused in the last four decades. Alan Evans (1992) and Hills (1993) for example raised issues about the validity of using a demand curve defined over flow, which has a per-unit-of-time dimension, because road users demand trips and not passages per unit of time. They advocate stock-based models using density or number of trips. Vickrey (1969) developed a bottleneck model, where congestion is assumed to arise when vehicles queue behind a bottleneck. All commuters wish to arrive at work at a certain time but there is a bottleneck with finite capacity that will not allow all of them to arrive at their preferred time. There are costs associated with early and late arrival, which together with the toll, are added to the cost of the trip, which commuters try to minimise by choosing their departure time. Queuing time evolves during the rush hour and this imposes a time pattern of departures. The evolution of congestion over the rush hour is thus endogenously determined. This bottleneck model was further extended, mainly by Arnott, de Palma and Lindsay (1988, 1990a, b, 1992, 1993, 1994, 1998) to include heterogeneous commuters, route choice, stochasticity in capacity, elastic demand, and time-varying tolls. A second generic form of the dynamic congestion model, initially developed by Henderson (1974), does not completely eliminate travel delays at the social optimum (Chu, 1995). Both approaches involve the distribution of travel delays and scheduling of costs of the peak, and the duration of the peak in the untolled equilibrium and the social optimum, being determined endogenoulsy. Road pricing is designed to produce a Pareto optimal use of a facility but this is dependent upon the standard assumptions that surround first-best partial equilibrium analysis and especially that all other prices are equal to marginal costs. Moving into the realm of “second-best,” where the assumptions of the Pareto world are relaxed, is less tidy and inevitably the efficient price becomes situation specific. One particular issue that is often addressed in this context is the pricing of substitutes to road transport, and especially public transport. In many cases

The Rationale for Road Pricing

13

subsidies have been given to public transport to attract trip-makers away from congested streets. Irrespective of whether this has effected any significant transfer, this is an instrument that could be used to optimise the modal split of traffic. A road user in the absence of road pricing will only consider the marginal private cost of a trip that is lower than the full marginal social cost of the road space taken up. A subsidy to public transport, accurately reflecting the cross-elasticity of demand between the modes, could theoretically be used to attract sufficient motorists from the roads to limit road use to qe in Fig. 3. Other topics within second-best pricing have been the pricing of only part of a road network, or pricing subject to constraints (Liu & McDonald, 1999; Small & Yan, 2001; Verhoef, 2000, 2002a, b; Verhoef et al., 1996). When some routes are priced and some are not, road users are left with a choice between a facility where access is unpriced but the service quality is poor and one where there is an access price but congestion is optimised.11 The road price introduced on such routes is required to reflect the opportunity cost of not using the unpriced roads as well as the cost of using the priced facility. Ignoring the cross-effects will lead to sub-optimally high use of the unpriced facility. Marchand (1968) was one of the first to look at this type of situation in terms of setting a charge that weighs the welfare benefits of reducing traffic on the unpriced facilities against the losses from deviating from first-best pricing on the priced route. Second-best considerations may also become important in the context of complements to, as well as substitutes for, a priced facility. Parking is, not unexpectedly, the most explored complimentary good (Calthrop et al., 2000; Glazer & Niskanen, 1992; Verhoef, 1996), although even here the literature is remarkably thin. Indeed because of the complementary nature of road use and parking, parking charges have in some cases been treated as an alternative to a road price (Verhoef et al., 1995). Although parking charges often involve low transactions costs, the circumstances where they can act as complete substitutes for road pricing are very stylised. In a first-best world there would be optimal parking fees reflecting the opportunity cost of taking up land to “rest” a vehicle as well as the congestion costs associated with cruising around seeking a parking spot.12 In practice, parking is often provided free or perverse charging regimes are employed. Such regimes are frequently structured to limit certain categories of users such as long stay users irrespective of willingness to pay. Parking fees should ideally be structured to reflect the traffic congestion caused by drivers cruising around looking for a parking place. Parking costs can also affect traffic demand by influencing the lengths of trips made over a network. Low parking fees, being a fixed cost of a trip, are spread more over a longer trip than a shorter one. This may affect congestion in sub-urban areas.

14

KENNETH BUTTON

A further input into urban car trips that has received a lot of attention is that of environmental costs. There are few that argue that motor traffic does not make considerable demands on the environment and that the costs involved are external to the market. These effects are both wide-ranging going through the emission of a variety of local, regional and global atmospheric pollutants, to noise, ground water contamination, through to visual intrusion, are in some cases very large. A road price that reflected the total marginal social cost would internalise both the congestion and the environmental externality, and would include a congestion charge and an environmental charge. This is a large subject area, the issues complex, and the literature substantial and is not dealt with here. Fuel taxes however can approximate optimal environmental charges because most environmental externalities are closely dependent on fuel consumption.13 There is another side to the second-best debate that should perhaps just be touched upon. There has been a mounting interest in how other road taxes, public transport fares and road investment decisions should be made in the absence of optimal road pricing. In this context Sherman (1971), Bertrand (1977), and Arnott and Yan (2000) have looked at pricing across multi-modal systems; Wheaton (1978) developed ideas of second-best road investment strategies; and Chia et al. (2001) examined the appropriate fuel taxation policy to pursue if appropriate congestion charges are absent. Much of the early work on road pricing treated all road users as being identical, Strotz (1965) being an exception. Variations in their income and, linked to this the valuations that they place on travel time saving and the size of vehicles driven, which affects road take, are now appreciated as being important considerations for both efficiency and equity reasons. At the very least, they raise questions about whether the congestion charge should reflect these features. If the concern is purely with efficiency and standard first-best conditions hold then there is no need to consider these variations with time varying congestion pricing. This is because all that matters are the congestion costs imposed. Arnott and Kraus (1998) provide a rigorous proof of this. Where there are alternative routes in the network (including different lanes on a road) then where users have different utility functions, optimality is attained by offering various congestion charge/congestion level combinations. This, as Pigou indicated, maximises welfare by allowing those with a high travel time value to buy their way onto the faster routes, leaving those with a lower value on the cheaper, more congested ones.14 The situation can also be compared with having users with different trip-time preferences using a single route. Arnott et al. (1992) show that where there are temporally inflexible congestion charges, a different charge on each of two roads is more efficient than a single uniform charge for both.

The Rationale for Road Pricing

15

Other situations that have attracted attention include those when drivers have different preferences for speed that are unrelated to income (Verhoef et al., 1999). Here the models have often tended to exclude the possibility of overtaking making them particularly relevant to bridge and tunnel cases. The results suggest that although the congestion charge needs to be higher for slow vehicles to reflect their impact on slowing of higher speed vehicles, the differential, because slow vehicles affect fewer fast vehicles on average and also because the speed of fast vehicles declines asymptotically to that of the slower vehicle, should decrease with the proportion of slow vehicles in the traffic stream. This approach would require differential approaches to congestion pricing depending on drivers’ speed preferences – in practical terms this may be done by having different pricing regimes for cars and trucks. Added to these challenges, even if public transport is not an issue, the network is simple and road users are homogeneous, the standard model of road pricing excludes any consideration of implementation and operation costs, like for example, the costs of collecting the fees and policing the system.15 These may be significant and may become entwined with other issues such as how much information the road user has on the “product” that is being purchased (de Palma & Lindsey, 1998). The idea of using electronic technology to collect a real time, variable congestion charge was initially advanced by William Vickrey (1959) in evidence to a U.S. congressional committee as early as the late 1950s. Despite the largely successful testing of electronic road pricing equipment in Hong Kong (Hau, 1990), operating costs, issues of privacy and concern over technical reliability have favoured area based systems. This means that a road price does not vary directly with traffic conditions but rather location or time of day is used as proxy for congestion levels. The Singapore and the recent London schemes use a combination of these surrogates largely for pragmatic reasons. The efficiency of these discrete charges is inevitably less than a continuous congestion charge, the loss of efficiency being a function of just how closely the proxies reflect changing congestion levels. This in turn often depends on the number of “steps” involved and the prices charged at each step. Empirical estimates of the extent to which these systems lose efficiency are found in Chu (1999) and Arnott et al. (1993, 1998) and they seem to be significant.

7. CAPACITY EXPANSION A market price serves not only to allocate current facilities optimally but it also provides signals to where capacity should be expanded, and the revenues from

16

KENNETH BUTTON

the price provide resources for that expansion. In a perfect market situation, effectively Knight’s position, there is no requirement for separate consideration of capacity expansion – the net revenue flows give the necessary guidance. Given the reality of public ownership of most road track and the possibility of internalising the congestion externality with a corrective congestion charge, there have been a number of studies looking at the link between the optimal congestion charge and road investment strategies (Mohring, 1970). An important result that provides an additional incentive for the road agency to internalise the congestion externality is that the optimal congestion charge, as defined in Section 3, equal to the difference between the MSC and the ASC at the efficient level of traffic, covers the fixed operating cost, equal to the sum of interest, depreciation and policing, provided there are constant returns to scale in road construction and maintenance and capacity can be increased in continuous increments (Mohring & Harwitz, 1962). This result was later expanded to include damage. In this sense, a road charge equal to the sum of the optimal congestion charge and the road damage charge, covers the fixed operating cost as defined above, and the variable maintenance cost, provided there are constant returns to scale in construction, maintenance and use of road capacity (Newbery, 1989). Under constant returns and optimal pricing, whenever there is economic profit, the capacity of the road should be expanded, and when there is economic loss, the road has been overbuilt and should be abandoned or closed. The optimal pricing and investment decision for roads can thus be dealt within a single model (Keeler & Small, 1977; Mohring & Harwitz, 1962). When there are diseconomies of scale at the optimal capacity and with an optimal congestion charge, economic rent will be earned. This reflects a scarcity rent on a fixed factor of production, in this case, the land. Equally, if there are economies of scale, and in some instances when there are large indivisibilities in investments, the optimal congestion charge, unless the road agency can price discriminate, will not generate sufficient revenue to cover the fixed operating cost. Subsidies will be required and techniques such as cost-benefit analysis come into play in determining their level. Much, therefore, depends on the view taken about the nature of returns to scale and indivisibilities in road investment. Kraus (1981) argues that the existence of lanes implicitly means that there are indivisibilities in supply. In contrast, Starkie (1982), following the legacy of Keeler and Small (1977) looks at a road network as a whole and argues that within that context, actions such as resurfacing, changes in lane width, etc, can be varied incrementally. Hau (1998) argues that increasing, decreasing and constant returns to scale all exist in road transport and each case should be assessed individually.

The Rationale for Road Pricing

17

8. THE VALUE OF TIME One of the most important benefits of road pricing (and indeed of any transport project) is travel-time savings. Time is a significant cost component of any trip and its value is therefore of fundamental importance. Indeed different values of travel time and travel time savings will lead to different answers on the viability of a transport project. The extensive literature on the subject (Brownstone & Small, 2003; Calfee & Winston, 1998; Hague Consulting Group et al., 1999; Hensher, 1997; Jara-D´ıaz, 2000; Lam & Small, 2001; Mackie et al., 2001, 2003; Small, 1982; Wardman, 1998; Whelan & Bates, 2001) suggests there is ample evidence that the value of time varies amongst population subgroups (Small, 1999) according to income, gender and age, and also with trip purpose, mode used, length of the trip, prevailing conditions of congestion and sometimes weather, and the value given to the travel time variation depends on the sign (loss or saving) and the magnitude, and its ratio to total travel time. As a consequence of this, each road pricing proposal will need to be evaluated on an individual basis, paying particular attention to the socio-economic characteristics and trip patterns of the road users to be affected. The value of the travel time savings will play a substantial role in the net benefits to be derived from any transport project, including road pricing schemes.

9. THE POLITICAL ECONOMY OF ROAD PRICING As Small and G´omez-Ib´an˜ ez (1998, p. 239) rightly observed, “Winning political approval for any form of congestion pricing project is difficult in a democracy, even with careful planning. There may be a number of reasons for this but the political economy of road pricing has largely centered around the distribution implications of the policy.” This was a point made as early as the mid-1960s by Clifford Sharp (1966) although not fully appreciated until later. Indeed Andrew Evans (1992) and others now see it as the critical practical issue to large scale application. This is hardly surprising since it is the road user that immediately loses as a result of charging for congestion. While economists tend to focus mainly on the efficiency implications of transport policies, politicians and administrators put more emphasis on the various effects on different groups in society. Overall the road user would be worse off with congestion pricing, although within that some users who value travel time savings and reliability of service

18

KENNETH BUTTON

highly would gain. Those concerned with distributional effects essentially accept that congestion pricing may result in a potential Pareto improvement in as much as theoretically those who gain could be “taxed” to yield compensation to those who lose out, and still enjoy a net benefit. The problem however is that road pricing projects tend to lack any actual compensation mechanism. As a first step in introducing incidence into the analysis it is necessary to know exactly who gains and who loses, and in what ways. The distributional impacts will typically depend on the nature of the road pricing regime. While there are important theoretical contributions in this field, the studies have been largely empirical in their approach. What emerges from the mainly empirical literature on the distributional impacts of road pricing is a lack of consensus. Layard (1977) for example, argues that the policy is regressive. Foster (1975) argues that it is progressive and Richardson (1974) that it is primarily the middle income groups that would be adversely affected. Giuliano (1992) takes a somewhat different position, namely that even if road pricing is regressive, the fact is used to disguise other objections to the policy rather than regressivity necessarily being important in itself. Santos and Rojey (2004) conclude that the distributional impacts depend on the design of the scheme and the geo-economic characteristics of the town in question (where do people live, where do they work and how do they get to work). A related topic to the distributional effects is the implications of various ways of using the revenues from a congestion charge. Some, such as Small (1983) and Andrew Evans (1992) see this as the core of the political economy debate. Newbery (1990, 1994) suggests that revenues from congestion pricing in the U.K. could be returned to motorists in the form of lower fuel duties. Motorists as a group would then be made better off as they would pay the same total road user charges but would experience less congestion. Goodwin (1989, 1990) and Small (1992) both propose that revenues should be allocated in thirds, one third to the reduction of taxes in general, or fuel taxes, or vehicle license fees or to the increase in social expenditure, one third to investment in public transit, and one third to investment in new roads or reimbursement to trip makers. Although this extremely arbitrary approach may have a superficial popularism that could help the introduction of road pricing it has the inherent long term danger of capture by vested interests and could become an ingrained impediment to other reforms in the future. In a purely competitive private market, large amounts of revenue would signal that road use was highly valued by users of the roads and the net revenues would go towards enhancement of the road network that would bring down the future cost to motorists of using it. In a second-best situation characterised by a variety

The Rationale for Road Pricing

19

of market and government failures across the economy this is not automatically the best way to meet concerns about compensation.

10. SOME BRIEF THOUGHTS ON APPLICATIONS TO DATE Despite all the energies of academics that have been expended, together with the time of a number of policy makers, the textbook application of road pricing has not yet emerged and it probably never will. The few applications of the idea are simplifications of the concept of the Pigouvian charge. This is perhaps inevitable, and the theoretical work should, at best, be viewed as providing important benchmarks against which success is measured. The introduction and impact of the Area Licensing system in Singapore offered a perspective on the power of the price mechanism to control traffic flows – until then elasticities of demand for road space had been little more than conjectures on academics’ note pads. The subsequent experiments with electronic technologies in Hong Kong to make charges more reflective of real-time congestion illustrated that more sophisticated methods could be deployed if desired (Dawson & Catling, 1986). This has been supported in many places where conventional tolling has reverted to rapid pass collecting techniques. Subsequent years have in many ways seen the initial demonstration impacts of the Singapore scheme and the Hong Kong experiment, limited as they were, be diluted. Indeed, this piece actually reflects some of the factors affecting this. Other miracle cures for traffic congestion problems, ranging from intelligent transportation systems that manage traffic more efficiently through the network to the aim of encouraging more teleworking to reduce the number of daily commuters, have come along, and largely gone away again after making a limited, albeit sometimes useful, contribution. None, however, have confronted the basic problem that road users do not take full account of their actions on others. The London Congestion Charging Scheme has recently attracted considerable attention from around the world and the results to-date of significant reductions in car traffic combined with enhancements in the quality of bus services on less congested roads are being noted. The introduction of high occupancy/toll lanes in California provides useful insights into viable technologies and has generated considerable amounts of new information about the sensitivity of road users behaviour when confronted by variable road charges. The Norwegian toll rings, although not strictly a Road

20

KENNETH BUTTON

Pricing initiative, have offered insights on how political coalitions may be formed to bring about policy change.

11. CONCLUSIONS It seems almost inevitable that motor traffic will grow well beyond the immediate future. This will impose additional strains on the road infrastructures of most countries. The associated impacts of longer commutes, and higher freight distribution costs have to be borne somewhere in the system. On top of that there are now additional costs that have recently come to the fore, most notably in the U.S., linked to security. While congestion has traditional economic costs, an excessively congested network also has costs in terms of impeding response to incidents. Reacting to single events by the emergency services often poses problems at present, but in context of a major catastrophic occurrence, large scale evacuation for example from many city centres would be chaotic. This realisation may accelerate the adoption of more efficient traffic pricing. Those responsible for road provision are moving to make use of fiscal tools to manage the normal economic concerns over traffic growth. One can debate why there has recently been an up-surge of interest in road pricing. It may reflect a sudden appreciation that roads are really no different to other goods and can be treated in similar ways. Markets may not be possible but at least some effort can be made to make more rational use of road space. It may also reflect a crisis. Other measures have either been designed to increase the available capacity with no form of allocation mechanism other than queuing, or have involved providing alternatives that road users could have used before but declined. Both of these broad approaches have been found wanting and have become ever more expensive to pursue. The shift in emphasis is thus towards treating the root course of the problem rather than the symptoms. One would like to think, therefore, that the systems of Road Pricing that have actually been deployed, ranging from the pioneering Area Licensing Scheme in Singapore to the system in Central London, were brought about by some sort of intellectual conquest and by the victory of ideas. In practice, it seems more to be a question of pragmatic necessity on the part of politicians when confronted with the failure of engineering and planning strategies. Economics is after all the “Dismal Science,” and it can sometimes take time before politicians realise that what superficially may appear as bad tasting medicine is actually very good for society. This chapter is in no way comprehensive and it barely scratches some of the basic theory and the more recent contributions. This book constitutes an update of the state of the art of the latest research and empirical evidence.

The Rationale for Road Pricing

21

NOTES 1. The account given in this chapter is decidedly non-technical. For a slightly more vigorous overview of many of the topics addressed here see Lindsey and Verhoef (2002) and Verhoef (1996). Button and Verhoef (1998) contains papers covering both normative and positive contributions on Road Pricing. 2. Speed-flow relationships have been studied by engineers, physicists and mathematicians since the 1930s. 3. Since vehicles differ in their impact on traffic flows the passenger car unit (PCU) equivalent is used in this chapter as a reflection of this. This measures weights vehicular impacts on the traffic flow to give a standard measure based on the congestion caused by a standard car. Given the variations in vehicle size, road design and other features, the weight used in PCU measures varies between countries. 4. In any case, it would be necessary to use dynamic models to study the hyper-congested branch of the speed-flow function (Lindsey & Verhoef, 2000). 5. There are in principle other ways of achieving the optimal level of congestion. These include subsidies, marketable permits, and command and control policies, as well as other transport policies, such as improvement and/or subsidy of public transport. None of these is the subject of this volume and so they shall be ignored. 6. The use of revenues is central to a congestion pricing scheme as it typically determines the final distributional impacts and the acceptability by the public. 7. A comprehensive account of the work of the French engineers can be found in Ekelund and H´ebert (1999). 8. One should always be wary of any literature that juxtapositions road pricing and the 1924, 1929, 1932 or 1952 editions of The Economics of Welfare. Road pricing is not in them. 9. This does assume a perfectly inelastic demand curve for total trips generated which deviates from much of the work by the French engineers. In other words for simplicity he took no account of what has more recently been rediscovered by traffic engineers, although well known to economists, as “latent demand.” 10. Academic articles on club goods often revolve around the “ski-lift” problem where there are day passes giving access but then allocation issues arise. 11. This is the de facto situation found with some of the high occupancy/toll lanes found in California, although these are slightly more complex because high-occupancy allows the toll to be avoided (Li, 2001). The idea there is marketed under the notion of “value pricing” on the basis that the toll for using the high occupancy/toll lanes reflects the value they confer. The system is explained in detail in Chapter 12 of this volume. 12. Much of the work on parking charges has tended to ignore this second element, and treat it as cost at the end of a trip (Calthrop et al., 2000; Glazer & Niskanen, 1992) and neglect the effect this price has on the demand for cruising. 13. The high fuel duties in the U.K. for example have been defended by different governments on environmental grounds. 14. The high occupancy/toll lanes in California may be seen as a practical reflection of this. 15. The operating costs for the London Congestion Charging scheme are certainly substantial. The details of the scheme are presented in Chapter 11 of this volume.

22

KENNETH BUTTON

REFERENCES Arnott, R., & Kraus, M. (1998). When are anonymous congestion charges consistent with marginal cost pricing? Journal of Public Economics, 67, 45–64. Arnott, R., de Palma, A., & Lindsey, R. (1988). Schedule delay and departure time decisions with heterogeneous commuters. Transportation Research Record, 1197, 56–67. Arnott, R., de Palma, A., & Lindsey, R. (1990a). Economics of a bottleneck. Journal of Urban Economics, 27, 111–130. Arnott, R., de Palma, A., & Lindsey, R. (1990b). Departure time and route choice for routes in paralell. Transportation Research B, 24, 209–228. Arnott, R., de Palma, A., & Lindsey, R. (1992). Route choice with heterogeneous drivers and group-specific congestion costs. Regional Science and Urban Economics, 22, 71–102. Arnott, R., de Palma, A., & Lindsey, R. (1993). A structural model of peak-period congestion: A traffic bottleneck with elastic demand. American Economic Review, 83, 161–179. Arnott, R., de Palma, A., & Lindsey, R. (1994). The welfare effects of congestion tolls with heterogeneous commuters. Journal of Transport Economics and Policy, 28, 139–161. Arnott, R., de Palma, A., & Lindsey, R. (1998). Recent developments in the bottleneck model. In: K. J. Button & E. T. Verhoef (Eds), Road Pricing, Traffic Congestion and the Environment: Issues of Efficiency and Social Feasibility (pp. 79–112). Cheltenham: Edward Elgar. Arnott, R., & Yan, A. (2000). The two-mode problem: Second-best pricing and capacity. Review of Urban and Regional Development Studies, 12, 170–199. Barzel, Y. (1974). A theory of rationing by waiting. Journal of Law and Economics, 17, 73–95. Bertrand, T. (1977). Second-best congestion taxes in transportation systems. Econometrica, 45, 1703–1715. Buchanan, J. M. (1956). Private ownership and common usage: The road case re-examined. Southern Economic Journal, 22, 305–316. Buchanan, J. M. (1965). An economic theory of clubs. Economica, 32, 1–14. Button, K. J. (1993). Transport economics. Aldershot: Edward Elgar. Button, K. J., & Verhoef, E. T. (Eds) (1998). Road pricing, traffic congestion and the environment: Issues of efficiency and social feasibility. Cheltenham: Edward Elgar. Brownstone, D., & Small, K. (2003). Valuing time and reliability: Assessing the evidence from road pricing demonstrations. Paper presented at the American Economic Association 2003 Meeting, January 3–5, Washington, DC. www.uctc.net/papers/668.pdf. Calfee, J. E., & Winston, C. (1998). The value of automobile travel time: Implications for congestion policy. Journal of Public Economics, 69, 83–102. Calthrop, E., Proost, S., & Van Dender, K. (2000). Parking policies and road pricing. Urban Studies, 37, 63–76. Chia, N. C., Tsui, A., & Whalley, J. (2001). Ownership and use taxes as congestion correcting instruments. Cambridge, MA: NBER Working Paper 8278. Chu, X. (1995). Endogenous trip scheduling: The Henderson approach reformulated and compared to the Vickrey approach. Journal of Urban Economics, 37, 324–343. Chu, X. (1999). Alternative congestion pricing schedules. Regional Science and Urban Economics, 29, 697–722. Dawson, J., & Catling, I. (1986). Electronic road pricing in Hong Kong. Transportation Research A, 20, 129–134. de Palma, A., & Lindsey, R. (1998). Information and usage of congestable facilities under different pricing regimes. Canadian Journal of Economics, 31, 666–692.

The Rationale for Road Pricing

23

Dupuit, J. (1844). On the measurement of the utility of public works. Annales des Ponts et Chauss´ees M’moires et Documents 2nd Series, 8, 332–375. (Translated by R. H. Barback (1952) International Economic Papers, 2, 83–110.) Edelson, N. E. (1971). Congestion tolls under monopoly. American Economic Review, 61, 872–882. Ekelund, R. B., & H´ebert, R. F. (1999). Secret origins of modern microeconomics: Dupuit and the engineers. Chicago: Chicago University Press. Evans, Alan W. (1992). Road congestion The diagrammatic analysis. Journal of Political Economy, 100, 211–217. Evans, Andrew W. (1992). Road congestion pricing: When is it a good policy? Journal of Transport Economics and Policy, 26, 213–243. Foster, C. D. (1975). A note on the distributional effects of road pricing: A comment. Journal of Transport Economics and Policy, 9, 186–187. Giuliano, G. (1992). An assessment of the political acceptability of congestion pricing. Transportation, 19, 335–358. Glazer, A., & Niskanen, E. (1992). Parking fees and congestion. Regional Science and Urban Economics, 22, 123–132. Goodwin, P. (1989). The rule of three: A possible solution to the political problem of competing objectives for road pricing. Traffic Engineering and Control, 30, 495–497. Goodwin, P. (1990). How to make road pricing popular. Economic Affairs, 10, 6–7. Hague Consulting Group, Accent Marketing and Research, Department of the Environment, Transport and the Regions (1999). The value of travel time on U.K. roads. The Hague, Netherlands. Hau, T. D. (1990). Electronic road pricing: Developments in Hong Kong. Journal of Transport Economics and Policy, 24, 203–214. Hau, T. D. (1998). Congestion pricing and road investment. In: K. J. Button & E. T. Verhoef (Eds), Road Pricing, Traffic Congestion and the Environment: Issues of Efficiency and Social Feasibility (pp. 39–78). Cheltenham: Edward Elgar. Henderson, J. V. (1974). Road congestion: A reconsideration of price theory. Journal of Urban Economics, 1, 346–365. Hensher, D. A. (1997). Value of travel time savings in personal and commercial automobile travel. In: D. L. Greene, D. W. Jones & M. A. Delucchi (Eds), The Full Costs and Benefits of Transportation (pp. 245–227). New York: Springer-Verlag. Hills, P. (1993). Road congestion pricing: When is it a good policy? A comment. Journal of Transport Economics and Policy, 27, 91–99. Jara-D´ıaz, S. R. (2000). Allocation and valuation of travel-time savings. In: D. A. Hensher & K. J. Button (Eds), Handbook of Transport Modelling (pp. 303–320). Oxford: Pergamon. Keeler, T. E., & Small, K. A. (1977). Optimal peak-load pricing, investment and service levels on urban expressways. Journal of Political Economy, 85, 1–25. Knight, F. (1924). Some fallacies in the interpretation of social costs. Quarterly Journal of Economics, 38, 582–606. Kraus, M. C. (1981). Indivisibilities, economies of scale, and optimal subsidy policy for freeways. Land Economics, 57, 115–121. Lam, T. C., & Small, K. A. (2001). The value of time and reliability: Measurement from a value pricing experiment. Transportation Research E, 37, 231–251. Layard, R. (1977). The distributional effects of congestion taxes. Econometrica, 44, 297–304. Lindsey, R., & Verhoef, E. T. (2000). Congestion modeling. In: D. A. Hensher & K. J. Button (Eds), Handbook of Transport Modelling (pp. 353–373). Oxford: Pergamon.

24

KENNETH BUTTON

Lindsey, R., & Verhoef, E. T. (2002). Traffic congestion and road pricing. In: K. J. Button & D. A. Hensher (Eds), Handbook of Transport Systems and Traffic Control (pp. 77–104). Oxford: Pergamon. Li, J. (2001). Explaining high-occupancy-toll lane use. Transportation Research D, 6, 61–74. Liu, L. N., & McDonald, J. F. (1999). Economic efficiency of second-best congestion pricing schemes in urban highway systems. Transportation Research B, 33, 157–188. Mackie, P., Jara-D´ıaz, S., & Fowkes, A. (2001). The value of travel time savings in evaluation. Transportation Research E, 37, 91–106. Mackie, P., Wardman, M., Fowkes, A., Whelan, G., Nellthorp, J., & Bates, J. (2003). Values of travel time savings in the UK: Report to the Department of Transport. Institute for Transport Studies, University of Leeds. www.its.leeds.ac.uk/working/downloads/VOTFull.pdf. Marchand, M. (1968). A note of optimal tolls in an imperfect environment. Econometrica, 36, 575–581. Mills, D. E. (1981). Ownership arrangements and congestion-prone facilities. American Economic Review, 71, 493–502. Minard, J. (1850). Notions e´ l´ementaires d’´economie politique appliqu´ee aux travaux publics. Annales des Ponts et Chauss´es: M´emoires et Documents, 2nd series, 1–125. Ministry of Transport (1964). Road Pricing: The Economic and Technical Possibilities. London: Her Majesty’s Stationary Office. Mohring, H. (1970). Urban highway investments. In: R. Dorfman (Ed.), Measuring Benefits of Government Investment (pp. 231–275). Washington, DC: Brookings Institution. Mohring, H., & Harwitz, M. (1962). Highway Benefits. Evanston: Northwestern University Press. Morrison, S. A. (1986). A survey of road pricing. Transportation Research A, 20, 87–97. Newbery, D. (1989). Cost recovery from optimally designed roads. Econometrica, 56, 165–185. Newbery, D. (1990). Pricing and congestion: Economic principles relevant to pricing roads. Oxford Review of Economic Policy, 6, 22–38. Newbery, D. (1994). The case for a public road authority. Journal of Transport Economics and Policy, 28, 235–253. Pigou, A. (1920). The Economics of Welfare. London: Macmillan. Richardson, H. (1974). A note on the distributional effects of road pricing. Journal of Transport Economics and Policy, 8, 82–85. Santos, G., & Rojey, L. (2004). Distributional impacts of road pricing: The truth behind the myth. Transportation, 31, 21–42. Sharp, C. H. (1966). Congestion and welfare, an examination of the case for a congestion tax. Economic Journal, 76, 806–817. Sherman, R. (1971). Congestion interdependence and urban transit fares. Econometrica, 39, 565–576. Small, K. A. (1983). The Incidence of congestion tolls on urban highways. Journal of Urban Economics, 13, 90–111. Small, K. A. (1992). Using the revenues from congestion pricing. Transportation, 19, 359–381. Small, K. (1999). Project evaluation. In: J. G´omez-Ib´an˜ ez, W. Tye & C. Winston (Eds), Essays in Transportation Economics and Policy: A Handbook in Honor of John R. Meyer (pp. 137–177). Washington, DC: Brookings Institution. Small, K. A., & G´omez-Ib´an˜ ez, J. A. (1998). Road pricing for congestion management: The transition from theory to policy. In: K. J. Button & E. T. Verhoef (Eds), Road Pricing, Traffic Congestion and the Environment: Issues of Efficiency and Social Feasibility (pp. 213–247). Cheltenham: Edward Elgar. Small, K. A., & Chu, X. (1997). Hypercongestion. Department of Economics, University of California at Irvine, Working Paper 97-98-12.

The Rationale for Road Pricing

25

Small, K. A., & Yan, J. (2001). The value of “value pricing” of roads: Second-best pricing and product differentiation. Journal of Urban Economics, 49, 310–336. Starkie, D. (1982). Road indivisibilities: Some observations. Journal of Transport Economics and Policy, 16, 259–266. Strotz, R. H. (1965). Urban transportation parables. In: J. Margolis (Ed.), The Public Economy of Urban Communities (pp. 127–169). Washington, DC: Resources for the Future. Verhoef, E. T. (1996). The economics of regulating road transport. Cheltenham: Edward Elgar. Verhoef, E. T. (1999). Time, speed flows and densities in static models of road traffic congestion and congestion pricing. Regional Science and Urban Economics, 29, 341–369. Verhoef, E. T. (2000). The implementation of marginal external cost pricing in road transport: Long run vs. short run and first-best vs. second-best. Papers in Regional Science, 79, 307–332. Verhoef, E. T. (2002a). Second-best congestion pricing in general static transportation networks with elastic demands. Regional Science and Urban Economics, 32, 281–310. Verhoef, E. T. (2002b). Second-best congestion pricing in general networks – heuristic algorithms for finding second-best optimal toll levels and toll points. Transportation Research B, 36, 707–729. Verhoef, E. T. (2003). Speed-flow relations and cost functions for congested traffic: Theory and empirical analyses. Tinbergen Institute, Discussion Paper TI 2003-064/3, Amsterdam-Rotterdam. www.tinbergen.nl/discussionpapers/03064.pdf. Verhoef, E. T., Nijkamp, P., & Rietveld, P. (1995). The economics of regulatory parking policies: The (im)possibilities of parking policies in parking regulation. Transportation Research A, 29, 141–156. Verhoef, E. T., Nijkamp, P., & Rietveld, P. (1996). Second-best congestion pricing: The case of an untolled alternative. Journal of Urban Economics, 40, 279–302. Verhoef, E. T., Rouwndal, J., & Rietveld, P. (1999). Congestion caused by speed differences. Journal of Urban Economics, 53, 533–556. Vickrey, W. S. (1959). Statement on the Pricing of Urban Street Use. Hearings, U.S. Congress Joint Committee on Metropolitan Washington Problems, November 11, Washington, DC. Vickrey, W. S. (1963). Pricing in urban and suburban transport. American Economic Review, 53, 452–465. Vickrey, W. S. (1968). Congestion charges and welfare. Journal of Transport Economics and Policy, 2, 107–118. Vickrey, W. S. (1969). Congestion theory and transport investment. American Economic Review, 59, 251–260. Walters, A. A. (1961). The theory and measurement of private and social costs of highway congestion. Econometrica, 29, 676–697. Wardman, M. (1998). The value of travel time: A review of British evidence. Journal of Transport Economics and Policy, 32, 285–316. Whelan, G., & Bates, J. (2001). Market segmentation analysis. ITS Working Paper 565. Institute for Transport Studies, University of Leeds. www.its.leeds.ac.uk. Wheaton, W. C. (1978). Price-induced distortions in urban highway investment. Bell Journal of Economics and Management Science, 9, 622–632.

This Page Intentionally Left Blank

2.

SECOND-BEST PRICING FOR IMPERFECT SUBSTITUTES IN URBAN NETWORKS

Jan Rouwendal and Erik T. Verhoef 1. INTRODUCTION The basic economic intuition behind marginal cost pricing is deceptively simple. When prices are equal to marginal social costs, consumers would expand consumption (be it road trips or any other good) up to the point where the benefits of the last unit consumed have become equal to these marginal social costs. The implied equality of marginal benefits and marginal social costs secures that the social surplus, defined as the difference between total benefits and total costs, is maximised. This social surplus is often considered as an appropriate indicator for social welfare, and its maximisation is a condition for economic efficiency to prevail. As explained in Chapter 1 of this volume, the “price” for the use of a road would in the first place include the costs borne directly by the road user, such as fuel expenses and the value of the travel time spent on the road – the marginal private costs. Besides, a toll may be charged for the use of the road, which also adds to the price. Marginal cost pricing in road transport therefore means that the toll should reflect the marginal external costs: the marginal costs of road use other than the marginal private cost. For a congested road, this includes the value of time losses (and travel time uncertainty) imposed on other road users, as well as the value of emissions, noise, and accident risks created. Such a tax, equal to marginal

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 27–60 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09002-X

27

28

JAN ROUWENDAL AND ERIK T. VERHOEF

external costs, is often referred to as a “Pigouvian tax,” after its spiritual father Arthur Pigou (1920). Based on this intuition, transport economists and engineers alike have for long advocated the use of road pricing schemes for the management of road traffic congestion (e.g. Knight, 1924; Pigou, 1920; Vickrey, 1963; Walters, 1961; etc.). Even if road capacities can be adjusted to cope with congestion, capacity policies alone will typically not be sufficient to reach an optimal outcome, and the use of congestion pricing remains warranted.1 As pricing not only succeeds in reducing demand per se, but in doing so also in maintaining exactly those trips for which the benefits, as reflected by the willingness to pay, are highest, it will typically lead to higher welfare gains than what could be realised by alternative, non-price-based demand management schemes, such as number plate policies as currently in operation in Athens and Mexico City. This explains much of the analysts’ preference for this type of policy. Moreover, the principle of marginal cost pricing securing efficiency carries over to more complex situations, including congestion pricing on full networks (rather than a single road), heterogeneous users, and time-varying congestion. This suggests that the practical application of the principle of optimal congestion pricing in reality, although yielding additional practical complications, would not create additional conceptual difficulties, compared to the basic mechanisms as identified in textbook models. However, this would – unfortunately – be true only under rather stringent and often unrealistic assumptions. Two of these unrealistic assumptions are discussed next. A first one concerns optimality of the pricing instrument, which should allow the road regulator to differentiate prices such that every individual road user indeed faces a toll exactly equal to the marginal external costs he or she causes. This requires tolls to be differentiated, at least, by time of day, route followed and hence the length of the trip, type of vehicle (e.g. small vs. large cars or passenger cars vs. trucks) and its state of maintenance (for pollution), driving style (for pollution, noise and accident risks), etc. It also requires that these issues can be monitored perfectly by the regulator. Although emerging technologies for electronic charging are likely to allow for a more sophisticated toll differentiation in the near future, it remains to be seen to which extent such complex pricing will indeed be put into practice. A second assumption concerns efficiency in all markets related to the transport market under consideration, where a market is “related” to the transport market if its equilibrium is indirectly affected by transport policies, and its equilibrium is inefficient if, again, prices are not equal to marginal social costs. The existence of such related markets is the rule rather than the exception because transport demand is often a derived demand, where the “consumption” of transport is often no goal in itself but serves to enable a market transaction between spatially separated

Second-Best Pricing for Imperfect Substitutes in Urban Networks

29

suppliers and demanders of a certain good or service. For example, commuting typically does not yield any benefit in itself, but serves to allow people to supply labour services at a different location, the work place, than their residences. Likewise, freight traffic enables a transaction between a supplier of goods and demanders, final consumers or firms demanding intermediates, who are separated in space. Whenever these related markets do not function properly, the simple policy rule of setting taxes equal to marginal external costs will no longer be a truly optimal choice. Instead, transport taxes should optimally deviate from this rule, so as to correct for the inefficiency in the related market without of course sacrificing too much efficiency in the transport market itself. Important reasons why prices in related markets may deviate from marginal social costs include, amongst others, the existence of market power, unpriced externalities (such as environmental pollution), or distortionary taxes (for instance on the labour market in the case of commuter traffic). Inefficiencies are therefore likely to be the rule rather than the exception. Despite the attention that first-best pricing has traditionally received in most of the literature on transport pricing, it will be clear from the above that second-best pricing is in fact often the most relevant case from a practical perspective. In recent years, there has consequently been an upsurge of studies in second-best congestion pricing. Lindsey and Verhoef (2001) provide a recent review of this literature, and discuss second-best pricing resulting from network issues, heterogeneity of users, dynamic constraints on toll flexibility (“step tolls”), uncertainty in travel times (in relation with real time information provision), relations with other sectors (notably labour markets), the simultaneous existence of multiple externalities (e.g. congestion and pollution), congestion pricing by private road operators, and through interactions with sub-optimal capacity choice. Second-best congestion pricing through network effects will be specially important in urban contexts, where dense networks will often prevent the application of optimally differentiated tolls on each and every link of the network. This chapter considers these issues. A first aim is to provide an overview of the basic economic ins and outs of second-best congestion pricing in networks by reviewing the classic two-route problem in Section 2. Most of the underlying literature conveniently, but somewhat unrealistically, treats different routes between two points (“nodes”) in a network as pure substitutes, and therefore applies so-called Wardropian equilibrium principles (Wardrop, 1952) in the determination of network equilibria (with and without prices). This means that even the smallest equilibrium price difference would make a user choose the cheapest route. In reality, people may have so-called idiosyncratic preferences over different routes, meaning that they would have different preferences for competing routes when these would have equal trip prices. It is therefore of

30

JAN ROUWENDAL AND ERIK T. VERHOEF

interest to study how this would affect the optimal design and welfare properties of second-best congestion pricing. Moreover, by allowing different routes to be imperfect substitutes, a framework is developed that can also be used to study second-best congestion pricing from a multi-modal perspective. Addressing these issues is the second aim of the paper. A theoretical model is developed for the study of second-best congestion pricing in networks where parallel roads are imperfect substitute routes, and its properties are illustrated with the help of a simulation model.

2. SECOND-BEST CONGESTION PRICING IN TRANSPORT NETWORKS One of the most widely studied instances of second-best congestion pricing concerns the case where not all links in a uni-modal transport network can be priced. This case has great practical relevance for actual policy making. For instance, pay-lanes as currently in operation at various places in the U.S. belong to this category of second-best pricing policies, since unpriced lanes are available as a direct substitute for the pay-lanes. In other (planned) road pricing schemes, be it based on marginal cost pricing principles or motivated by the desire to generate revenues, prices will often be charged on a limited number of links in a network only (e.g. toll roads, or toll cordons). Furthermore, the case is of interest as it can be used to illustrate some more general principles of second-best pricing that will be relevant for other types of second-best pricing too. These principles include the fact that in second-best pricing, the na¨ıve use of Pigouvian taxes, set equal to the direct marginal external costs, will generally not be optimal, and the fact that second-best pricing itself creates distortions, which affect the optimal second-best price itself.

2.1. The Classic Two-Route Problem The simplest version of the problem at hand concerns the two-route problem, where an untolled alternative road is available parallel to a toll road. This problem has for instance been studied by L´evy-Lambert (1968), Marchand (1968), and more recently also by Braid (1996), Verhoef et al. (1996), and Liu and McDonald (1999). Figure 1 illustrates the basic set-up of the problem. Two parallel, congested roads of given capacities connect an origin (A) and a destination (B). On one of these (road T), a congestion toll can be set, whereas the other road (U) remains untolled. The roads are pure substitutes, meaning that equilibrium “generalised

Second-Best Pricing for Imperfect Substitutes in Urban Networks

31

Fig. 1. The Classic Two-Route Problem.

prices” p (the sum of monetised travel costs, c, plus a toll, ␶, if levied) should be equal for both roads when both are used (this is the “Wardropian” equilibrium principle mentioned earlier). Drivers are nearly identical (they have the same value of time, drive the same type of cars, etc.), with the exception of their willingness to pay for a trip, which varies over drivers whenever demand is not perfectly elastic. The two central questions are then: at what level should the second-best toll be set, and what are the welfare effects compared to the first-best option of tolling both roads? An important result from the studies looking at these problems is that the second-best tax, set so as to maximise social welfare given the persistence of the second-best distortion of leaving the other road untolled, is typically below the optimal Pigouvian tax that would be set on the same road under first-best pricing, but also below the marginal external costs on the toll road in the second-best optimum. In addition, the second-best toll yields considerably smaller welfare gains than first-best pricing. It is important to understand why the second-best toll should ideally be below the marginal external congestion costs on the tolled road. The reason is that the congestion charge on the toll road brings both good and bad news from the perspective of efficiency. The good news is that initially excessive free-market congestion on the toll road is reduced through the toll. The bad news is that some people who are priced off this road will be diverted to the already congested untolled road, thus aggravating the congestion there. The second-best toll trades off the good news against the bad news. The good news alone would ask for a tax equal to marginal external costs for the reasons given in the introduction, but as soon as the bad news is relevant (i.e. when there is congestion on the untolled road, and when a by-product of the toll is indeed to divert traffic to this untolled road), a downward adjustment of the toll is optimal. As a corollary, a policy of simply ignoring the second-best distortion, and setting the toll equal to the marginal external congestion costs on the tolled road, would lead to a social surplus below that in the second-best optimum, exactly because the policy is “na¨ıve” and ignores the spill-overs on the free lane. Indeed, such “quasi first-best pricing” may even lead to a welfare loss compared to the no-toll outcome, which would certainly occur when the second-best optimal tax is negative, as in some examples in Verhoef et al. (1996).

32

JAN ROUWENDAL AND ERIK T. VERHOEF

The second-best optimal tax rule for this classic two-route problem reflects the above mentioned trade-off between the good and the bad news from the toll. Verhoef et al. (1996) express it as: ␶ = mecT − mecU ·

−D  c U − D 

where mecT (mecU ) denotes the marginal external costs on the tolled (untolled) road, D the slope of the inverse demand function, and C U that of the cost function on the untolled road; all evaluated at the second-best equilibrium. The tax rule clearly shows how the second-best toll is equal to what will be called the “direct marginal external costs” of drivers on road T (mecT ),2 minus a fraction of those on the untolled road U (mecU ) (because D = 0 and c U ≥ 0, the weight for mecU is always between 0 and 1). Further intuition can be obtained by considering a few instances where the fraction mentioned would approach its possible extreme values of 0 (meaning that ␶ = mecT ) or 1 (meaning that τ = mecT − mecU ). Intuitively, the fraction should approach zero when there is no diverting of traffic to the untolled road in the second-best optimum due to marginal changes in ␶. This is the case when either D  → 0 (regardless of ␶, the untolled road will in equilibrium be filled with drivers up to that level NU that equates cU (NU ) to the given value of D), or when c U → ∞ (the cost function for road U is perfectly inelastic, and NU is given). In these cases, the regulator can ignore spill-overs of tolling on road T upon road U, simply because these will not occur. The fraction, in contrast, approaches unity when either D  → −∞ or when c U → 0. In these cases, the second-best toll only tries to optimise route split, by internalising the difference between marginal external costs for the two routes, and fully ignores the effects on overall demand. This makes sense when demand is perfectly inelastic (D  → −∞), in which case overall demand would not respond to marginal changes in ␶. It also makes sense if every user priced off road T immediately switches to road U, in the situation where c U → 0 implies that private costs of using road U are independent of the route split, so that the overall demand is given and defined by the equality of D and cU . It should be noted, however, that c U → 0 at the same time implies that mecU = 0, so that the tax rule in this case effectively becomes ␶ = mecT . Second-best pricing then becomes identical to first-best pricing because spill-overs do not result in additional welfare losses. The relative size of the welfare losses from second-best pricing compared to first-best pricing depends on various elasticities. Verhoef et al. (1996) for example demonstrate that if route U is particularly congested, the optimal second-best toll can be negative, which typically implies relatively low efficiency gains due to pricing (even zero when the optimal second-best congestion toll happens to be zero).

Second-Best Pricing for Imperfect Substitutes in Urban Networks

33

More generally they show that the efficiency gains depend on the relative free-flow travel times and capacities of the two routes, and increase if the tolled road becomes relatively more attractive. Furthermore, the gains depend on the price elasticity of travel demand, in combination with the cost parameters. Specifically, if demand approaches perfect inelasticity, the efficiency gains from second-best pricing may approach those from first-best pricing if the roads differ in free-flow travel times. The only factor determining efficiency in that case would be the route split, which can be optimised with one toll only. However, if the roads have identical free-flow travel times, as would often be the case for pay-lanes, the efficiency gains from second-best pricing may nevertheless fall to zero, as the free-market route split (without tolling) may then already be optimal. Equalisation of private costs then implies the optimal equalisation of marginal costs. The second-best optimal toll approaches zero and therefore induces no welfare effects whatsoever. These findings illustrate that for second-best pricing often “anything goes,” in the sense that the relative efficiency of second-best policies is often impossible to predict without prior detailed knowledge of the case at hand. Liu and McDonald (1998) constructed a model descriptive of one of the California road pricing demonstration projects (State Route 91 in Orange County), and found that the efficiency gains from second-best congestion pricing are in the order of 10% of the potential gains from first-best pricing. Bearing in mind the exposition above, this should in fact not be too disappointing, given the similarity of the pay-lanes and the free lanes, and the inelasticity of demand.

2.2. Beyond the Classic Two-Route Problem The classic two-route problem has received considerable attention for at least two reasons. Firstly, it is probably the simplest possible and most transparent setting for illustrating some of the main economic insights into the ins and outs of second-best pricing policies, which often also carry over to other, more complicated settings.3 Secondly, it appears to be a reasonable approximation for many congestion pricing demonstration projects that are currently in operation. The basic model is certainly rather restrictive, and a number of extensions have been made to investigate some further properties of the problem and its solution. For example, static models, as discussed above, may underestimate the efficiency gains from second-best one-route tolling because they ignore some alternative ways in which second-best pricing can affect driver behaviour. Braid (1996) and de Palma and Lindsey (2000) allow for trip-timing adjustments by considering time-varying tolls in the Vickrey (1969) bottleneck model, applied to the same two-parallel-routes network. They find that second-best tolling yields higher absolute efficiency gains than in the

34

JAN ROUWENDAL AND ERIK T. VERHOEF

static model, and a greater fraction of the first-best efficiency gains. This is because the toll not only curbs excessive total usage, but also eliminates queuing on the tolled route. The efficiency of second-best tolling may also be improved when drivers have different values of travel time. Users with a high value of time would then choose the tolled connection and benefit relatively strongly from the reduced travel time, while those with a low value of time would not suffer too much from the increased travel on the untolled connection that they prefer. Small and Yan (1999) and Verhoef and Small (1999) find that the efficiency of pay-lanes relative to the first-best optimum is indeed higher than in the equivalent model with no heterogeneity in the value of time. Finally, optimal second-best tolls and the associated welfare gains have also been investigated for larger networks on which not all links can be tolled. Verhoef (1998) derives optimal static tolls on any subset of links (including parking spaces) in an arbitrary network. The toll formulae, which are quite complicated, include terms reflecting marginal external costs on other links, and weights that depend on various demand and cost elasticities. The solution for the classic two-route problem presented above is a special – but transparent – illustrative case of these general formulae.

3. THE TWO-ROUTE PROBLEM WITH IMPERFECT SUBSTITUTES In this section the network of Fig. 1 is further discussed. As announced, this network will be reconsidered for the case where the two routes are imperfect substitutes. There are, as before, two nodes, A and B, connected by two links, which are now denoted 1 and 2. Besides tolls, cases where capacities on these links can vary will also be considered, and second-best issues arise if at least one of these four policy variables is restricted. The links may refer to roads, as in the classic two-route problem, and secondbest problems emerge, for instance, if only one link is unpriced or if it is priced sub-optimally. The links may also represent different modes, which is what will be considered in Section 4. For competing modes it is common to regard these as imperfect substitutes. In addition, the type of cost functions for users, as well as the cost of providing capacity for operators, may then be different between the two links. Arnott and Yan (2000) have recently provided a review of the problems that arise if travel with one mode is subsidised, while the policy maker can choose the price of travelling on the other mode and capacity on both links. This “two-mode problem” appears to be hard to solve in general terms.

Second-Best Pricing for Imperfect Substitutes in Urban Networks

35

If two roads are perfect substitutes, any price difference implies that one route will not be used. However, if they are imperfect substitutes, price differences do not necessarily imply that demand for the most expensive route falls to 0. In reality, two roads that connect the same pair of towns need not necessarily be considered as perfect substitutes. For instance, consumers may prefer one route to the other because their residential location in A is closer to the point where route 1 starts, because their destination in B is closer to where route 2 ends, or because they have a preference over highways (easier driving) or secondary roads (nicer scenery) per se. The random term in the logit model presented below may be interpreted as reflecting such idiosyncratic differences.

3.1. The General Model In what follows a model that can cover all the aforementioned cases is constructed. In the general model the social planner attempts to maximise the social surplus, which is defined as consumer surplus plus toll revenues minus cost of infrastructure capacity: SS = CS + TR − K

(1)

The Marshallian notion of consumer surplus is used. For a single good, it is the area under the inverse demand curve and above the prevailing price, between the origin and the equilibrium quantity. If two (or more) goods are involved, the total consumer surplus is not usually the sum of the consumer surpluses for the individual goods as they could be calculated on the basis of the two (or more) demand functions. The reason is that the demand curves are interrelated, and when the price of one good is increased, the demand curves for the other good(s) and the associated consumer surplus(es) change(s). However, total surplus can still be defined unambiguously if the demand functions are independent of income, and that is what will be assumed here. The situation with two goods, indicated by suffixes 1 and 2, is considered. The quantities q demanded can be derived from the consumer surplus by taking the first derivatives and putting a minus sign in front: qi = −

∂CS , ∂p i

i = 1, 2.

(2)

Tolls are denoted ␶, and have to be multiplied by the number of trips in order to find toll revenues: TR = TR1 + TR2 = ␶1 q 1 + ␶2 q 2

(3)

36

JAN ROUWENDAL AND ERIK T. VERHOEF

Cost depends on capacity: K = K 1 (cap1 ) + K 2 (cap2 )

(4)

Constant returns to scale are not imposed. Capacity is allowed to be vector valued (e.g. the frequency of trains and the number of seats per train may be relevant dimensions of capacity), but it will usually be assumed to be a scalar (e.g. road width). Only cases in which a user equilibrium is obtained will be considered. User equilibrium is found where the inverse demand, as represented by the full price p for both routes, is equal to the sum of private travel cost c, which depends on capacity and number of trips, and toll ␶: p i = c i (capi , q i ) + ␶i ,

i = 1, 2.

(5)

The first-best problem is the maximisation of social surplus by choosing the two tolls and capacities under the side condition that a user equilibrium should be obtained. The Lagrangian is: L = SS + ␭1 (p 1 − c 1 − ␶1 ) + ␭2 (p 2 − c 2 − ␶2 )  ∞ = q i (x)dx + ␶i q i (p i ) − K i (capi ) i

pi

+ ␭i (p i − c i (q i (p i ), capi ) − ␶i )

(6)

The first order conditions with respect to p1 , p2 , ␶1 , ␶2 and cap1 and cap2 are:     ∂q ∂q ∂c 1 ∂q 1 ∂c 2 ∂q 2 −q1 + ␶1 1 + ␶2 2 + ␭1 1 − + ␭2 − =0 ∂p 1 ∂p 1 ∂q 1 ∂p 1 ∂q 2 ∂p 1     (7) ∂c 1 ∂q 1 ∂c 2 ∂q 2 ∂q ∂q + ␭2 1 − =0 −q2 + ␶1 1 + ␶2 2 + ␭1 − ∂p 2 ∂p 2 ∂q 1 ∂p 2 ∂q 2 ∂p 2 q i = ␭i ,

i = 1, 2.

∂c i ∂K i = −␭i , ∂capi ∂capi

(8) i = 1, 2.

(9)

In the first-best optimum, all conditions (5) and (7)–(9) are satisfied. In second-best situations one or more of the four Eqs (8) and (9) are not satisfied. Equations (5) and (7) are always valid, since they are the consequence of the basic requirement that there should be a user equilibrium. Equation (8) are the requirements for optimal tolling. They state that under optimal tolling the Lagrange multipliers will be equal to the associated numbers of trips.4 If Eq. (8) are valid, the Lagrange multipliers can easily be eliminated from

Second-Best Pricing for Imperfect Substitutes in Urban Networks

37

the system of equations. This implies a substantial simplification. Equation (7) can then be written as:     ∂c 1 ∂q 1 ∂c 2 ∂q 2 + ␶2 − q 2 =0 ␶1 − q 1 ∂q 1 ∂p 1 ∂q 1 ∂p 1     (10) ∂c 1 ∂q 1 ∂c 2 ∂q 2 ␶1 − q 1 + ␶2 − q 2 =0 ∂q 1 ∂p 2 ∂q 1 ∂p 2 This implies that both tolls should be equal to the external congestion effect.5 Equation (9) becomes: ∂K 1 ∂c 1 = −q1 ∂cap1 ∂cap1 ∂K 2 ∂c 2 = −q2 ∂cap2 ∂cap2

(11)

Equation (11) states that the marginal cost of capacity should be equal to the marginal benefit in the form of reduced travel costs. It is standard to assume that private travel costs are homogeneous of degree zero in capacity and number of users. This implies: qi

∂c i ∂c i + capi = 0, ∂q i ∂capi

i = 1, 2.

(12)

Using Eq. (12) the first order conditions with respect to capacity can be rewritten as: capi

∂K i ∂c i = ␭i q i , ∂capi ∂q i

i = 1, 2.

(13)

If there are constant returns to scale, ∂K i /∂capi is a constant, ki , and the left hand side of Eq. (10) is equal to the total cost of capacity of link i. The right hand side is equal to ␭i times the external (congestion) effect of traffic on link i. In the first-best equilibrium the Lagrange multiplier equals the number of users and the cost of capacity is therefore equal to optimal toll revenues. This self-financing result was first obtained by Mohring and Harwitz (1962) for a single link. Yang and Meng (2002) recently showed that it also holds in a general network. Homogeneity of degree 0 of the user cost function will not always be imposed: the function used for public transport in the two mode model does not satisfy this requirement.

3.2. Trip Generation and Trip Distribution A distinction between trip generation and trip distribution is often made in transportation planning models. It is convenient because it allows for a decomposition

38

JAN ROUWENDAL AND ERIK T. VERHOEF

of the total planning problem in parts that can be treated relatively independent of each other and it is therefore useful to introduce it in economic models as well. In order to introduce this distinction in a way that is consistent with economic theory, assume that consumer’s surplus is determined by a function P of the prices of the two transport services: CS(p 1 , p 2 ) = CS∗ (P)

(14)

P = P(p 1 , p 2 )

(15)

with

P can be interpreted as a composite price for transport services and Eq. (14) states that the consumer surplus derived from both transport services can be written as the consumer’s surplus of a single transport service. It can then be easily verified, using Eq. (2), that for the sum of the demand for both links, Q = q 1 + q 2 :   ∂CS∗ ∂P ∂P Q=− + (16) ∂P ∂p 1 ∂p 2 whereas the distribution shares s of the trips over the two modes are: si =

∂P/∂p i qi = , Q (∂P/∂p 1 ) + (∂P/∂p 2 )

i = 1, 2.

(17)

Considerable simplification of the latter two equations is possible if it is assumed that the partial derivatives of P add up to 1: ∂P ∂P + =1 ∂p 1 ∂p 2

(18)

According to Eq. (18) a small change dp in the prices of trips on the two roads leads to an identical change in the composite price of transport on both routes, which is intuitive. Equation (16) refers to trip generation, and Eq. (17), to modal split or route choice. These equations show that, when both assumptions are made, trip generation does not depend on the specification of P, whereas trip distribution does not depend on the specification of CS∗ . Specification (14) with P satisfying (18) is therefore consistent with the use of different sub-models for these two aspects of travel demand, which are relatively independent of each other. The effect of an increase in the price of one route or mode can be decomposed into a trip generation effect and a trip distribution effect as follows:   ∂s i ∂Q ∂Q ∂s i ∂q i ∂(Qs i ) = = si +Q +Q = si sj (19) ∂p j ∂p j ∂p j ∂p j ∂P ∂p j

Second-Best Pricing for Imperfect Substitutes in Urban Networks

39

This equation holds for i, j = 1, 2. The logit model, which is a standard specification for mode choice, results if the following P is chosen:6 1/␤  ␣1 e␤p 1 + ␣2 e␤p 2 P = ln (20) ␣1 + ␣2 with ␤ < 0 and ␣1 , ␣2 > 0. The shares can be determined using Eq. (17) as: si =

␣i e␤p i ␣1 e␤p 1 + ␣2 e␤p 2

(21)

␣i can be rewritten as exp(␣∗i ) with ␣∗i = ln(␣i ) and it is possible that ␣∗i depends on mode or route characteristics, as is usual in empirical logit models used to study trip distribution.7 In the limiting case in which ␤ = 0, link choice is independent of the prices. It can be verified from Eq. (20) that if p 1 = p 2 = p, P is also equal to p, that P is increasing in both prices, and that P always takes on a value in-between that of the two prices. This motivates the interpretation of P as a composite price for transport on the two modes or routes. P is homogeneous of degree 1 in the prices of all goods if p1 and p2 are interpreted as relative prices. The two links of the network are perfect substitutes if the users always choose the one with the lowest cost. In this situation the share of link 1 should be equal to zero if p 1 > p 2 , and equal to 1 in the opposite case, whereas the shares are indeterminate when both prices are equal. If the logit specification for the shares is used, this situation can be approximated arbitrarily closely by taking the limit for ␤ → −∞ of (21). This is easily verified by rewriting Eq. (21) for i = 1 as: s1 =

␣1 e␤(p 1 −p 2 ) ␣1 e␤(p 1 −p 2 ) + ␣2

(22)

and taking the limit for ␤ → −∞. When ␤ equals 0 there is no price sensitivity at all. The model developed in the present subsection seems attractive as the basis of a simulation model. In particular, it offers the possibility to compare how second-best problems change if the two links become closer substitutes.

3.3. A Simulation Model with Two Routes The previous two subsections provide the main ingredients for a simulation model for the two route and two mode situations that will be used in the remainder of

40

JAN ROUWENDAL AND ERIK T. VERHOEF

this section. The two-route case is discussed here and the model is completed by introducing specific forms for the (total) demand and travel time functions. A quadratic specification for consumer’s surplus is adopted as a function of the transport price index P: CS∗ = −aP − 0.5bP 2 .

(23)

with a > 0 and b < 0. Total demand for trips is therefore a linear function of the price of transport: Q = a + bP

(24)

The logit formulation (20) is used for P. User cost is given by the commonly used Bureau of Public Roads (BPR) formulation:8    q i c ie (25) c i = vot fftt 1 + c i1 capi where vot denotes the value of time and fftt free-flow travel time. This function is homogeneous of degree zero in the number of trips and capacity. For the cost of capacity in the two-route problem the following specification is used: K i (capi ) = k i capi

(26)

This cost function has constant returns to scale. The parameter values used for the two-route problem (see Table 1) refer to a situation in which two cities are connected by roads that differ in capacity and free-flow travel time. The route with the highest free-flow travel time has the lowest cost of capacity per road kilometre. Preferences with respect to the use of the two routes are symmetric, which means that consumers do not systematically prefer one road over the other. Link 1 has a free-flow travel time of half an hour and link 2 has a free-flow travel time of three-quarters of an hour, and the value of time is set at the Dutch average of 7.5. The BPR parameters c1 and ce have their conventional values. The unit price of capacity was determined as follows. A value of cap equal to 1,750 for the BPR cost function implies a doubling of travel times at a flow of around 2,800 vehicles per hour. This is roughly the flow at which, empirically, travel times double for a single highway lane. The hourly unit price of capacity of 6 for link 2, the fast connection, was determined by dividing the estimated average annual capital cost of one highway lane kilometre in The Netherlands (0.2 million) by 1,100 (220 working days multiplied by 5 peak hours per working

Second-Best Pricing for Imperfect Substitutes in Urban Networks

41

Table 1. Parameter Values for the Two-Route Problem. Parameter

Route/Mode 1

Route/Mode 2

Demand a b ␤ ␣

1

User cost vot fftt c1 ce

0.75 0.15 4

0.5 0.15 4

Cost of capacity k

6

6

7500 −100 −0.1 1 7.5

day, assuming two peaks) and then by 1,750 (the number of units of capacity of a standard highway lane), and finally by multiplying that result by 60 (the number of kilometres that can be travelled at free-flow speed in half an hour). The price of capacity at link 1 was set at the same value in order to facilitate comparability. Although the road is longer, which might call for a higher k, construction may be cheaper, which may call for a lower k.

3.4. Base Case The base case refers to a user equilibrium without tolls and with capacities as shown in Table 2, with a higher capacity for link 2. The main characteristics of the base case are given in Table 2. It is interesting to see what happens when the two routes become closer substitutes or in other words, when the absolute value of ␤ is increased. Figure 2 shows the equilibrium values of prices and demand for travel on both links for successively higher values of ␤. The prices (user costs) for the two links in Fig. 2 approach each other closely. Since capacity is fixed, this implies that the link that had the lower user cost in the base case is used more intensively. Its lower price attracts more users when price sensitivity increases, and for this reason user cost increases and the price difference disappears. Convergence to the case of perfect substitutes for ␤ → ∞ has been confirmed by means of a separate model that uses the Wardrop equilibrium conditions.9

42

JAN ROUWENDAL AND ERIK T. VERHOEF

Table 2. The Base Case of the Two-Route Problem. Variable cap q Q c (=p) P Elasticity of q w.r.t. p1 Elasticity of q w.r.t. p2 Elasticity of Q w.r.t. P CS K SS

Route/Mode 1

Route/Mode 2

1,250 2,413

1,750 3,566 5,925

17.35

13.45 15.21

−1.15 0.58 −0.25 178,752 7,500

0.67 −0.68

10,500 160,752

Note: cap, q and Q are measured as numbers of trips per hour, c, p, P, CS, K and SS in euros. Source: Own calculations.

3.5. First-Best Finding the first-best solution is facilitated by the assumptions of constant returns to scale of the capacity cost function and homogeneity of degree 0 of the travel cost function. Equation (9) implies:   q i c ie +1 (27) k i = vot ftt c i1 c ie capi and this determines the ratio between number of users and capacity. The optimal toll is a function of this ratio only:   q i c ie ␶i = vot fftt c i1 c ie (28) capi as is the user cost of travel given in Eq. (25). Together they determine the values of the prices pi by Eq. (5). The prices pi in turn allow total demand and route choice to be computed. Consumer surplus, toll revenues, capacities and associated costs are presented in Table 3. For the first-best case, it is also interesting to see what happens if the price sensitivity of route choice increases. The first thing to observe is that in this case the prices (user costs) on the two routes cannot become equal. As explained above, these prices are determined by the ratio between number of users and capacity, and this ratio is fixed by the first order condition for capacity choice. The constant price difference between the two routes will have a larger effect on demand when price sensitivity increases. Since capacity is adjusted to the number of users

Second-Best Pricing for Imperfect Substitutes in Urban Networks

43

Fig. 2. Effect of Increasing Substitutability on User Cost (Upper Panel) and Route Choice (Lower Panel) in the Base Case of the Two-Route Model.

because of the constant ratio between the two, this implies that the most expensive road will gradually disappear with enough substitutability. Figure 3 shows that demand shifts towards the first route when the absolute value of ␤ increases. For values larger than 2.5 demand for trips on route 1 is virtually equal to 0.

44

JAN ROUWENDAL AND ERIK T. VERHOEF

Table 3. First-Best Situation of the Two-Route Problem. Variable cap q Q c ␶ p P Elasticity of q w.r.t. p1 Elasticity of q w.r.t. p2 Elasticity of Q w.r.t. P CS TR K SS

Route/Mode 1

Route/Mode 2

2,511 2,817

2,942 3,579 6,396

6.96 5.35 12.31

4.98 4.93 9.91 11.04

−0.77 0.46 −0.17 204,542 15,063 15,063

0.47 −0.52

17,650 17,650 204,542

Note: cap, q and Q are measured as numbers of trips per hour, c, t, p, P, CS, K and SS in euros. Source: Own calculations.

Fig. 3. Effect of Increasing Substitutability on Route Choice in the First-Best Situation of the Two-Route Model.

Second-Best Pricing for Imperfect Substitutes in Urban Networks

45

3.6. Second-Best: The Two Route Problem In order to solve second-best problems an iterative procedure was used. As a starting point a user equilibrium for given values of the policy variables was taken. Then a linearised version of the first order conditions was solved for the second-best optimum. In the following step a weighted average of the original values and those suggested by the linearised first order conditions for the policy variables that can be freely chosen was used. This procedure was repeated until convergence occurred. A sequence of second best situations, where in each step an additional policy instrument is added, will be discussed. In the base case no instruments are used, a toll on road 2 is then introduced and the capacities of roads 2 and 1 are subsequenlty added. If, finally, a toll on road 1 comes available as a policy instrument, the first-best situation results. 3.6.1. One Toll, Fixed Capacities First, the situation in which the toll on route 1 is equal to zero, capacities on both routes are fixed and the policy maker’s single instrument is the toll on route 2 is analysed. This implies that of the first order conditions (7)–(9), only (7) and (8) for i = 2 are satisfied. In that situation only a modest welfare gain in comparison to the base case can be achieved: social surplus rises to 166,654. The index of relative welfare improvement, ␻,10 is equal to 0.13. The optimal toll is equal to 8.53, and severe congestion remains on both routes. When price sensitivity increases, user costs on both routes converge, whereas the numbers of trips diverge somewhat, just as in the base case. These results are surprisingly close to the results from the model with perfect substitutability (␶2 = 8.39 and v = 0.15, with SS = 161,579 in the base case, SS = 169,029 in the second-best case, and SS = 211,809 in the first-best case, where road 1 is in fact eliminated). This suggests that, although the assumption of imperfect substitutability between routes makes the model richer and allows for mode choice, it does not strongly affect the policy conclusions on the design and desirability of second-best pricing. 3.6.2. One Toll, One Capacity The second phase concentrated on the analysis of the situation in which the toll and capacity of route 2 can be determined by the policy maker. First-order conditions (7) are both satisfied, but conditions (8) and (9) are only valid for i = 2. In the second-best optimum, the capacity of route 2 is extended substantially to 4,048 trips per hour, whereas the toll is negative: −2.79. Congestion on route 1 remains significant, with the number of trips at 2,003. The welfare gain is substantial: social surplus is now equal to 194,510, with ␻ = 0.77. When price sensitivity increases,

46

JAN ROUWENDAL AND ERIK T. VERHOEF

the optimal toll increases. It becomes positive for values of ␤ higher than 0.25. At that level ␻ is still positive (and equal to 0.91), which reflects the welfare gain from optimising route 2’s capacity, even when the second best optimal toll is zero. When ␤ increases, the optimal capacity of route 2 increases, but congestion on route 2 also grows. In the limiting case of perfect substitutability, the second-best toll is equal to 1.71, the capacity, to 4,525.25 trips per hour, and ␻ increases to 0.92. These results suggest that when capacity choice is a policy variable, the impact of allowing imperfect substitutability upon policy design and evaluation increases. 3.6.3. One Toll, Two Capacities Finally, the case in which route 1 cannot be tolled, whereas the toll on route 2 and the capacities on both routes can be freely chosen was considered. In this situation all first-order conditions, except (8) for i = 1 are satisfied. The optimal values of these policy variables are presented in Table 4. In comparison to the first-best situation, the capacity of route 1 is increased, whereas the toll on route 2 is decreased. The price of transport is lower, which increases consumer surplus. The costs of capacity are no longer covered by toll revenues, and the deficit decreases social surplus. These two changes result in a social surplus that is almost as high as in the first-best situation (␻ = 0.97). Table 4. Second-Best: The Two Route Problem with One Toll and Two Capacities. Variable cap q Q c ␶ p P Elasticity of q w.r.t. p1 Elasticity of q w.r.t. p2 Elasticity of Q w.r.t. P CS TR K SS

Route/Mode 1

Route/Mode 2

2,806 3,244

2,955 3,596 6840

7.13 0 7.13

4.98 1.11 6.10 6.60

−0.42 0.29 −0.10 233,916 0 16,841

0.27 −0.34

4,016 17,732 203,359

Note: cap, q and Q are measured as numbers of trips per hour, c, t, p, P, CS, K and SS in euros. Source: Own calculations.

Second-Best Pricing for Imperfect Substitutes in Urban Networks

47

When looking at the sensitivity of these results to changes in the price sensitivity of mode choice it was found during the simulations that the number of trips on route 1 in the second-best optimum suddenly falls to (a value very close to) zero when the absolute value of ␤ increases from 0.97 to 0.98.11 Further investigation revealed that for the interval [−0.97, −0.37] there exist two equilibria: (a) one in which both routes have a substantial number of users and; another (b) in which road 1 is virtually unused. For values of ␤ smaller than 0.37 (in absolute value) the equilibrium in which route 1 is virtually unused disappears. Figure 4 shows the effects on user costs and quantities. The prices and quantities ending with “a” refer to the equilibrium in which both routes are used, and those ending with “b,” to the equilibrium in which only route 2 is used. The price for trips on route 1 in equilibrium “b” is only indicated for values of ␤ lower than 0.42, since it increases rapidly to values higher than 100. Use of route 1 is always close to 0. Only for values of ␤ smaller than 0.42 is the number of trips on road 1 in equilibrium “b” larger than 0.01. The equilibrium user costs on the two routes in equilibrium “a” do not converge and in fact the difference between them increases. As a result, the numbers of users on the two routes diverge. Eventually, it becomes inefficient to maintain route 1. With perfect substitutability, the ␻ for this policy is equal to 1. This implies that eliminating a road in the first-best optimum is not harmful to efficiency if its toll is restricted to be equal to zero. When the two equilibria exist simultaneously, the more balanced equilibrium “a” always has the highest social surplus. When ␤ increases, the difference between social surpluses decreases. The findings just reported show that there may be two solutions of the relevant subset of first-order conditions in the two-mode problem with one toll and two capacities as policy instruments. The two solutions correspond with different values of the policy instruments. When there are two equilibria, both are stable in two senses. First, the two solutions correspond to different values of the policy instruments, but for both configurations the user equilibrium in the network is stable. Second, both second best equilibria represent local welfare maxima, not minima.12 What has been found is that a policy maker who starts from an arbitrary user equilibrium and adjusts the values of the instruments so as to reach a second best optimum may end up in two different situations if the parameter ␤ lies in a particular interval. The initial values of the policy instruments determine in which equilibrium he ends up. Although no extensive mathematical or numerical analysis was undertaken, the authors experience indicates that when there exist two equilibria, both can be reached with substantially different initial values of the policy instruments.13

48

JAN ROUWENDAL AND ERIK T. VERHOEF

Fig. 4. Effect of Increasing Substitutability on User Cost (Upper Panel) and Route Choice (Lower Panel) in the Two-Route Problem with One Toll and two Capacities.

4. A SIMULATION MODEL FOR THE TWO-MODE PROBLEM 4.1. Set-Up of the Two-Mode Problem There are two main differences between the two-route and the two-mode problems. The first is that studies of the two route problem usually assume perfect

Second-Best Pricing for Imperfect Substitutes in Urban Networks

49

substitutability. This was an important motivation for considering the alternative case of imperfect substitutes in the previous section. The second difference involves the transport technology. In the two route problem, constant returns to scale in capacity cost and homogeneity of degree zero in capacity and demand are usually assumed, whereas public transport is often produced under increasing returns to scale and with a technology that is not homogeneous. In the present section a simulation model for the two-mode problem that incorporates a different transportation technology for the second mode is developed. It is otherwise comparable to the two-route problem of the previous section. For this two-mode model, demand, user cost of roads (mode 1) and cost of road capacity are all specified as in the two-route model. For rail (mode 2), two aspects of capacity are distinguished: frequency and number of seats per train. A peak period of one hour is considered, during which the arrival flow of commuters at the train platform is assumed to be constant. The average waiting time is therefore equal to 0.5/N, where N denotes the frequency of train departures. Passengers dislike waiting and the cost of waiting time is equal to (1 + w) vot, where vot denotes the value of time and w gives the extra weight given to the value of waiting time (Mohring, 1972). The discomfort associated with travelling by train is assumed to be: (a) proportional to travel time (tt); and (b) increasing in the ratio between the number of passengers and the number of seats. The total number of passengers is q2 and the total number of seats available is Ns, where s denotes the number of seats per train. The cost function for travelling by train is defined as: 

q c 2e 0.5(1 + w) + c 21 tt 2 c 2 = vot (29) N Ns This function is homogeneous of degree 0 in the number of trips and the number of seats. It is not homogeneous of degree 0 in the number of trips and trip frequency. The cost of providing rail transport is determined by the number of train departures and by the product of the number of departures and the number of seats per train: K 2 = k 21 N + k 22 Ns

(30)

The first order conditions for the optimum are analogous to conditions (7)–(9). The only difference is that there are now two first order conditions for the capacity of the second mode. The parameter values for the simulation model are listed in Table 5. The demand parameters are left unchanged. Travel time by rail is assumed to be equal to free-flow travel time by car. The additional cost of waiting is assumed to be 100% of the reference value of time, as is conventional. The exponent c2e has

50

JAN ROUWENDAL AND ERIK T. VERHOEF

Table 5. Parameter Values for the Two-Mode Problem. Parameter Demand a b ␤ ␣ User cost vot fftt/tt w c1 ce Cost of capacity k

Route/Mode 1

Route/Mode 2

7,500 −100 −0.1 1

1 7.5

0.5 0.15 4 6

0.5 2 1.0 4 k21 = 365, k22 = 4.5

a value of 4. The cost parameters are based on MuConsult (1999), where the total costs of passenger transport of Dutch Railways are approximated with a linear function of seat kilometres, train kilometres and passenger kilometres.14 The passenger kilometres component is relatively small and is ignored here. The marginal cost of train kilometres was found to be much higher than that of seat kilometres. The distance between A and B is assumed to be 60 kilometres. In the base case it is assumed that trains with a capacity of 300 seats depart every 10 minutes. In the Netherlands the fare for a return trip of 60 kilometres is around 15 euros. Given that for peak hours most of the trips are return trips, it seems appropriate to use half the value of a return ticket. Many of these trips, however, are made by season ticket holders, who pay a lower price per trip. For this reason 25% (instead of 50%) was used as the appropriate price for a one-way trip of 60 km. Results for this base case are presented in Table 6. The price of transport is higher than in the case of the two-route problem. The main reason for this difference is that the cost function for passenger transport is more sensitive to congestion than that for road transport. For the latter, the ratio between the number of users and capacity can be made equal to two. For passenger transport, at least in the Netherlands, it is virtually impossible to carry twice as many passengers as there are seats. This difference is reflected in the specifications of the cost functions for rail and road transport as discussed above. In the base case the ratio between use and capacity is 1.4 for trains. This figure seems reasonable for commuter trains in the Randstad area during peak hours. Revenues from train tickets are sufficient to

Second-Best Pricing for Imperfect Substitutes in Urban Networks

51

Table 6. The Base Case of the Two-Mode Problem. Variable cap q Q c ␶ p P Elasticity of q w.r.t. p1 Elasticity of q w.r.t. p2 Elasticity of Q w.r.t. P CS TR K SS

Route/Mode 1

Route/Mode 2

1,250 2,887

s: 300, N: 6 2,582 5,469

1,976 0 19.76

17.13 3.75 20.88 20.31

−1.12 0.85 −0.37 149,574 0 7,500

0.81 −1.28

9,683 9,390 142,367

Note: cap for the road, N, q and Q are measured as numbers of trips per hour, c, ␶, p, P, CS, K and SS in euros. Source: Own calculations.

cover all costs of rail passenger transport, as seems realistic for peak traffic in the Randstad area. In order to study the effect of changes in the substitutability between the modes, the same range of values of the coefficient ␤ that was used for road transport in the two-mode version simulation model were simulated. Only minor differences with the two-route situation were found. Again, the prices for the two modes, which are already close to each other in the base case, become virtually identical. The price elasticities of demand for the separate modes become very large (in absolute values) but the overall price of transport and its price elasticity remain almost identical. It must therefore be concluded that, at least in this base-case version of the two-mode model, the differences between the situations of imperfect and perfect substitutes are minor and comparable to the differences in the two-route case.

4.2. The First-Best Situation and the Two-Mode Problem Determining the first-best equilibrium is more complicated than in the two-route case. The ratio q2 /Ns is determined by the first order condition for s, but this does not determine N. This means that the cost of railway transport cannot be computed from the first order condition with respect to capacity alone, and an iterative procedure is therefore needed in order to compute the first-best equilibrium. The equilibrium

52

JAN ROUWENDAL AND ERIK T. VERHOEF

Table 7. First-Best Situation of the Two-Mode Model. Variable cap q Q ␶ p P Elasticity of q w.r.t. p1 Elasticity of q w.r.t. p2 Elasticity of Q w.r.t. P CS TR K SS

Route/Mode 1

Route/Mode 2

2385 2,901

s: 556, N: 8.77 3,743 6,644

4.98 4.93 9.91

2.16 5.21 7.37 8.56

−0.62 0.37 −0.13 220,712 14,307 14,307

0.35 −0.38

19,500 22,701 217,511

Note: cap for the road, N, q and Q are measured as numbers of trips per hour, c, ␶, p, P, CS, K and SS in euros. Source: Own calculations.

was first computed conditional upon a given value of N, the value of N suggested by the relevant first order condition in this equilibrium was then computed, and N adjusted towards this optimal value. This procedure was repeated until convergence occurred. The simulation results are presented in Table 7. In comparison with the base case, the number of rail passengers increases substantially. The quality of rail transport is also improved: the number of departures per hour increases from 6 to 8.77, and the number of seats per train almost doubles, and becomes equal to 556. The ratio between passengers and seats drops to 0.76. These improvements result in a much lower user cost of travel time for rail transport (d1.96 instead of d16.99). Even though the railway fare increases considerably to d5.21 per (one way) trip, the generalised price of train transport falls considerably because of the lower travel time, and is now lower than that of car travel. Road capacity is also expanded substantially and the price of road transport is halved in comparison with the base case. Toll revenues are exactly equal to the cost of road capacity. Fare revenues from rail transport are now insufficient to cover all costs. These revenues exactly cover the cost of seat kilometres (i.e. they are equal to k22 Ns in Eq. (30)). There are economies of scale associated with increasing the number of seats per train and marginal cost pricing is therefore unable to cover total exploitation costs. When train and road become closer substitutes, demand shifts towards the cheapest mode. With the assumed parameter values, the cheapest mode is the

Second-Best Pricing for Imperfect Substitutes in Urban Networks

53

train. With economies of scale in the railways, this demand shift causes a further drop in the cost of rail transport and road as a mode of transport eventually disappears. 4.2.1. Railway Fare, Fixed Capacities When the capacities of both modes are fixed and the only policy instrument is the train fare, possible welfare gains are small when compared to the base case (SS = d151,737, ␻ = 0.11). The optimal railway fare turns out to be considerably higher than in the first-best situation (d16.15). There is substantial congestion on the road (q 1 = 2,980). User costs are equal to d21.92 for the road and d25.03 for the railway. When substitutability between the two modes is increased, user costs converge to d22.68. The optimal toll decreases somewhat. 4.2.2. Railway Fare, Frequency and Seat Capacity When the railway fare, number of seats and frequency of departure can all be chosen by the policy maker, welfare gains are considerably higher than when only capacity can be used as a policy instrument. Social surplus is equal to d208,063 (␻ = 0.88), which is close to the first-best value. The optimal railway fare now turns out to be negative (−d2.09). User cost is equal to d8.45 for road and d1.87 for railway transport, implying that the price of the latter mode is also negative (−d0.22). When ␤ is increased the railway fare becomes positive, while user costs of the two modes converge gradually. 4.2.3. Railway Fare, Frequency and Seat Capacity, and Road Capacity The results for the two-mode problem, where capacity of the road can also be chosen optimally, are presented in Table 8. As can be seen in the table, the share of potential welfare gains is even larger than in the case where only the capacity of rail could be used as a policy instrument. The social surplus equals d215,889 (␻ = 0.98). Both road and railway capacities are considerably higher than in the first-best situation. The revenues from railway fares are less than one third of the costs of railway transport. When substitutability between the two modes increases, results similar to those from the two-route problem are obtained. Figure 5 illustrates this. For the interval [−0.92,−0.41] there are again two equilibria. In one of these the road remains virtually unused. For absolute values of ␤ higher than 0.92 there is only one equilibrium, in which there is no road. For absolute values of ␤ lower than 0.41 there is only one equilibrium in which both road and railway attract a substantial number of trips. For small values of ␤ the equilibrium in which two modes are used has the highest social surplus, for larger values the equilibrium in which only one mode is used has the highest social surplus.

54

JAN ROUWENDAL AND ERIK T. VERHOEF

Table 8. Second-Best: The Two Mode Problem. Variable cap q Q c ␶ p P Elasticity of q w.r.t. p1 Elasticity of q w.r.t. p2 Elasticity of Q w.r.t. P CS TR K SS

Route/Mode 1

Route/Mode 2

2,639 3,298

s: 556, N: 8.79 3,757 7,055

5.12 0 5.12

2.16 1.66 3.82 4.45

−0.31 0.21 −0.06 248,870 0 15,833

0.17 −0.21

6,247 22,783 216,501

Note: cap for the road, N, q and Q are measured as numbers of trips per hour, c, ␶, p, P, CS, K and SS in euros. Source: Own calculations.

4.3. Summary The main purpose of this paragraph was to provide a comparison of the results of two-mode situations with those of the two-route situations discussed in the previous section. The most important conclusion is that these results appear to be so similar. Introduction of imperfect substitutability into the two-road model makes it behave much like the analogous two-mode model, despite the substantial difference in the cost structure. The similarity between the two models is nicely illustrated by the development of the index of relative welfare improvement as a function of the price substitutability parameter ␤, which is shown in Fig. 6.15 In both cases only modest improvements in welfare are possible when one road or mode has to remain untolled and capacity has to be taken as given. If capacity of the road or mode that can be tolled can also be used as a policy instrument, a much larger share (typically more than three quarters) of the potential (first best) welfare gains can be realised. If the capacities of both roads or modes can be used as policy instruments, 100% of the potential welfare gains is realised for large values of ␤. The reason is that in these circumstances the unpriced road or mode disappear, which implies that its price becomes irrelevant. The simulation results clearly suggest that there is a qualitative difference between situations in which substitution between roads or modes is difficult (with the absolute value of ␤ less than 1) and situations in which it is easier,

Second-Best Pricing for Imperfect Substitutes in Urban Networks

55

Fig. 5. Effect of Increasing Substitutability on User Cost (Upper Panel) and Route Choice (Lower Panel) in the Two-Mode Problem.

but that there is not a qualitative difference between perfect substitutability and imperfect but relatively easy substitutability (a large absolute value of ␤). Two important reasons or this conclusion are that one road or mode disappears completely in the first best case for finite values of ␤ and that in the case with two capacities and one toll the transition from an equilibrium in which both roads/modes are used to one in which only one is used takes place for relatively small values of ␤.

56

JAN ROUWENDAL AND ERIK T. VERHOEF

Fig. 6. Effect of Increasing Substitutability on User Cost and Route/Mode Choice in the Two-Route (Upper Panel) and Two-Mode (Lower Panel) Models.

5. CONCLUSIONS This paper considers second-best pricing as it arises through incomplete coverage of full networks. The main principles were first reviewed by considering the classic two-route problem and some extensions that have been studied more recently. In

Second-Best Pricing for Imperfect Substitutes in Urban Networks

57

most of these studies the competing routes are assumed to be perfect substitutes, which is probably not the case for most parallel roads in reality, and even less likely for the case where competing connections represent different transport modes. A modelling framework in which the alternatives are imperfect substitutes was developed and numerical results for two roads and two modes were presented. In the model, trip generation and trip distribution are distinguished in a way that is consistent with economic theory. A linear demand equation was used for trip generation, and the logit model for trip distribution. The model offers the possibility to study the effect of changes in the substitutability between the two routes or modes. Perfect substitutability is a limiting case, in which the absolute value of one parameter becomes infinitely large. The model was used to consider situations in which one route or mode cannot be tolled. Simulation results show that, for the chosen parameter values, there is a substantial difference between the effectiveness of policies in which the capacities have to be taken as given, and those in which capacity of at least one mode can be changed. If only a toll on route or mode 2 can be used typically less than a quarter of the total possible welfare gains is realised. When the capacity of at least one route or mode can be determined by the policy maker, typically more than three quarters of the maximum possible welfare gains are realised. These figures are not very dependent on the substitutability between the two routes or modes. A striking feature of the policy in which the capacities of both modes and the railway fare can be used as policy instruments is the existence of two equilibria for a range of values of ␤. In one equilibrium there are substantial numbers of users of both modes, whereas in the other use of one mode is negligible. In this case there is a regime shift that is related to the possibility to substitute use of one mode for another, but it does not coincide with the difference between perfect and imperfect substitutability, interpreted as a finite and an infinite value of ␤ respectively. For the first-best case there is no regime shift that coincides with the difference between perfect and imperfect substitutability either. The situation in which only one mode is used is now approached in a continuous way, and it is reached for a finite and relatively small value of ␤. The results for the classic two-route problem with imperfect substitutes, in which only a toll can be charged on one of two parallel roads, are surprisingly close to the results for the model with perfect substitutes. This suggests that although the assumption of imperfect substitutability between routes makes the model richer and allows for mode choice, it does not strongly affect the policy conclusions on the design and desirability of second-best pricing. This makes the results of prior studies appear robust to the assumption of perfect substitutability.

58

JAN ROUWENDAL AND ERIK T. VERHOEF

This picture changes when capacity choice is also included as a policy instrument. When only the capacity of the priced road can be chosen, the choice may make the difference between a negative second-best toll (for imperfect substitutes) and a positive one (for closer or perfect substitutes), as illustrated by the results. When both capacities can be chosen, the choice may make the difference between keeping a road (for imperfect substitutes) and eliminating it in the long run (for closer or perfect substitutes).

NOTES 1. Note, however, that the reverse is also true: when capacity can be chosen, pricing alone will typically not result in an optimal outcome unless initial capacities “happen” to be optimal. 2. The imperfection of the tax instrument causes the term “marginal external costs” to become ambiguous, as a distinction between “full” and “direct” marginal external costs can be made. The full marginal external costs for drivers on the tolled road in this example are given by the tax rule. These include the second term, which is non-zero because of the imperfection of the tax instrument. Direct marginal external costs are thus defined here as all marginal external costs that are not caused by inefficient behaviour induced by the tax instrument itself. 3. Under second-best pricing, it is no longer optimal to equate taxes to the direct marginal external costs, but instead more complicated tax rules apply, which aim to partly correct for the relevant distortions, and na¨ıve Pigouvian taxation is therefore no longer optimal. In addition to that, the relative efficiency of a given type of second-best pricing often depends rather strongly on case-specific circumstances. 4. In transportation economics often an alternative (but equivalent) formulation of consumer surplus is used, based on the inverse demand function. In that case, partial derivatives of the Lagrange function with respect to the volumes of trips have to be used, and the Lagrange multipliers are then equal to zero under optimal tolling. In the present context, with possibly imperfect substitutability between the two routes/modes, the formulation based on the “ordinary” demand function is more convenient. 5. If the matrix of partial derivatives ∂qi /∂pj has an inverse this is the unique solution. The matrix has no inverse if total demand is fixed or if the shares of the two routes/modes are fixed. 6. Equation (20) is not the only possible specification of P that satisfies (18). Any function P = ln[g(exp(p1 ), exp(p2 ))] with g homogeneous of degree 1 in its two arguments does. 7. The function P is equal to the sum of a constant (−ln(␣1 + ␣2 )) and the logsum measure that can be used for welfare economic analyses (Small & Rosen, 1981). 8. See, e.g. Small (1992, p. 70). 9. This check has also been carried out in other uni-modal cases considered below. 10. The index of relative welfare improvement is the increase in social surplus compared to the base equilibrium as a fraction of the increase in the first-best optimum. Note that the value of this index is dependent on the parameter values chosen in the base case. For instance, the value of this index for the present case “one toll, fixed capacities” would

Second-Best Pricing for Imperfect Substitutes in Urban Networks

59

have been larger if the capacities in the base case had been closer to their first best optimal levels. 11. The number of trips on route 1 falls from 635 to 0, the number of trips on route 2 increases from 6,192 to 6,438. 12. This is confirmed, for instance, if social surplus is maximised by choosing capacity and toll for road 2 with various exogenously determined values of the capacity of road 1. For ␤ = 0.75 there exists a local interior maximum when the capacity of road 1 is chosen at the level of equilibrium (a) and another one at the boundary, when the capacity of road 1 equals zero. When ␤ is close to (but larger than) 0.37 the second maximum also becomes an interior maximum. 13. The case in which the two capacities are the only policy instruments (both tolls are equal to zero) has also been considered and no indications of multiple policy equilibria have been found. When ␤ gets large (in absolute value) route 1 gradually disappears. 14. These are only operating costs, not capital costs. Moreover, fixed costs and costs directly related to the number of passengers, which turned out to be relatively small, are ignored here. 15. When there are two equilibria, the one yielding the highest social surplus was used in Fig. 6.

REFERENCES Arnott, R., & Yan, A. (2000). The two-mode problem: Second-best pricing and capacity. Review of Urban and Regional Development Studies, 12, 170–199. Braid, R. M. (1996). Peak-load pricing of a transportation route with an unpriced substitute. Journal of Urban Economics, 40, 179–197. de Palma, A., & Lindsey, R. (2000). Private toll roads: Competition under various ownership regimes. Annals of Regional Science, 34, 13–35. Knight, F. (1924). Some fallacies in the interpretation of social costs. Quarterly Journal of Economics, 38, 582–606. L´evy-Lambert, H. (1968). Tarification des services a` qualit´e variable: Application aux p´eages de circulation. Econometrica, 36, 564–574. Lindsey, C. R., & Verhoef, E. T. (2001). Traffic congestion and congestion pricing. In: D. A. Hensher & K. J. Button (Eds), Handbook of Transport Systems and Traffic Control, Handbooks in Transport, 3 (pp. 77–105). Amsterdam: Elsevier/Pergamon. Liu, L. N., & McDonald, J. F. (1998). Efficient congestion tolls in the presence of unpriced congestion: A peak and off-peak simulation model. Journal of Urban Economics, 44, 352–366. Marchand, M. (1968). A note on optimal tolls in an imperfect environment. Econometrica, 36, 575–581. Mohring, H. (1972). Optimization and scale economies in urban bus transportation. American Economic Review, 62, 591–604. Mohring, H., & Harwitz, M. (1962). Highway benefits. Evanston: Northwestern University Press. MuConsult (1999). Voorbereiding prestatiecontract NS (Preparation Achievements Contract Dutch Railways). Report for the Dutch Ministry of Transport (in Dutch). Pigou, A. C. (1920). Wealth and welfare. London: Macmillan. Small, K. A. (1992). Urban transportation economics. Chur: Harwood. Small, K. A., & Rosen, H. S. (1981). Applied welfare economics with discrete choice models. Econometrica, 49, 105–130.

60

JAN ROUWENDAL AND ERIK T. VERHOEF

Verhoef, E. T., Nijkamp, P., & Rietveld, P. (1996). Second-best congestion pricing: The case of an untolled alternative. Journal of Urban Economics, 40, 279–302. Vickrey, W. S. (1963). Pricing in urban and suburban transport. American Economic Review, 53, 452–465. Vickrey, W. S. (1969). Congestion theory and transport investment. American Economic Review (Papers and Proceedings), 59, 251–260. Walters, A. A. (1961). The theory and measurement of private and social cost of highway congestion. Econometrica, 29, 676–697. Wardrop, J. (1952). Some theoretical aspects of road traffic research. Proceedings of the Institution of Civil Engineers Part II, 1, 325–378. Yang, H., & Meng, Q. (2002). A note on highway pricing and capacity choice under a build-operate transfer scheme. Transportation Research, A, 36, 659–663.

3.

THE IMPACT ON NETWORK PERFORMANCE OF DRIVERS’ RESPONSE TO ALTERNATIVE ROAD PRICING SCHEMES

Anthony May and David Milne 1. INTRODUCTION The majority of past proposals for road pricing have assumed that charges would be imposed to cross cordons (Dawson & Brown, 1985; Ministry of Transport, 1964, 1967; Richards et al., 1996). The initial and current schemes in Singapore (Holland & Watson, 1978; Menon, 2000) operate on this basis (but now with additional charge points elsewhere in the network), as do the toll rings in Norwegian cities (Larsen & Østmoe, 2001). The most significant exception to this is the area licensing scheme (under the title of “Congestion Charging”) implemented in London in 2003 (Transport for London, 2003), which, like the earlier proposals by the Greater London Council (May, 1975), involves charging within a defined area. Both such systems involve charging, or enforcement, at points in the road network. The mechanism for achieving this can simply rely on enforcement staff as initially in Singapore (Holland & Watson, 1978), employ physical toll booths as in Norway (Larsen & Østmoe, 2001), use video recognition as in London (Transport for London, 2003), or employ microwave technology and in-vehicle units as in the later applications in Norway and Singapore (Menon, 2000).

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 61–85 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09003-1

61

62

ANTHONY MAY AND DAVID MILNE

Despite this range of technologies, the concept of point-based, or cordon pricing is relatively simple, and easy for the driver to understand. It is also amenable to extension in several ways, by adding additional cordons, introducing radial charging screenlines to discourage orbital traffic, using isolated point charges, and varying charges by direction, location or time of day. Some of these design options are discussed in Chapter 4 of this volume. Even such complex structures have been criticised. It has been argued that they are inflexible, since the fixed charging points cannot readily be relocated if conditions change; that they are inequitable, in imposing the same charge on long and short journeys, and in having widely different impacts on activities just inside and just outside the cordon; and that they are disruptive, since they encourage rerouting around the boundary to avoid the charge. All of these limitations arise from the discontinuities which point-based charging introduces into the road network. In response to these criticisms, three other systems have been developed, all of which involve continuous charging within a defined area: distance-based, time-based and delay-based charging. Time-based and delay-based charging have been advocated on the basis that they are related more directly to the vehicle’s contribution to congestion (Oldridge, 1990). Time-based charging would be relatively simple to operate using microwave or Global Positioning Systems (GPS) to identify the boundary, and clock-based charging while the vehicle was moving. Delay-based charging would be more complex, in that it would require speed to be measured and a simple definition of congestion specified. Both have been criticised on the basis that the charges would be unpredictable, that they could be inequitable if delays were caused by third parties (for example through accidents or road works), and that they might encourage unsafe driving. They were rejected on this basis in the London Congestion Charging study (Richards et al., 1996). Subsequent research in a driving simulator, associated with the research reported here, confirmed that risk taking increased when time-based charges were imposed (Bonsall & Palmer, 1997). Distance-based charging would avoid these problems, and could be readily implemented using GPS technology. Its one potential drawback is that it could encourage shorter journeys through congested central areas. It has emerged as the front runner as an alternative to cordon pricing; the U.K. Commission for Integrated Transport (2002) has advocated it for national use; it is being employed for road freight charges in Germany; and it is being developed as a long term replacement for Singapore’s pioneering electronic road pricing system (May, 2004). In this chapter the impacts of these four road pricing schemes on the performance of the road network are explored. A brief review of earlier research on the issue (May & Milne, 2000) is presented in Section 2. Two subsequent projects designed to understand road users’ responses to these charging systems are described in

The Impact on Network Performance

63

Section 3. The model is described in Section 4. Aggregate results for three city road networks are presented in Section 5, and more detailed evidence for one of them in Section 6. Section 7 concludes.

2. PREVIOUS RESEARCH Earlier research by the authors on the relative merits of different road pricing systems used two congested assignment models, SATURN (Van Vliet, 1982) and CONTRAM (Taylor, 1990), to assess how responses to the charges, through reduced vehicle use and rerouting, would influence network performance, as measured by flows, travel times, speeds and generalised cost (Smith et al., 1994). Subsequently the SATURN application for Cambridge was used, together with the related elastic assignment routine SATEASY (Hall et al., 1992), to analyse the underlying causes of the aggregate impacts in more detail (May & Milne, 2000). The modelling approach is described below in Section 4. The initial analysis (Smith et al., 1994) used a single pricing regime for each of the four pricing systems. For the continuous charging systems, a uniform charge rate was applied throughout the road network inside a cordon located immediately within an outer ring road, per unit distance, time and time in excess of free-flow time respectively. For cordon pricing, a set of three cordons and six screenlines was implemented, with uniform charges in both directions at all crossing points, as shown in Fig. 1. It was not possible at the time to determine whether this was the best cordon structure; this problem is addressed in Chapter 4 of this volume. Three charge levels were used for each system, designed to achieve trip reductions of 5, 10 and 15% respectively. Values of time and distance in the base case were derived from earlier government advice (Department of Transport, 1989) as 9.93 pence per PCU-minute1 and 6.86 pence per PCU-km at 2002 prices. Values of charges were assumed to be equivalent to the value of other out of pocket costs. A single constant elasticity value with respect to generalised cost of −0.5 was used for all four systems, based on earlier findings (Goodwin, 1992). A single peak hour was modelled, and those no longer travelling as generalised cost increased were assumed to switch to other modes or times of day, or to cancel their trips altogether. Table 1 shows the charge levels required to achieve the three levels of trip reduction for the four systems. Three of the pricing systems exhibit similar relationships, with the charge for a 10% reduction roughly twice that for a 5% reduction, and the charge for 15% around four times as high. However, for delay-based charging, the charges are both much higher per unit, and need to rise much more rapidly to achieve higher levels of trip reduction. The pattern of

64

ANTHONY MAY AND DAVID MILNE

Fig. 1. Cambridge Charging Cordons.

these trip reductions is analysed in more detail in May and Milne (2000). All the continuous charging systems have their greatest impact on trips that start and finish within the charged area; delay-based charging has a greater impact on trips in the peak direction. Cordon charging, by contrast, has a greater impact on trips into and out of the charged area, and little impact on trips that start and finish within the charged area. The initial analysis focused on changes in network speed (PCU-km/PCU-hour) for the three charge levels and four charging systems. Table 2 summarises the results. It can be seen that the time-based systems appear to perform much better Table 1. Charge Levels Required to Produce Given Reductions in Number of Trips for the Cambridge Network (2002 Prices). Charging Method

Units

Cordon Distance Time Delay

Pence/crossing Pence/km Pence/min Pence/min delayed

Low Charge (5%)

Medium Charge (10%)

High Charge (15%)

27 13 7 78

59 26 14 260

117 48 25 651

Source: Updated from May and Milne (2000).

The Impact on Network Performance

65

Table 2. Percentage Change in Average Network Speeds Compared to Uncharged Base and Total Revenue from Charging (£ per Hour at 2002 Prices). System

Charge Level

Charge Area Speed (% Change)

Outer Orbital Speed (% Change)

Revenue (£ per Hour)

Cordon

Low Medium High

+8 +11 +19

0 −4 −8

10,958 19,330 32,147

Distance

Low Medium High

+15 +17 +15

−2 −4 −7

10,812 18,330 28,104

Time

Low Medium High

+20 +29 +31

−2 −3 −7

10,231 18,056 25,875

Delay

Low Medium High

+30 +35 +39

−1 −4 −6

11,441 19,544 29,566

Source: Updated from May and Milne (2000).

than the distance-based one, with the cordon-based system performing least well. It should also be noted that higher charge levels for distance-based charging produce no further improvement, and at the highest charge level, cordon-based charging out-performs distance-based charging. These results suggested that, despite their potential weaknesses, further development of time-based charging systems was warranted. One other interesting observation at this stage was that the revenues generated for a given level of trip reduction were very similar for all four charging systems, as also shown in Table 2. This suggested that revenue generation could be used as the common metric for comparing the four systems. Subsequent analysis focused on a more detailed comparison of the four systems for a wider range of charge levels. Trends in travel time, delayed time, travel distance, network speed and generalised cost were all analysed. As might be expected, time-based charging proved most effective in reducing travel time, and distance-based charging in reducing travel distance. Delay-based charging performed less well and it became clear that it achieved its impact on network speed by inducing substantial increases in distance for some trips to avoid the most congested locations. A loss of performance with distance-based charging was seen to result from a failure to reduce travel time further at the highest levels of distance reduction, reflecting a tendency to use shorter routes with lower capacity. Calculation of generalised costs for the whole network showed that time

66

ANTHONY MAY AND DAVID MILNE

and distance-based charging performed best, delay-based charging less well, and cordon-based charging least effectively.

3. SUBSEQUENT RESEARCH INTO USER RESPONSES One of the main limitations of the research reported above was the assumption that drivers’ responses to road pricing charges, and hence generalised cost elasticities, would be identical for the four charging systems. To address this, a subsequent research project used a number of experimental techniques to measure drivers’ responses (May et al., 1998). The first experiment used the Leeds Advanced Driving Simulator to test whether drivers were more likely to take risks when confronted with charges based on time, or time in congestion. The results, reported in detail elsewhere (Bonsall & Palmer, 1997), revealed that even at modest charge levels, equivalent to around 29 pence per journey (at 2002 prices), drivers were more likely to take risks and to admit having done so. The second experiment used a simple GPS-based facility to impose cordon and distance-based charges on drivers making actual journeys to work in Newcastle. It had been intended to test time-based and congestion-based charging as well, but this was abandoned following the results of the driving simulator experiments. A total of 30 volunteers took part in the experiment, and each was given a sum of money from which the charges incurred were deducted over the two weeks of the experiment. Charges were imposed either for crossing three virtual cordons or per unit distance on the driver’s most commonly used route. Charges were imposed on the journeys both to and from work. Drivers could avoid charges by taking an alternative route or by rescheduling by at least 30 minutes. The charge levels were modified during the experiment to determine the charge level that would just prompt the driver to select an alternative. Comparison of the generalised costs on the normal and alternative journey then allowed values of time and generalised cost elasticities to be determined. Overall, values of time of 13 pence per PCU-minute at 2002 prices were obtained. Values of time however, were substantially higher for work-bound trips than for home-bound, and for distance-based than for cordon-based charging. Higher values of time were typically associated with greater perceived risks of incurring higher costs. The third experiment used the VLADIMIR route choice simulator (Bonsall et al., 1994), which uses a laptop computer to display photographs of locations within a known network, superimposed with information on driving conditions, travel time and, in this case, charges imposed. A total of 83 volunteers took part,

The Impact on Network Performance

67

using a simulation of the road network in north-west Leeds. Once again, each participant was given a sum of money from which charges were deducted. They were asked to drive a route between a given origin and destination, under a known charging regime with differing levels of charge and in differing travel conditions. They could reduce the charge that they incurred by changing their route, but had no other response available. Comparison of the routes taken in terms of the travel time experienced and the charge incurred once again provided evidence of values of time and generalised cost elasticities. The average value of time across all experiments was 12 pence per PCU-minute at 2002 prices, a value very similar to that obtained in the second experiment. Once again, values of time differed by charging system. The fourth experiment was a conventional stated preference study, in which respondents chose between pairs of options in differing contexts. All four charging structures were presented, at differing charge levels and with differing outcomes in terms of reductions in congestion and performance of public transport. Different versions of the questionnaire were prepared so that the full range of conditions could be explored without making too many demands on individual respondents. A total of 562 responses were obtained from 1500 questionnaires distributed to drivers travelling to central Leeds and central Newcastle. The results from this series of experiments were subsequently used to provide a revised set of network-based analyses of the relative performance of the four charging structures, as described in the next section. However, a later study suggested a different representation of demand response (Hodgson, 2000). It involved a comparison of four different traffic restraint strategies: road pricing, parking charges, permit control and physical reduction of road space. The only road pricing structure tested on this occasion was cordon-based charging. Once again, network modelling was used as a basis for comparing the alternatives, and stated preference surveys were used to determine the parameters to describe the demand response. The survey questionnaires presented the four types of restraint method and indicated possible outcomes in terms of reductions in congestion and improvements in public transport for each of them. Like in the earlier study, different respondents received different versions of the survey. A total of 425 responses were obtained from 2500 questionnaires handed out to drivers parking in Cambridge, Norwich and York city centres. On this occasion the analysis distinguished between the value that drivers implicitly placed on travel time and the value that they placed on the charge; a higher than unit value of charge would imply greater sensitivity to expenditure on road pricing charges than to other out-of-pocket expenditure. The results indicated a value of time of 3.29 pence per PCU-minute at 2002 prices, much lower than that from the previous study, and a value of charge of 3.31 pence per PCU-minute, also at 2002 prices. This resulted in a net value of time with respect to road pricing charges of 11 pence per PCU-minute

68

ANTHONY MAY AND DAVID MILNE

at 2002 prices, which is broadly similar to the values from the earlier experiment (May et al., 2000). The relative timing of these studies meant that it was not possible to incorporate these results into the analysis described below. The implications are discussed further in the next section.

4. MODEL BASED ANALYSIS The findings reported in the following sections were obtained using SATURN (Van Vliet, 1982), a steady state equilibrium assignment model which predicts route choice and traffic flows on road networks based on the generalised costs of travel, taking account of detailed junction delay information. Vehicle flows are expressed in PCUs and public transport is preloaded on to the network along fixed, scheduled routes. In its conventional form, the model assumes fixed road travel demand. However, when representing demand management schemes, such as road user charging, variable demand may be represented through the SATEASY elastic assignment algorithm (Hall et al., 1992). This allows the representation of changes in demand that occur as a direct result of changes in the costs experienced on the road network, on the basis of an own-price elasticity function. A single composite elasticity is used to represent all possible demand responses including modal shift, retiming and trip suppression. The results generated need to be interpreted in the context of the model employed and its potential limitations. Steady state models such as SATURN assume average conditions throughout the time period represented and are, thus, unable to take account of variability in travel demand and traffic conditions. For this reason, it is necessary to focus on an individual peak situation when analysing congestion in urban networks and it is not possible to represent dynamic effects, either within the peak (e.g. as a result of changes in departure and arrival times) or between different occurrences (e.g. from day to day, due to variations in demand, incidents on the network, etc). The work reported here has investigated only the morning peak hour and the city model applications used have been constructed to represent long-run average conditions. Although SATURN includes facilities for representing variability in terms of drivers’ values of time, this was not used as no disaggregate data was available for the cities investigated and adding this dimension complicates the model calculations significantly. One further weakness of traditional assignment models is that all trips are treated as discrete, with a single origin and destination. Therefore, it is not possible to connect the separate elements of more complex, multi-purpose journeys made by individual drivers (e.g. a trip to deliver children to school, followed immediately by a trip to work). While in reality, it would be expected that a change in travel

The Impact on Network Performance

69

choice for a journey to one destination might also affect others in the trip chain, this is not recognised in the model. A particular situation where this may be significant involves the treatment of searching for a parking space in congested city centres. As such models are designed to assign traffic to the minimum cost routes between two points, a typical city model application will include a significant number of very short trips to represent circulating traffic involved in searching behaviour, in order to calibrate modelled traffic and congestion levels to those observed in reality. When a policy is introduced that induces changes in travel demand, such as road pricing, these short trips will be treated completely separately in the elastic assignment component of the model and may be more or less likely to be removed than the larger portion of the journey between the true origin and the “first destination” within the city centre, depending on the form of the demand function. Despite the limitations discussed above, assignment models such as SATURN have considerable potential to provide insights about the network effects of alternative road pricing schemes. Their key strength lies in the level of geographic detail by which both patterns of travel demand and topology of network supply are represented. This provides a more sophisticated treatment of spatial variation than other economic models and allows the investigation of tangible scheme designs at a level of detail appropriate for informing practical policy-making. Outputs were generated for three urban SATURN applications: Cambridge, Leeds and York. The Cambridge network covers the developed urban area plus a significant hinterland around the city and has extensions of major routes to other areas of the country, to allow the full costs of long distance trips to be included. The Leeds network covers the Leeds borough, including the developed urban area plus a significant hinterland to the north of the city. However, to the south and west, the modelled area is truncated at the boundaries of other West Yorkshire boroughs. The York network covers the developed urban area but only extends as far as the outer orbital routes formed by the A64 Leeds to Scarborough dual carriageway and the A1237 York Outer Ring Road. While both the Cambridge and York road networks are of broadly symmetrical ring radial structure, the Leeds network is significantly more extensive to the north of the city centre and dominated by the existence of two radial motorway approaches to the south. The differing topologies of the networks may have a significant impact on the outputs generated. Two sets of modelling parameters were produced from the various surveys discussed above:  an elastic exponential (or semi-log) demand relationship with respect to changes in cost, meaning that the impact of charges varied by origin-destination (O-D) movement with respect to the proportional change in generalised cost; and

70

ANTHONY MAY AND DAVID MILNE

 a “value of charge” factor for each charging system, which represents the behavioural weight of a given unit of charge compared to other elements of generalised cost. The elastic exponential function in SATURN takes the following form: T ij = T ij0 exp

−␤(c ij − c 0ij ) c 0ij

where Tij T 0ij cij c 0ij −␤

= road travel demand after charges are applied = road travel demand before charging = road travel costs after charging, including charges = road travel costs before charging = demand sensitivity coefficient

This function results in charges having greater impacts on shorter, lower cost journeys than on longer, higher cost ones. This is more consistent with an intuitive understanding of how road user charges might affect travel choices than the constant elasticity function assumed in Smith et al. (1994) and May and Milne (2000). The demand response function was calibrated to fit the behavioural parameters included within the SATURN applications, giving a −␤ parameter value of −0.2. Figure 2 illustrates the resulting relationship between change in cost and reductions in the number of trips. Given these parameters, a charge that increases generalised cost by 10% produces approximately a 2% decrease in the number of trips; a 50% increase in generalised cost produces approximately a 10% decrease in the number

Fig. 2. Relationship Between Change in Generalised Cost and Reduction in the Number of Trips.

The Impact on Network Performance

71

Table 3. Value of Charge Weights Relative to Other Aspects of Generalised Cost. Charging Method Cordon Distance Time Delay

Value of Charge 0.85 0.85 0.46 0.30

Source: Own surveys.

of trips; a doubling of generalised cost produces approximately an 18% decrease in the number of trips; a trebling of generalised cost reduces the number of trips by around a third; and a 5-fold increase in generalised cost (not unusual for short trips under a cordon-based system) reduces the number of trips by more than half. The “value of charge” weightings found are shown in Table 3. It should be noted that these values do not reflect the very different results from the later study in Cambridge, Norwich and York, as described above. Those results also suggested that a simple exponential function gave a better fit, indicating that the impact of charges varies with absolute, rather than proportional, changes in generalised cost. It will be important to test the implications of these different values. These figures suggest that drivers in the surveys valued charges at a lower rate than equivalent units of travel time and distance (based upon the values of time generated). This may run counter to expectations and it certainly conflicts with evidence from the later study outlined above, where cordon charges were found to be valued at three times other elements of generalised cost. It is interesting to see that, while in the case of fixed cordon and distance-based charges the value of charge is very similar and reasonably close to parity with other elements (e.g. travel time), values of charges fall when uncertainty is introduced into the charging system. This implies that drivers are less responsive to variable (and hence unpredictable) charges. The reason for this is risk aversion. According to the surveys, drivers are less likely to experiment with new routes under variable charging systems because of the combination of uncertainties about actual travel times and congestion and the additional costs that would be associated with them under time and delay-based charging. Applying a “value of charge” to the different charging systems affects their impacts on the number of trips and network performance. In order to assess how the behavioural parameters generated affected model outputs, charge levels were allowed to remain at those levels generated during previous studies and hourly revenue totals were used as the main method of comparison between charging systems. The analysis reported in the following sections focuses on two areas.

72

ANTHONY MAY AND DAVID MILNE

First, the aggregate results for reductions in the number of trips and changes in total travel time, total travel distance and generalised cost for all three networks were investigated. Second, the detailed impacts on route choice for two typical journeys were explored for Cambridge.

5. AGGREGATE RESULTS FOR ALL THREE CITIES Tables 4 and 5 show the charge levels required to produce given reductions in the number of trips for the Leeds and York networks, based on the earlier estimates of demand responsiveness used in May and Milne (2000) and in Table 1. The values for York and Leeds in Tables 4 and 5 are similar to those for Cambridge in Table 1 at low and medium charge levels. At high charge levels, the Leeds values are significantly lower. These differences may be the result of the way in which the Leeds network was defined. The boundary of the Leeds network used in the simulations places greater limitations on rerouting options to the south and west of the city than the actual network would suggest in reality. Lack of rerouting alternatives is also the primary explanation for the (typically at least 50%) lower charge levels in York. However, in this case, the lack of choice is a fair representation of reality. Table 6 shows percentage changes in the total number of trips, for all three cities, with the charge levels identified previously, but taking account of the new demand response and value of charge parameters that resulted from the surveys. The numbers in Table 6 suggest, not surprisingly, that the revised parameters indicate a lower responsiveness to charges than had been assumed during previous work. In particular, they suggest that time and delay-based charges might be rather ineffective at reducing the number of trips. General reductions in responsiveness will be due to the change in the elasticity, while the greater reductions for time and delay-based charges will occur as a result of the charge weights from Table 3. Table 4. Charge Levels Required to Produce Given Reductions in Number of Trips for the Leeds Network (2002 Prices). Charging Method

Units

Cordon Distance Time Delay

Pence/crossing Pence/km Pence/min Pence/min delayed

Source: SATURN calculations.

Low Charge (5%)

Medium Charge (10%)

High Charge (15%)

21 8 4 66

46 15 10 221

79 24 17 497

The Impact on Network Performance

73

Table 5. Charge Levels Required to Produce Given Reductions in Number of Trips for the York Network (2002 Prices). Charging Method

Units

Cordon Distance Time Delay

Pence/crossing Pence/km Pence/min Pence/min delayed

Low Charge (5%)

Medium Charge (10%)

High Charge (15%)

11 6 3 19

22 12 7 50

41 19 12 110

Source: SATURN calculations.

However the general tendency for time-variable charges to be more easily avoided by rerouting, as observed in May and Milne (2000), will also contribute to differences between the systems. Figures 3 and 4 show percentage changes in total distance travelled and total time spent travelling over the whole Cambridge network. Total distance travelled (in PCU-km) is probably the best measure of perceived traffic levels. Total time spent travelling is affected by both changes in traffic and changes in delayed time. The time and distance plots can be used together to determine changes in average speed on the network. It can be seen that all four charging systems induce small increases in total distance travelled over the network, despite the fact that total number of trips has decreased. Three systems, distance, time and delay-based, reduce total time spent travelling in the network by up to 15%, implying significant Table 6. Percentage Changes in Total Number of Trips. City

Charging Method

Low Charge (% Change)

Medium Charge (% Change)

High Charge (% Change)

Cambridge

Cordon Distance Time Delay

−2.5 −2.3 −1.0 −0.7

−5.4 −4.7 −2.5 −2.0

−9.7 −8.8 −4.5 −4.0

Leeds

Cordon Distance Time Delay

−1.6 −1.7 −0.7 −0.9

−4.2 −4.1 −2.1 −2.9

−7.2 −6.7 −3.7 −4.9

York

Cordon Distance Time Delay

−2.5 −2.5 −1.4 −1.3

−4.6 −5.0 −2.8 −2.5

−8.1 −7.5 −4.6 −4.3

Source: SATURN calculations.

74

ANTHONY MAY AND DAVID MILNE

Fig. 3. Percentage Change in Total Distance Travelled for the Cambridge Network.

increases in speeds. Cordon charging achieves a 5% reduction for a relatively low charge, but subsequently fails to produce a consistent decrease with rising charge levels. These trends are broadly the same in Leeds and York. Figure 5 shows variations in total generalised cost for the Cambridge network. It shows that distance-based and time-based charging have similar impacts, while delay-based and cordon-based charging are substantially less effective. At the lowest levels of charge, distance and time-based charging are respectively roughly 100 and 50% more effective than cordons in reducing generalised cost. However,

Fig. 4. Percentage Change in Total Time Spent Travelling for the Cambridge Network.

The Impact on Network Performance

75

Fig. 5. Reduction in Total Generalised Cost for the Cambridge Network.

these results should be interpreted with care. On the one hand the reductions are explained to a significant extent by reductions in the number of trips, without attempting to reflect the loss of utility from no longer travelling. On the other hand, a significant proportion of trips included in the calculation of generalised cost occur between zones outside the charging area and are therefore unaffected by the charging systems. In order to overcome these problems, average generalised costs were calculated solely for the affected trips,2 Figs 6–8 show trends in the total average generalised

Fig. 6. Percentage Change in Average Generalised Cost (Excluding Charges) for Affected Trips in the Cambridge Network.

76

ANTHONY MAY AND DAVID MILNE

Fig. 7. Percentage Change in Average Generalised Cost (Excluding Charges) for Affected Trips in the Leeds Network.

cost of travelling (excluding the charges paid) for the three cities. This represents the best measure of network performance. Similar impacts are observed in all three cities, although the plots display slightly different shapes in each case, probably because of the different network topologies. The most significant similarity is the rank order of the systems in generalised cost terms which, from best to worst, is delay, time, distance, and cordon. Although delay-based charging

Fig. 8. Percentage Change in Average Generalised Cost (Excluding Charges) for Affected Trips in the York Network.

The Impact on Network Performance

77

Table 7. Percentage Changes in Total Travel Time and Total Travel Distance for Standard Charges. Charging System

Distance Travelled (% Change)

Time Spent Travelling (% Change)

Centre

Urban

Orbital

Total

Centre

Urban

Orbital

Total

Cambridge Cordon Distance Time Delay

−26.9 −34.2 −21.9 −16.2

−26.3 −32.5 −18.3 −4.2

+24.8 +28.7 +17.1 +8.8

+1.4 +0.6 +0.6 +1.7

−36.4 −44.4 −35.4 −31.7

−26.7 −34.6 −27.7 −18.9

+44.8 +44.5 +21.6 +7.9

−6.8 −12.5 −14.1 −13.3

Leeds Cordon Distance Time Delay

−23.9 −20.9 −10.6 −7.1

−8.9 −28.9 −17.4 −6.8

+6.6 +3.3 +9.2 +6.0

−3.2 −4.8 −1.9 0

−34.0 −25.5 −19.7 −31.0

−8.5 −31.7 −24.1 −18.0

+18.5 +29.0 +15.7 +11.3

−3.2 −3.1 −5.3 −8.4

York Cordon Distance Time Delay

−6.8 −14.1 −12.4 −9.0

−10.4 −18.2 −9.8 −3.4

+4.2 +11.2 +8.8 +6.5

−2.2 −1.8 +0.3 +1.8

−13.3 −25.9 −32.9 −37.7

−5.4 −20.5 −17.5 −12.5

+4.4 +19.8 +14.9 +12.4

−3.8 −9.3 −10.4 −9.2

Source: SATURN calculations.

is the only system that universally reduces generalised cost, the scale of the impact at network-wide level is small. In contrast, cordon charges universally increase generalised cost. Time-based charging appears almost as good as delay-based charging in Cambridge and York, but produces no significant benefit in Leeds, where the network impacts of charging in general appear least positive. Distancebased charging performs well at low charge levels in Cambridge, but fails to reduce generalised cost in Leeds and York. All systems show a tendency to lose performance at higher charge levels. Table 7 provides results for changes in total distance travelled and total time spent travelling, for all three networks, broken down by geographical area. Figures are presented for a standard charge level, defined as the charge that yields similar volumes of revenue for all four systems. Referring back to the charge levels presented in Tables 1, 4 and 5, this involves specifying the high charge level for three systems (cordon, distance and time-based) and the medium charge level for delay-based charging. Outputs were also disaggregated to distinguish the city centre (centre), the remainder of the urban charging area (urban) and a major orbital route immediately outside the charging area (orbital). These figures demonstrate that impacts are broadly as expected, with significant reductions in traffic volumes and time spent

78

ANTHONY MAY AND DAVID MILNE

travelling within the charging area, offset to varying degrees by increases in traffic volumes and time spent travelling on the orbitals outside. Although there is significant variation in the distribution of impacts between cities and charging systems, in general the greatest reductions in traffic volumes and time spent travelling tend to occur in city centres. The scale of impacts overall is significantly greater for cordon, distance and time-based charges in Cambridge than in the other two cities. Delay-based charges on the other hand have a broadly similar effect in all cases. In terms of redistribution of PCU-km between areas of the network, distancebased charges tend to yield the highest reductions, followed by cordon and time-based charging, while the effects of delay-based charges are significantly smaller. In terms of time, there is no consistent pattern of impacts by charge level across the cities. Cordon charges produce the highest reductions in time spent travelling in the city centres of Cambridge and Leeds, but delay-based charges produce the highest reductions in York. Cordon charging also produces the highest increases in time on the orbitals in Cambridge, but has the smallest effect in York. The main message coming out of this information is that impacts of road user charging are likely to be sensitive both to location and system design.

6. DETAILED RESULTS FOR CAMBRIDGE In order to gain some insight into the effects of different charging systems on individual drivers, two typical origin to destination movements were investigated. One movement comprises a north-south trip from a residential location outside the charging area to a destination to the south of the city centre (i.e. an external to internal trip). The other movement comprises an east-west trip across the city centre, with both origin and destination inside the charging area (i.e. a wholly internal trip). The impacts of charging for the standard charge level used in Table 7 were assessed. Outputs comprise routes chosen under different charging systems, including the no charge base case, and the effects on distances, times, delays and costs that would be experienced by the drivers involved. Figure 9 shows the routes chosen for the external to internal trip. In the no charge situation, drivers choose two main routes that are quite direct and travel primarily through the charging area. However, under cordon and distance-based charging, drivers choose to avoid the charging area to the greatest extent possible, preferring to travel around the orbital to a point that minimises the charge paid. In complete contrast, delay-based charges cause most drivers to concentrate on one of the city centre routes used in the no charge base case. There is no cost incentive to avoid the charged area provided that travel within is undelayed. The effects

The Impact on Network Performance

79

Fig. 9. Route Choice for a Typical External to Internal Movement.

of time-based charges fall between those of distance and delay-based, with some drivers taking a direct route through the charging area and others diverting around the orbital to access the quickest radial route to the centre (which, interestingly, is not the same radial as that chosen for shortest distance!).

80

ANTHONY MAY AND DAVID MILNE

Table 8. Journey Parameters for a Typical External to Internal Trip (2002 Prices). Scenario

Uncharged Cordon Distance Time Delay

Average Trip Length (km)

Average Travel Time (Minutes)

Average Time Delayed (Minutes)

Average Charge Paid (£)

Average Trip Cost (£)

8.2 18.8 17.7 (2.4) 11.3 8.8

19.8 28.1 26.6 15.6 (11.0) 14.4

8.4 5.4 4.5 3.7 3.4 (3.2)

0 1.06 0.97 2.31 7.06

2.14 3.46 3.28 4.28 8.78

Note: The figures in brackets represent the elements of travel distance, travel time and delay that occur within the charging area and that are thus charged. Source: SATURN calculations.

Table 8 summarises the outputs for the external to internal trip. The figures in brackets, where presented, represent the elements of travel distance, travel time and delay which occur within the charging area and which are thus charged. The table shows that all charging systems increase average trip length, although the impact is marginal in the case of delay-based charging. Both cordon and distance-based charging also produce significant increases in average travel time, due to the much longer routes chosen, while time and delay-based charging reduce average travel time, through reductions in congestion within the charging area. All systems reduce delays. Average charges paid under cordon and distance-based systems are less than half those paid under time-based charging and less than one sixth of those under delay-based charging, suggesting that a significant trade-off is taking place between time and money. In terms of total generalised cost, cordon and distance-based charging cause approximately a 50% increase, time-based charging implies a 100% increase and delay-based charging implies a 400% increase. Although delay-based charging reduces delays to a greater extent than the other systems, the unit of charge (£2 per PCU-minute at 2002 prices) means that the amounts paid are much higher, as a result of an apparent inability to push delay within the charging area below a certain threshold level (in this case, around 20% of total travel time). Figure 10 shows the routes chosen by the wholly internal movement. In this case, the differences between the charging systems and the no charge base case are rather less noticeable, except in the case of distance-based charging, which causes some drivers to pursue a very significant diversion. The other important point to note is that both time and delay-based charging cause increased driving

The Impact on Network Performance

81

Fig. 10. Route Choice for a Typical Wholly Internal Movement.

through the historic city centre, presumably to avoid congestion on the inner orbital that runs around it. Table 9 presents the numerical results. In this case, average trip length remains more or less constant, except for distance-based charging, which results in a

82

ANTHONY MAY AND DAVID MILNE

Table 9. Journey Parameters for a Typical Wholly Internal Trip (2002 Prices). Scenario

Uncharged Cordon Distance Time Delay

Average Trip Length (km)

Average Travel Time (Minutes)

Average Time Delayed (Minutes)

Average Charge Paid (£)

Average Trip Cost (£)

5.2 5.1 11.9 (3.2) 5.0 4.9

11.2 9.3 17.6 9.0 (7.6) 9.7

4.0 2.3 5.5 2.1 2.5 (2.3)

0 1.99 1.30 1.59 5.08

1.25 3.07 3.48 2.64 6.18

Note: The figures in brackets represent the elements of travel distance, travel time and delay which occur within the charging area and which are thus charged. Source: SATURN calculations.

6.7 kilometre detour to avoid paying for an additional 2 kilometres within the charging area! All systems apart from distance-based also experience reductions in travel time and delay. As a result of the major detour taken by some drivers, distance-based charging achieves the lowest average charge. However, when charges are combined with other elements to form average trip cost, time-based charging produces approximately a 100% increase in total average trips costs, followed by cordon charging, which produces a 150% increase, distance-based charging, which produces a 200% increase, and delay-based charging, which produces a 400% increase.

7. CONCLUSIONS The studies reported here show that different charging systems have very different impacts. Cordon-based charging imposes relatively high charges at discrete points in the network. So, less obviously, does delay-based charging, since charges are only incurred where there is sufficient delay to trigger them. As an indirect result, charge levels for delay-based charging have to be significantly higher to achieve a given level of trip reduction, since it is relatively easy to reroute to avoid them. The differences in value of charge for the different systems suggest that drivers will be less willing to reroute, or reduce their travel, in response to time-based and delay-based charging, because the charges are variable and hence uncertain. This is a reflection of the effect of risk aversion on value of time. Conversely, cordon-based and distance-based charging are likely to induce greater rerouting. This was particularly evident in the detailed analysis of selected journeys in the Cambridge network, where rerouting outside the charging area substantially increased overall journey distance. Earlier research in a driving simulator had also

The Impact on Network Performance

83

indicated that time-based and distance-based charging could encourage drivers to take risks to reduce their travel time, and thus drive less safely. This was the principal reason for not testing them in an on-road experiment. At the standard charge levels tested, distance-based charging was the most effective system in reducing distance travelled within the charged area, whereas delay-based charging was the least effective. However, in all cases increases in travel on the orbital outside the charged area largely offset the reduction in distance travelled within it. As a result, the net effect on average distance travelled was almost the same for all systems. This highlights the importance of selecting orbital routes that will be able to handle traffic without undue environmental impacts or increased congestion. Alternatively, charges could be extended over much wider areas. It should be noted that since a reduction of vehicle kilometres inside the charging area will be typically offset by an increase outside it, the aggregate impacts on the environment, which are largely related to vehicle kilometres travelled, and particularly CO2 emissions, will be small. At the same standard charge levels, distance-based charging had the greatest impact on urban travel times and, except in Cambridge, cordon charging the least. Once again, much of this benefit was offset by increases in travel time on the orbital. However, all systems achieved a net reduction in overall travel time, with that for cordon charging being typically lower than the reductions achieved by the continuous charging systems. The impacts on generalised cost are harder to interpret. When the impact on trip reduction is included in full it appears that distance-based and time-based charging perform similarly, while delay-based charging is less effective, and cordon-based charging is both least effective and least able to generate higher benefits as charges rise. When generalised costs are simply averaged over those trips that are directly affected and remain on the network, time-based and delay-based charging appear to perform better, with cordon-based charging showing no benefits in any city, and distance-based charging only performing well in Cambridge. As noted earlier, the real impact on generalised cost is likely to lie between these two assessments. It is difficult to make a final recommendation on which system to choose when all the results above are taken into consideration. It is clear that the cordon system, as defined for these tests, was less effective in reducing trips, travel distance, travel time and generalised cost than the three continuous charging systems. This may however be the consequence of the particular design that was adopted in this study, with three cordons and six screenlines. This problem is further discussed in Chapter 4 of this volume. However, it is likely that the discontinuities in charging imposed by cordons, and their induced rerouting, will generally make them less effective.

84

ANTHONY MAY AND DAVID MILNE

Amongst the continuous charging systems, distance-based charging is the most effective in reducing travel time and distance travelled within the charging area, and total generalised cost. However, it has the most extreme impacts on rerouting, and it therefore requires particular care in the selection of the boundary route that defines the charging area. Time-based and delay-based charging have a more subdued impact on rerouting and trip making, but appear likely to encourage dangerous driving. Although they are probably best rejected on this basis alone, if they were to be selected it appears that time-based charging would be more effective than delay-based charging.

NOTES 1. PCU stands for passenger car unit and denotes the ratings given to vehicles to indicate their relative congestive effect on a link. A car has a PCU rating of 1, whereas a lorry for example, has a PCU rating of 2.5 or 3 depending on its size. 2. Computed by dividing total generalised cost in the affected cells by the number of trips in the same cells.

ACKNOWLEDGMENTS The research reported in this chapter was funded by the U.K. Engineering and Physical Sciences Research Council, whose support is gratefully acknowledged. Thanks are also due to several colleagues in the Universities of Leeds, Newcastle and York who contributed to the different stages of the research programme. The views expressed in this paper, however, are the authors’ own.

REFERENCES Bonsall, P. W., Clarke, R., Firmin, P. E., & Palmer, I. A. (1994). VLADIMIR and TRAVSIM: Powerful tools for route choice research. Proceedings of the 22nd European Transport Conference, Seminar H. PTRC (Planning and Transport Research and Computation), London, 65–67. Bonsall, P. W., & Palmer, I. A. (1997). Do time-based road-user charges induce risk-taking? – Results from a driving simulator. Traffic Engineering and Control, 38, 200–203. Commission for Integrated Transport (CfIT) (2002). Paying for road use. London: CfIT. www.cfit. gov.uk/reports/pfru/pdf/pfru.pdf. Dawson, J. A. L., & Brown, F. N. (1985). Electronic road pricing in Hong Kong 1: A fair way to go? Traffic Engineering and Control, 26, 608–615. Department of Transport (1989). Highways economics note no. 2. In: COBA 9 Manual. London: HMSO.

The Impact on Network Performance

85

Goodwin, P. B. (1992). A review of new demand elasticities with reference to short and long run effects to price changes. Journal of Transport Economics and Policy, 26, 155–169. Hall, M. D., Fashole-Luke, T., Van Vliet, D., & Watling, D. P. (1992). Demand responsive assignment in SATURN. Proceedings of 20th Summer Annual Meeting, Seminar E, PTRC (Planning and Transport Research and Computation), London, 25–39. Hodgson, F. (2000). Demand management in historic cities: Measurement of travel choices. Working Paper 550. Institute for Transport Studies, University of Leeds. Holland, E. P., & Watson, P. L. (1978). Traffic restraint in Singapore. Traffic Engineering and Control, 19, 14–22. Larsen, O. I., & Østmoe, K. (2001). The experience of urban toll cordons in Norway: Lessons for the future. Journal of Transport Economics and Policy, 35, 457–471. May, A. D. (1975). Supplementary licensing: An evaluation. Traffic Engineering and Control, 16, 162–167. May, A. D. (2004). Singapore: The development of a world class transport system. Transport Reviews, 24, 79–101. May, A. D., Bonsall, P. W., & Hills, P. J. (1998). Evaluation of driver response to road user charging systems. Proceedings of 9th Road Transport Information and Control Conference. London: Institution for Electrical Engineers. May, A. D., Hodgson, F. C., Jopson, A. F., Milne, D. S., & Tight, M. R. (2000). Comparison of four travel demand management measures. Traffic Engineering and Control, 41, 396–401. May, A. D., & Milne, D. S. (2000). Effects of alternative road pricing systems on network performance. Transportation Research, A, 34, 407–436. Menon, A. (2000). ERP in Singapore – A perspective one year on. Traffic Engineering and Control, 41, 40–45. Ministry of Transport (1964). Road pricing: The economic and technical possibilities. London: HMSO. Ministry of Transport (1967). Better use of town roads. London: HMSO. Richards, M., Gilliam, C., & Larkinson, J. (1996). The London congestion charging research programme: 1. The programme in overview. Traffic Engineering and Control, 37, 660–671. Smith, M. J., May, A. D., Wisten, M. B., Milne, D. S., Van Vliet, D., & Ghali, M. O. (1994). A comparison of the network effects of four road-user charging systems. Traffic Engineering and Control, 35, 311–315. Taylor, N. B. (1990). CONTRAM 5: An enhanced traffic assignment model. Research Report, 241, Crowthorne: Transport and Road Research Laboratory. Transport for London (2003). Congestion charging. London: TfL. www.cclondon.com/. Van Vliet, D. (1982). SATURN – A modern assignment model. Traffic Engineering and Control, 23, 578–581.

This Page Intentionally Left Blank

4.

OPTIMAL LOCATIONS AND CHARGES FOR CORDON SCHEMES

Anthony May, Simon Shepherd and Agachai Sumalee 1. INTRODUCTION In Chapter 3 of this volume distance-based and time-based charging were found to perform better than cordon-based charging in reducing generalised cost in the networks studied. Typically, distance-based charging achieved reductions that were around 50% greater than those from cordon-based charging. However, these results were specific to the cordon structure specified which, for most of the tests reported, involved three cordons and six screenlines. Since there is evidence that shows that the impact of cordon-based charging on generalised cost is very sensitive to the detailed design of the cordon (May et al., 1996; MVA Consultancy, 1999), there seems to be a good case for developing methods which enable cordons to be better designed. In this chapter a series of methods developed to determine optimal locations for imposing charges in urban networks, and optimal charge levels at those locations are presented. The term “cordon schemes” is used to define any system in which charges are levied as vehicles pass points in the road network. These points are combined to form a cordon around a city or its centre, but several variants are available. These include the use of one or more further cordons; radial screenlines to control orbital movements; spurs which control orbital movements adjacent to the cordon; and point charges at isolated locations in the network. A further element of cordon design is the level of charge at the charge points, which can vary by location of the cordon, direction of

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 87–105 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09004-3

87

88

ANTHONY MAY ET AL.

crossing, time of day or even, in principle, between crossing points on the same cordon. In conventional studies the options tested are selected largely on the basis of professional judgment. A review of the results of a survey of the judgmental approaches adopted in a series of case studies is presented in Section 2. These surveys were conducted in order to understand the judgmental process for the design of cordons, and to see to what extent formal mathematical methods could generate more successful designs. Following that a series of mathematical methods is described in Sections 3 and 4. Those in 4 use the principles of genetic algorithms to determine the optimal charge locations and charge levels. The results of applications of these methods to two road networks of differing complexity are then presented in Sections 5 and 6. Finally, Section 7 draws conclusions on the methods and their policy implications.

2. JUDGMENTAL DESIGN PRINCIPLES While many studies on road pricing and all the schemes implemented so far have selected locations for road pricing cordons, few have specified the basis on which these selections were made. May et al. (2002) and Shepherd et al. (2001a) used those studies and real world examples to understand the cordon design criteria used and found that in general the idea is to design cordons that will avoid adverse impacts, gain public acceptance, and be practical. Sumalee (2001) conducted a questionnaire and in-depth interview survey with practitioners in six U.K. local authorities who were active participants in the U.K. Charging Development Partnership. Both questionnaire and interview were structured to cover the context of the proposal, the objectives of the scheme, and the detailed design process. The conclusion from that study was that avoiding adverse impacts and gaining public acceptance were perceived as more important by the local authorities than other practical considerations. Table 1 lists the design criteria identified in the interviews. Sumalee (2001) found that the approach most frequently used was to:  focus on the city centre, together with any major traffic generators on its fringes;  place the cordon within the city centre ring road if one existed, or alternatively try to avoid charging routes which would allow drivers to avoid the centre;  minimise the crossing points by using existing barriers to movement, and keep the cordon as simple as possible;  ensure that on-street parking control extends beyond the cordon;  use a simple charge structure with uniform charges for all crossing points; and  keep the charge at a level sufficiently low to be acceptable.

Optimal Locations and Charges for Cordon Schemes

89

Table 1. Design Criteria Identified in the Interviews. Avoid adverse impacts The design should ensure the provision of sufficient alternative routes for drivers who want to bypass the charging area. The design should avoid the dispersion of environmental or congestion problems to other areas. The design should leave the facilities for interchange outside the cordon (e.g. park and ride or parking facility). The design should ensure that all entry points to the charging area are charged or closed. The cordon boundaries should be associated with the controlled parking zone which can avoid the dispersion of parking demand around the fringe of the cordon. Gain public acceptance The cordon structure should be simple and easy to understand. The charge structure should also be simple and easy to understand. The charge should be at a level that is acceptable to the public. The charge should be perceived as fair by the public. Be practical The number of charging points should be minimized to reduce capital costs. The cordon should be located inside the city authority area. Source: Sumalee (2001).

These principles formed the basis for the judgmental designs, which were used as a base against which to compare mathematically generated designs.

3. INITIAL METHODS Methods for specifying optimal charges (or first best charges) where all links in a transport network can be included are readily available (Sheffi, 1985). However, those that cater for charging at a limited number of points, such as across cordons, are less well developed. The problem of optimal cordon design involves choosing the charge locations and levels that maximise social welfare, defined as total benefits minus total costs. If the Marshallian demand is used to measure benefits, the social welfare function can then be defined as  Ni  D i (x) dx − ␦jp F p c j (1) W(F, ␶) = i

0

j

p

where F is a vector of path flows, ␶ is a vector of link charges, Di is an inverse demand function, Ni is the trip demand for origin-destination (O-D) pair i, Fp is path flow on path p, cj is the travel cost on link j (excluding the charge), and ␦jP is the dummy variable which is equal to 1 if link j is related to path p and 0 otherwise.

90

ANTHONY MAY ET AL.

Once the charges are implemented, road users will respond to them by changing their routes or deciding not to travel. This response is captured by assuming that road users behave according to Wardrop’s user equilibrium rule (Wardrop, 1952). This user equilibrium condition is thus included in the optimisation program as a constraint. This constraint imposes a special structure and complexity on the problem. Generally, this type of optimisation program can be formulated as a Mathematical Program with Equilibrium Constraints (MPEC). The mathematical approach adopted in this chapter builds directly on work by Verhoef (2002). The approach is, logically, divided into two stages, the definition of optimal charge levels for a given set of locations and the prediction of optimal charge locations. The CORDON process associated with SATURN was developed in order to solve the problem of optimal charge levels. SATURN is a steadystate equilibrium assignment model which predicts route choice and traffic flows on a road network based on the generalised costs of travel, taking account of delays due to capacity constraints (Van Vliet, 1982). It includes an assignment sub-model, which estimates driver route choices using Wardrop user optimum equilibrium assumptions. SATURN can assume fixed demand or variable demand. SATEASY is an elastic assignment algorithm (Hall et al., 2002) that allows the representation of changes in demand which occur as a direct result of changes in the costs experienced on the road network. Shepherd et al. (2001a) develop the CORDON process. Given a set of predefined tolled links, SATURN is first run for the do-nothing case (no charges). This produces a set of used paths, path travel costs and O-D demand and costs as input for the CORDON process. CORDON then solves the first order equation system to obtain the predicted charges on specified links. Basically, the first order equation system is obtained by setting the first derivatives of the Lagrangian, which is the welfare function with constraints for user equilibrium, equal to zero. The predicted charges are then implemented with SATURN to produce the new input for CORDON. The process is repeated until convergence, where the predicted charges in iteration i equal those in iteration i + 1. Shepherd et al. (2001b) extend CORDON to develop LOCATE. The LOCATE process involves building up a list of charge points incrementally by choosing links one by one on the basis of a location index. Although previously selected charge points are always included, the charge levels are allowed to vary each time an additional link is added. The location indices are the approximation of the welfare gains that would result from placing optimal charges in particular locations. They use the predicted charge from the first iteration of the CORDON process combined with the shadow prices associated with the link(s) considered. Shepherd and Sumalee (2004) conduct tests on the performance of both CORDON and LOCATE with small and medium size traffic networks. While

Optimal Locations and Charges for Cordon Schemes

91

the CORDON method works well with these simpler networks, it is not as successful with large scale networks. The method requires perfect convergence of the assignment problem as any convergence errors at the path level are magnified by the Lagrange multipliers and hence upset the prediction of optimal charges. The LOCATE process fails to identify the optimal charge locations in cases where an optimal charge location in an early iteration is no longer the optimal in later iterations. The results for both tests of CORDON and LOCATE with the MINILEEDS network are presented in Section 5. Before that Section 4 describes alternative methods developed to overcome the problems with CORDON and LOCATE.

4. GENETIC ALGORITHMS APPLICATIONS This section describes how genetic algorithms were used to develop GA-CHARGE and GA-LOCATE, which are alternative methods to CORDON and LOCATE. A further program, GA-AS, was also developed to meet the additional constraint of charge points forming a closed cordon. Genetic algorithms (GA) are one of the exhaustive searching techniques in artificial intelligence; they are stochastic algorithms whose search methods model some natural phenomena: genetic inheritance and Darwinian strife for survival. Davis and Steenstrup (1987, p. 1) state that: The metaphor underlying genetic algorithms is that of natural evolution. In evolution, the problem each species faces is one of searching for beneficial adaptations to a complicated and changing environment. The ‘knowledge’ that each species has gained is embodied in the makeup of the chromosomes of its members.

The basic idea of the GA approach is to code the decision variables of the problem as a finite string, called “chromosome,” and calculate the fitness (objective function) of each string. Chromosomes with a high fitness level have a higher probability of survival. The surviving chromosomes then reproduce and form the chromosomes for the next generation through the “crossover” and “mutation” process. For the optimal charge, optimal charge location, and optimal cordon problems discussed in this chapter, the chromosomes in GA represent a combination of charge levels for charged links, a combination of charge points, and a charging cordon respectively.

4.1. A GA Application to Solve the Optimal Charge Problem: GA-CHARGE Shepherd and Sumalee (2004) develop GA-CHARGE to solve the optimal charge problem. The process of GA-CHARGE randomly generates an initial set of

92

ANTHONY MAY ET AL.

Fig. 1. Chromosome Structure in GA-CHARGE.

chromosomes representing possible combinations of charge levels on a predefined set of links. As in the traditional approach in GA, the charge level on each link is represented by a binary number. Thus, a chromosome representing charge levels for a given set of charge points is a binary matrix. Figure 1 shows an example of a chromosome with four charge points. The binary number in each column represents a charge level for each charge point. The benefits in terms of social welfare improvement are evaluated for each charge level by running SATURN. GA-CHARGE then selects the parent chromosomes for the next generation based on the performance of each chromosome. The genetic operators, crossover and mutation, are then randomly applied to the parents to produce the offspring.

4.2. A GA Application to Solve Optimal Charge Location Problem: GA-LOCATE The GA based approach applied here, termed GA-LOCATE, uses location indices based on predictions of welfare gains that would result from placing optimal charges in particular locations, as suggested by Verhoef (2002), as its fitness indicator for each chromosome. In this application the length of the chromosome represents the number of charge points required, and the number in each bit identifies the link to be charged. The candidate list of chargeable links has to be prepared in advance by assigning a number to each candidate link. An example of a chromosome in GA-LOCATE is shown in Fig. 2. In this example, four charge points are to be selected (four columns in the string) and links 1, 3, 8, and 6 are to be charged in this chromosome.

4.3. A GA Application to Solve Optimal Cordon Design Problem: GA-AS The last GA based application is GA-AS. This algorithm is developed to find a heuristic optimal location of a closed charging cordon with its associated optimal

Optimal Locations and Charges for Cordon Schemes

93

Fig. 2. Chromosome Structure in GA-LOCATE.

uniform charge around the cordon. A closed charging cordon, in the context of graph theory, is a cordon where all paths from all zones outside it connecting to nodes inside it are charged at least once on a link related to those paths. In other words, all car users driving into or passing through a designated area (charging area) are charged. It should be noted that there is a slight change in the objective function for the optimisation problem. The number of charge points to be selected (to form a closed cordon) is now included as one of the variables for the optimisation process. Therefore, the fitness of chromosomes in GA-AS is the net benefit, which is social welfare minus total scheme cost. Thus, GA-AS finds: (i) optimal location of a closed cordon; (ii) associated optimal uniform charge; and (iii) optimal number of charge points. A chromosome that represents a closed cordon and preserves the formation of a closed cordon, even after the genetic operators (i.e. mutation and crossover) are applied, was designed. The concept of “branch-tree” proposed in Sumalee (2004) was used to encode a closed cordon into a chromosome format. A branch-tree can be seen as a sub-part of a traffic network comprising nodes and links, as shown in Fig. 3. Based on Fig. 3, the simple idea of the branch-tree approach is that, if the charges are imposed on the end links of the branch-tree (termed

Fig. 3. An Example of a Branch Tree.

94

ANTHONY MAY ET AL.

Fig. 4. An Example Network for Closed Cordon Formulation with Branch-Tree.

leaves), all traffic trying to access the nodes in the upper part of the branch will be charged. Figure 4 shows the connection between a branch-tree and a closed cordon in a traffic network. In Fig. 4a, the grey node is the city centre, which is assumed to be the charging area. In a network, a set of links forming an initial closed cordon around the charging area must be predefined. Cordon 1 is defined as the initial cordon. From this initial cordon, a virtual root node (named C1) is defined for the branch, representing an area around which the charge will be imposed, and the first level nodes in the branch are the preceding nodes of the links forming the initial cordon (Cordon 1). Figure 4b shows the branch C1 representing the initial cordon. The original branch in Fig. 4b is then expanded at node E creating the new branch in Fig. 4c. This new branch forms the new cordon, Cordon 2, as depicted in Fig. 4c. In Fig. 4d, the original branch (Cordon 1) is instead expanded at node G resulting in Cordon 3. As mentioned above, the charges will be imposed on all leaves of branch-trees. All three cordons in Fig. 4 are closed cordons. The information about a branch-tree representing a closed cordon has to be converted to a chromosome for GA-AS. The information required for chromosome encoding is: (i) node number; and (ii) node degree. Node degree is the number of child nodes (or the number of nodes in the lower level connected to that node) of

Optimal Locations and Charges for Cordon Schemes

95

Fig. 5. An Example of a Chromosome for a Branch-Tree.

that node. Thus, two integer strings are required for the chromosome, one for the node number, and the other for the associated node degree. Figure 5 illustrates an example of a chromosome representing the branch shown in the same figure.

5. TESTS WITH THE MINILEEDS NETWORK From the previous sections, the contrast between the judgmental and mathematical approaches to cordon design should be clear. Here, a numerical experiment is set up in order to: (i) illustrate the potential uses of CORDON, GA-CHARGE, LOCATE, and GA-LOCATE; and (ii) compare the benefits of two different design methods.

5.1. Network Description and Experimental Setting Figure 6 shows the MiniLeeds network, which is a miniature version of the network for the City of Leeds in the U.K. The MiniLeeds and actual networks should

Fig. 6. MINILEEDS Network with Three Judgmental Cordons.

96

ANTHONY MAY ET AL.

not be compared, as MiniLeeds is a significantly simplified version of the real network. The MiniLeeds network has 89 directional links and 14 zones and is coded in SATURN buffer mode, which means that the delay on each link is estimated without modelling interactions at junctions. The SATEASY module (Hall et al., 1992) was used to allow for an elastic demand. The demand function assumed was:  ␤ 0 Ci (2) Ni = Ni C 0i where N 0i is the number of trips between O-D pair i in do-nothing case, C 0i and Ci are the generalised travel costs between O-D pair i in the do-nothing and dosomething cases, and the constant elasticity with respect to travel costs (␤) was assumed to be −0.57. The value of time adopted for a single user class traveller was 7.63 pence per minute and the vehicle operating cost was 5.27 pence per kilometre. Two numerical tests were conducted. In the first test, three judgmental charging cordons were defined, as shown in Fig. 6. It should be noted that the charges were imposed on inbound traffic only. The optimal uniform charge on each cordon was then calculated by evaluating the benefits of all charge levels through SATURN and then selecting the charge level with the highest benefits. For each charging cordon, the CORDON and GA-CHARGE processes were then applied in order to find the optimal variable charges. In the second test, LOCATE and GA-LOCATE were applied to find the optimal charge locations in the network when only six charge points are permitted. This restriction was imposed to make the number of charging points equivalent to that for the outer cordon. CORDON was then applied to each set of combinations to seek the second-best optimal charges. The optimal uniform charge for the best six charge points found by LOCATE was also computed.

5.2. Modelling Results Ten different schemes were considered, made up from three of the judgmental cordons, all with uniform and second-best variable charges (found by CORDON and GA-CHARGE), and the two sets of six charged links (found by LOCATE and GA-LOCATE) with second-best charges. From now onwards the optimal combinations of six tolled links from LOCATE and GA-LOCATE will be referred to as the best–6L and the best–6G toll sets respectively.

Optimal Locations and Charges for Cordon Schemes

97

Fig. 7. The Optimal 3 and 6 Toll Points Selected by LOCATE for the MINILEEDS Network.

Figure 7 illustrates the results obtained from LOCATE and GA-LOCATE. The bandwidths along the links in the figure represent the directional links selected. The number near each bandwidth identifies the order of the link being selected by the LOCATE process. The arrows along links define the set of links selected by GA-LOCATE. The associated benefits of these schemes are presented in Table 2. The optimal uniform charges for the inner, intermediate, and outer cordons were found to be £0.20, £0.19, and £1.04 respectively at 2000 prices. The outer cordon Table 2. Results for the Tests with the MINILEEDS Network at 2000 Prices. Scheme

Charge Regime

Inner

Uniform Second-best

Intermediate

Uniform Second-best

Outer

Uniform Second-best

Best–6L

Uniform Second-best Second-best First-best

Best–6G All links (FB)

Note: FB: First best.

Charge (£)

0.21 CORDON GA-CHARGE 0.19 CORDON GA-CHARGE 1.04 CORDON GA-CHARGE 0.55 CORDON CORDON Varied

Number Costs of Charge (£/Hour) Points 3 3 3 7 7 7 6 6 6 6 6 6 89

300 300 300 700 700 700 600 600 600 600 600 600 8900

Benefit (£/AM Peak Hour) Gross % of FB 166 363 375 445 946 1,084 923 1,166 1,305 2,574 4,321 4,427 6,287

3 6 6 7 15 17 15 19 21 41 69 71 100

Net −134 63 75 −255 246 384 323 566 705 1,974 3,721 3,827 −2,612

98

ANTHONY MAY ET AL.

produced the highest benefit, which was £923 per single AM peak hour. This is roughly 15% of the potential benefits from the first-best scheme. The optimal uniform charge for the best–6L was found to be £0.55 with benefits of £2,574 per single AM peak hour. This is roughly 40% of the potential benefits from the firstbest scheme. The benefits from the best–6L with the optimal uniform charge was about 280% of the benefits from the best judgmental cordon (the outer cordon) with the uniform charge. When the charge levels on the three judgmental cordons were varied GACHARGE found a slightly better set of charge levels than CORDON for all three judgmental cordons. The second-best charges produced higher benefits in all judgmental cordons compared to the uniform tolls. The benefits were improved by over 125% for the case of the inner and intermediate cordons compared to the uniform tolls. The benefits from the best–6L and the best–6G with second-best charges were £4,321 and £4,427 per single AM peak period respectively. The benefit produced by the best–6G was slightly higher than the benefit from the best–6L (approximately 0.9% higher). The benefits from the best 6G were about 150% higher than the benefits from the outer cordon with second-best charges. This illustrates the substantial advantage of using LOCATE and GA-LOCATE to identify appropriate charge locations in a network as well as using CORDON and GA-CHARGE to calculate the optimal second-best charges over uniform charges.

6. TESTS WITH THE EDINBURGH NETWORK: GA APPLICATIONS The Edinburgh network is a simplified version of the network of the City of Edinburgh, again modelled by SATURN in a buffer mode. The network has 350 links and 25 zones. Figure 8 shows the network and four judgmental charging cordon designs. All of the cordons were tested with different uniform charge levels to find their optimal uniform charges. In order to display the effect of variable charges and alternative locations three other theoretical approaches were adopted. The first simply took the system optimal tolls derived automatically by SATURN (when all links may be charged) and selected the highest ten charges only. These top-ten charges and their locations are depicted in Fig. 9. The second method applied GA-CHARGE to the Inner 1 and Outer 1 cordons. The aim was to test the effect of varying charges around a cordon. The third method applied GA-AS to the network to find the best two charging cordons, OPC1 and OPC2, with their associated optimal uniform charges. OPC1 and OPC2 are shown in Fig. 10. Benefits were also compared to the first best solution derived by SATURN.

Optimal Locations and Charges for Cordon Schemes

99

Fig. 8. The Edinburgh Network with Four Judgmental Cordons.

Table 3 shows the results of the tests. The optimal uniform charges for the four judgmental cordons vary between £0.50 and £2.25 and the best of the judgmental cordons is the Outer 1 cordon, with a benefit of £6,200 for a single peak hour. OPC1 and OPC2 produce the highest and second highest net benefits from the optimisation process. The optimal charge for both OPC1 and OPC2 is £1.50. The gross benefits from the judgmental cordons with uniform charges are between 8 and 17% of the gross benefits from first best charging. When variable charges are applied to the Inner 1 and Outer 1 cordons gross benefits increase to 10 and 20% of the gross benefits from first best charging. In other words, moving from uniform to variable charges increases benefits by 20%. OPC1 and OPC2 with optimal uniform charges of £1.50 achieve gross benefits 20 and 10% greater than the best judgmental cordon respectively. Relaxing the requirement for a closed cordon and selecting the ten best charging points with uniform charges (£0.80) increases gross benefits to 24% of the gross benefits from first best charging, while allowing the charges to vary as shown in Fig. 9 increases gross benefits further to 41% of the benefits from first best charging.

100

ANTHONY MAY ET AL.

Fig. 9. The Top-Ten Links with the Highest Marginal Cost Charges in the Edinburgh Network.

It is unlikely that such variable charges would be implemented in practice, though a two zone system may well be adopted, as is being considered in the Dutch vehicle-km based charging proposal. The top-ten links were assigned to two sets as shown in Fig. 9. The optimal combination of charges was found to be £0.50 and £2 for the inner and outer sets respectively. This system of charges yielded benefits that were 32% of the benefits from first best charging. This result is mid-way between the uniform and variable solution for these links. The topten results demonstrate the advantage of implementing more complex charging structures. The benefits discussed above were computed before any implementation and operating costs were considered. In practice the equipment and operational cost of the system, which depends on the number of toll points, should be deducted from these gross benefits to give a net benefit. The capital and operating costs were estimated to be £183,400 per charge point and £85,300 per charge point per annum respectively at 2000 prices (Oscar Faber,

Optimal Locations and Charges for Cordon Schemes

101

Fig. 10. Optimal Charging Cordons Found by GA-AS for the Edinburgh Network.

2001). This is equivalent to £100 per toll point per peak-hour if charges are assumed to apply over 1,000 peak hours per year, the schemes are assumed to have a life of 30 years, and a discount rate of 6% is used. The four judgmental cordons have 9, 7, 16 and 20 toll points respectively, which imply costs of £900, £700, £1,600 and £2,000 per peak hour as shown in Table 3. The top-ten links cordon would incur costs of £1,000. The OPC1 and OPC2 cordons have 10 toll points, which cost £1,000 per peak hour. These costs are relatively low and all test results imply that the costs would be met in full. However if similar costs per toll point were assumed for the first best solution then the costs would be in the order of £35,000 per peak hour, which would result in net benefits significantly lower than those from the top ten, OPC1, OPC2, and Outer 1 cordon with uniform and varied charges. The top ten links with variable charges is the design with the highest net benefits (£14,500 per peak hour). The net benefits produced by OPC1 and OPC2 are about 40 and 20% higher than the net benefit from the Outer cordon 1 (the best judgmental

102

Table 3. Results for the Test with the Edinburgh Network at 2000 Prices. Cordon

Note: NA: Not applicable.

Capital/Operation Costs Per Peak Hour (£k)

Gross Total Benefit Per Peak Hour (£k)

% of Gross Total Benefit Compared to First-Best

£0.50 £0.75 £2.25 £0.75 Varied

0.90 0.70 1.60 2.00 0.90

3.00 4.69 6.17 3.96 3.62

8.1 12.6 16.6 10.6 9.7

2.10 3.99 4.57 1.96 2.72

Varied

1.60

7.41

19.9

5.81

1.30 1.30 1.00

7.64 6.85 8.99

20.5 18.4 24.2

6.34 5.55 7.99

1.00

15.50

40.7

14.50

1.00

11.96

32.2

10.96

35.00

37.19

100.0

2.19

£1.50 £1.50 £0.80 Varied 50 p and £2 for inner and outer set NA

Net Benefit Per Peak Hour (£k)

ANTHONY MAY ET AL.

Inner cordon 1 Inner cordon 2 Outer cordon 1 Outer cordon 2 Inner cordon 1 with varied charges (GA) Outer cordon 1 with varied charges (GA) OPC1 OPC2 Top ten links with uniform charge Top ten links with varied charges Top ten links with two level charges First-best condition

Optimal Charge

Optimal Locations and Charges for Cordon Schemes

103

cordon) respectively. Note that top 10 was dominated by one link – the Forth bridge which was not available to be selected by GA-AS.

7. CONCLUSIONS Shepherd and Sumalee (2004) showed that the CORDON approach can solve the second-best charging problem in most but not all cases. It was reported to fail due to changes in the path set or as a result of assignment convergence errors. The GA-CHARGE approach was shown to be successful in solving for optimal charge levels for a given set of links, giving slight improvements over the CORDON process. The incremental LOCATE approach performed well in the case study of MiniLeeds, but in general suffers from the weakness that previously selected links cannot be de-selected when building a combination of charge points. The GA-LOCATE approach gives only a slight improvement in the case presented as many of the links selected by LOCATE are also in the GA solution. The application of GA-AS has proved successful in producing closed cordons with optimal uniform charges around the cordon, which consistently out perform the judgemental based approach. The GA based method was found to be time consuming and there is no proof of convergence of the algorithm. However, the evidence of successful implementation of GA based methods has been growing in the literature. Although the tests reported in this chapter are only limited to a simplified network, the results in terms of the relative social welfare improvement values show that GA based methods can find at least a good heuristic optimal solution, particularly for the case of the charge location problem. Cordon-based pricing appears likely to remain as one of the preferred methods of implementing road pricing in the medium term. The other preferred method is an area licensing system, such as that which has been in operation in London since February 2003. While various analytical studies have investigated a wide range of designs of cordon tolls, the surveys conducted between practitioners indicate that the focus tends to be on a much simpler set of design options, prompted largely by considerations of achieving acceptability and avoiding adverse impacts. While this is understandable, there is a real danger that simple designs will produce substantially lower benefits, and impose unnecessarily adverse distributional impacts. The tests based on Leeds and Edinburgh confirmed the initial assumption that simple cordons may be suboptimal. The judgmental cordons tested varied substantially in their performance, with the best producing more than twice the benefits

104

ANTHONY MAY ET AL.

of the worst. When a more sophisticated method was used in Edinburgh to help define the location of a charging cordon, the benefit of the scheme was around 25% higher than that from the best judgmental cordon. Relaxing the requirement for a closed cordon, and instead identifying the most effective charging points, with a uniform charge, added around 50% to the benefits of the best judgmental cordon in Edinburgh, and almost trebled the benefits in the simpler Leeds network. Further relaxing the requirement for uniform charges at all points increased benefits by around 20% on the cordons and by over 60% on the top ten links in Edinburgh. It is clear from these results that there is the potential for substantial improvements in the performance of cordon-based pricing as a means of improving network efficiency. It is probable that there will be similarly large improvements in environmental performance. Further work is needed to ensure that the newly developed methods are fully effective, and to test the distributional impacts they generate.

ACKNOWLEDGMENTS The research reported here was supported financially by the U.K. Engineering and Physical Sciences Research Council and the Department for Transport. The authors are grateful to both of these bodies for their support. The contributions of Erik Verhoef and Dave Milne, Ronghui Liu, and David Watling are also acknowledged. The conclusions, however, are the authors’ own.

REFERENCES Davis, L., & Steenstrup, M. (1987). Genetic algorithms and simulated annealing: An overview. In: M. L. Davis (Ed.), Genetic Algorithms and Simulated Annealing (pp. 1–11). Los Altos, CA: Morgan Kaufmann Publishers. Hall, M. D., Fashole-Luke, T., Van Vliet, D., & Watling, D. P. (1992). Demand responsive assignment in SATURN. Paper presented at Seminar E, 20th Summer Annual Meeting, PTRC (Planning and Transport Research and Computation), UMIST, Manchester. May, A. D., Coombe, D., & Travers, T. (1996). The London congestion charging research programme 5: Assessment of the impacts. Traffic Engineering and Control, 37, 403–409. May, A. D., Liu, R., Shepherd, S. P., & Sumalee, A. (2002). The impact of cordon design on the performance of road pricing schemes. Transport Policy, 9, 209–220. MVA Consultancy (1999). Road pricing in Leeds: Final report. Leeds: Leeds City Council. Oscar Faber Consultancy (2001). Road user charging study-West Midlands. Final report prepared for Birmingham City Council. Unpublished. Sheffi, Y. (1985). Urban transportation networks: Equilibrium analysis with mathematical programming methods. Englewood Cliffs, NJ: Prentice-Hall.

Optimal Locations and Charges for Cordon Schemes

105

Shepherd, S. P., May, A. D., & Milne, D. S. (2001a, July). The design of optimal road pricing cordons. 9th World Conference on transport research (CD-ROM, Paper No. 3432), Seoul. Shepherd, S. P., May, A. D., Milne, D. S., & Sumalee, A. (2001b). Practical algorithms for finding the optimal road pricing location and charges. Paper presented at the European Transport Conference, Cambridge, U.K. Shepherd, S. P., & Sumalee, A. (2004). A genetic algorithm based approach to optimal toll level and location problems. Networks and Spatial Economics, 4, 161–179. Sumalee, A. (2001). Analysing the design criteria of charging cordons. Institute for Transport Studies. University of Leeds, ITS Working Paper No. 560. Sumalee, A. (2004). Optimal road user charging cordon design: A heuristic optimisation approach. Computer-Aided Civil and Infrastructure Engineering, 19, 377–392. Van Vliet, D. (1982). SATURN – A modern assignment model. Traffic Engineering and Control, 23, 578–583. Verhoef, E. T. (2002). Second best congestion pricing in general networks: Heuristic algorithms for finding second-best optimal toll levels and toll points. Transportation Research B, 36, 707–729. Wardrop, J. (1952). Some theoretical aspects of road traffic research. Proceedings of the Institution of Civil Engineers Part II, 1, 325–378.

This Page Intentionally Left Blank

5.

TIME-VARYING ROAD PRICING AND CHOICE OF TOLL LOCATIONS

Andr´e de Palma, Robin Lindsey and Emile Quinet 1. INTRODUCTION For decades the literature on road congestion pricing has focused on deriving optimal tolls when tolling is implemented either on the whole road network, or on a pre-specified subset of links. Attention has recently turned to determining how many links should be tolled and where. Verhoef (2002) proposes heuristic algorithms for identifying optimal toll levels and locations. Shepherd and Sumalee (2004), May et al. (2002), and Zhang and Yang (2004) build on Verhoef’s work by developing genetic algorithm based methods. An update of this work is presented in Chapter 4 of this volume. These studies concentrate on developing and implementing reliable algorithmic methods, and offer only limited insights into the factors that determine the best choices. They also adopt static models and therefore do not incorporate departure time decisions or the time structure of tolls. The purpose of this paper is to include these dynamic aspects and to emphasise intuition. The paper focuses on three questions: Priority: Which links yield the largest benefits from tolling? Returns to scale: Are there constant, increasing or decreasing marginal benefits from tolling successively more links? If there are decreasing benefits then it may

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 107–131 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09005-5

107

ANDRE´ DE PALMA ET AL.

108

be optimal to toll a few key links while leaving the rest toll-free. If the benefits are increasing, then an all-or-nothing tolling strategy may be appropriate. Concentration: Should tolling be concentrated on one part of the network, or spread out? The choice can have jurisdictional and political implications, as well as different welfare distributional impacts on the travelling population. A few results pertinent to these questions can be gleaned from the literature. For a network with one origin-destination (O-D) pair joined by two parallel links or routes it has been shown (de Palma & Lindsey, 2000; Verhoef et al., 1996) that it is generally better to toll the route that carries more traffic without tolls. With time invariant tolling, there are increasing returns to tolling; i.e. tolling both routes yields more than twice the benefits from tolling only one (Verhoef, 2002; Verhoef et al., 1996). With time-varying tolling and inelastic demand on the other hand, the benefits can be diminishing (Braid, 1996). Tolling links or routes in series generally yields diminishing benefits regardless of the tolling regime when tolling one route is a good substitute for tolling the other. Having said that, tolling at more than one location along urban arterial routes where congestion builds towards the centre does enhance the effectiveness of congestion pricing (Kraus, 1989). Neither simple parallel nor simple series link configurations are suitable for studying concentration, and nothing appears to have been written about it. This paper builds on the results mentioned above regarding priority and returns to scale, and provides a first attempt to study concentration. Section 2 describes the model and characterises equilibrium mode choice, departure time choice and route choice decisions. Section 3 defines the system optimum and discusses the main factors governing optimal tolls. Base case parameter values for the simulations are given in Section 4. Section 5 provides a relatively detailed analysis of tolling on a two route network. Section 6 follows up with four routes. A highlight of the results is the importance of the degree of time variation of tolls in governing returns to scale from tolling, and the relative merits of geographical concentration and dispersal. Conclusions are summarised in Section 7.

2. THE MODEL The geometry of the urban road network is shown in Fig. 1. Residents live on a ring road with a population density of ␳ per unit distance. All trips are bound to the centre at point C, and are taken either by auto or another mode.1 If auto is chosen then departure time and route choice decisions must also be made. Travel by auto is described in Section 2.1, and travel by the alternative mode and mode choice decisions in Section 2.2.

Time-Varying Road Pricing and Choice of Toll Locations

109

Fig. 1. A Radial Network with Four Routes.

2.1. Auto Travel To reach the centre by car, drivers must take one of the arterial routes that link the ring road to the centre. The set of routes is denoted by R = {1, . . ., R}. Figure 1 depicts an example with R = 4. The entrance to route r is labelled Er . Travel distance counterclockwise from Er to Er+1 MOD R is Lr , r ∈ R, where MOD denotes modulo. Route r draws drivers residing within an attraction zone, Ar , that extends − a distance x + r counterclockwise from Er , and a distance x r clockwise from Er . The boundaries of the Ar , r ∈ R, are related by the R equations − x+ r + X r+1 MOD R = L r ,

r ∈ R.

(1)

Drivers travel one per vehicle and have identical trip cost functions. Under free-flow conditions, combined travel time and vehicle operating cost net of taxes and tolls on route r from Er to the centre is Kr . The unit cost of travel on the ring road is k. If a tax of u ≥ 0 per unit distance is applied, the free-flow cost of a trip via route r is K r + uD r + (k + u)x, where x is distance to Er on the ring road, and Dr is the length of route r. (It should be noted that the Dr need not be the same for all routes if the arterials do not run in straight lines, if the ring road is not circular, or if the “centre” is not in the middle of the ring road.)

ANDRE´ DE PALMA ET AL.

110

2.1.1. Congestion and Departure Time Decisions Vickrey’s (1969) bottleneck model is used to describe the dynamics of congestion and trip timing decisions. Drivers incur a travel time cost of ␣ per hour. They have a common preferred arrival time at the centre, and incur schedule delay costs of ␤ per hour from arriving early, and ␥ per hour from arriving late. Route r has a bottleneck with a flow capacity of sr . If the arrival rate of vehicles at the bottleneck exceeds sr , a queue will develop. The equilibrium arrival rate on route r, and consequently the travel costs as well, depend on the tolling regime on route r. (It should be noted that tolling of routes is equivalent to tolling of arterial links on the network of Fig. 1.) Four regimes are considered: no toll, flat tolling, step tolling, and fine tolling. In the no toll regime there is no toll. Flat tolling entails a base toll that is constant throughout the day. Step tolling comprises a base toll, and a peak toll that is levied during part of the arrival period. The magnitude and timing of the peak toll on each route are chosen to minimise social travel costs conditional on the number of users. Finally, in the fine tolling regime the toll is set at a base level outside the arrival period, and changes smoothly within it to deter queuing. It can be shown (Arnott et al., 1990) that for all tolling regimes, in a departure time choice equilibrium the average private variable travel cost on route r can be expressed as ␶r + ␦r N r /s r , where ␶r is the base toll, Nr is the number of users (or trips) on route r, and ␦r is a parameter that depends solely on the tolling regime on r. Define ␦ ≡ ␤␥ (␤ + ␥)−1 . In the no toll, flat, and fine tolling regimes, Fine = ␦. In the step tolling regime ␦Step = ␦ if ␥ ≤ ␣, and ␦No = ␦Flat =␦

Step −1 −1 = ␦ 1 − (␥ − ␣) ␤ (␤ + ␥) (␣ + ␥) /2 < ␦ if ␥ > ␣. When the free␦ flow and variable cost components are added together, average private travel cost on route r is C pr (x) = K r + uD r + (k + u)x + ␶r +

␦r N r , sr

r ∈ R,

(2)

where ␦r ∈ {␦No , ␦Flat , ␦Step , ␦Fine } as applicable. Route choice decisions − Given Eq. (2), the attraction zone boundaries, x + r and x r , r ∈ R, can be solved using the equations − C pr (x + r ) = C r+1 MOD R (x r+1 MOD R ), p

r ∈ R.

Drivers who live within attraction zone Ar take route r to the origin.

(3)

Time-Varying Road Pricing and Choice of Toll Locations

111

2.2. Travel by the Alternative Mode and Mode Choice In addition to auto, there is another mode (transit, bicycle, walking) with a dense network in the sense that travel cost to the centre is the same from all points on the ring road. To be concrete, this other mode will be called transit. Transit is assumed to be congestion-free, and the fare (if any) is assumed to be exogenous and equal to the marginal cost per passenger of providing the service. In addition to any fare the cost of taking transit, CS , includes the cost of walking time, waiting time, physical discomfort, etc. Parameter CS is assumed to be uniformly distributed in the population residing at each location on the ring road over the range [C- S , C- S + ]. The range is assumed to be large enough that both modes are used at every location. A resident within Ar who is indifferent between auto and p transit incurs a transit cost of C Sr (x) = C r (x). The density of residents who drive is thus ␳r (x) = ␳−1 (C- S +  − C Sr (x)), and the total number of drivers on route r is  xr+  xr− Nr = ␳r (x) dx + ␳r (x) dx, r ∈ R. (4) x=0

x=0

3. SOCIAL OPTIMUM The social optimum is defined as the maximum social surplus. Given fixed trip demand, the optimum can be derived by minimising total social cost for auto and transit trips. Since tolls are a transfer, not a resource cost, the social cost of driving is less than the private cost by the amount of toll revenue, which includes revenue from the base toll and revenue generated through the time variation of the step and fine tolls. The marginal social variable cost of a trip on route r can be written as ␾r N r /s r (Arnott et al., 1990), where ␾r is a parameter that depends on the tolling regime as follows: ␾No = ␾Flat = ␦, ␾Fine = ␦/2, ␾Step = 3␦/4 if ␥ ≤ ␣, and ␾Step = ␦/4[3 − (␥ − ␣)␤(␤ + ␥)−1 (␣ + ␥)−1 ] if ␥ > ␣. Since ␾No = ␾Flat > ␾Step > ␾Fine , the marginal social variable trip cost declines with the degree of time variation in the tolling regime. In addition to congestion, it is assumed that driving creates an environmental cost of e ≥ 0 per unit distance on the ring road and on the arterial routes. The environmental externality can be internalised with a distance tax, and this is one of the pricing regimes to be considered.2 Like tolls, the distance tax is a transfer, not a social cost. Taking into account both toll revenue and environmental costs, the social cost of an auto trip on route r is C or (x) = K r + eD r + (k + e)x + ␾r N r /s r , where ␾r ∈ {␾No , ␾Flat , ␾Step , ␾Fine } as applicable.

r ∈ R,

(5)

ANDRE´ DE PALMA ET AL.

112

As noted earlier, transit is assumed to be congestion free and it is also assumed to generate no environmental externalities. Social cost for residents within Ar and a distance x to Er is therefore  CrS (x) z dz, TCor (x) = ␳r (x)C or (x) + ␳−1 S z=C and total social cost within Ar is  xr+  xr− o o TCr = TCr (x) dx + TCor (x) dx. x=0

x=0

Denote the set of routes that are tolled by T, where T ⊆ R. The tolling regime employed on each route in T is treated as given. Define ␶r as the base toll on route r, r ∈ T. It is possible for the optimal ␶r to be negative with any of the three tolling regimes. For example, as Braid (1996) shows, if a route has a high capacity it may be desirable to attract drivers onto it in order to alleviate congestion on adjacent routes. Negative tolls are presumed to be feasible and acceptable. In defence of this assumption it should be noted that: (a) total travel demand is fixed in the model, which rules out frivolous trips; (b) negative tolls can be imposed using electronic toll collection technology (e.g. smartcard credit balances can be incremented); and (c) charging negative tolls enhances the political acceptability of road pricing since some motorists benefit financially, and immediately.3 If a distance tax can be levied as well as tolls, the social optimum problem given T can be written as Min

R 

␶T r,r∈T ,u

TCor

(6a)

r=1

subject to the user equilibrium conditions in Eqs (1), (3), and (4) that determine − x+ r , x r and N r , r ∈ R. If a distance tax cannot be levied, the system optimum problem is Min

R  TCor

␶T r,r∈T r=1

(6b)

subject, again, to Eqs (1), (3), and (4). Let TC0 (T) denote the minimum total social cost obtained by solving problem (6a). Problem (6b) can be treated in a similar way. The benefit from tolling T is then G T ≡ TC0 (∅) − TC0 (T). In comparing alternative choices for T it will prove convenient to express benefits as a fraction of the benefits from tolling all routes. T R This fraction is denoted g R T ≡ G /G . Hereafter, tolling all routes will be called universal tolling, and tolling a subset of routes, partial tolling.

Time-Varying Road Pricing and Choice of Toll Locations

113

In summary the overall model can be viewed as a bi-level programming problem. At the upper level a planner minimises social costs by choosing tolls (and possibly a distance tax), taking the set of routes that can be tolled and the tolling regime on each route as given. At the lower level residents minimise their individual travel costs. Despite the simplicity of the model, the bi-level program has to be solved numerically. This is done iteratively as shown by the flow chart in Fig. 2. Upper-level steps are identified with thick borders, and lower-level steps by thin borders and shading. Residents make their mode choice, auto departure time choice and auto route choice decisions in that order. With the bottleneck model the departure time decisions do not have to be solved explicitly because, as noted in Section 2.1, they are fully described in reduced form by the ␦r parameters. Mode choice and route choice decisions are solved iteratively in the lower level as shown. Before turning to the simulations, it is instructive to review the three main goals of tolling: demand reduction, traffic reallocation, and revenue generation. Verhoef et al. (1996) and de Palma and Lindsey (2000) discuss the first two: Demand reduction: When congestion is above the optimal level, defined as the level at which marginal social cost is equal to marginal social benefit, it is desirable to reduce the number of auto trips. The congestion externality on route r increases with the volume of traffic on it, Nr , and with its congestion characteristic, ␦r /sr with no tolling, and ␾r /sr with tolling (see Eqs (2) and (5)). If tolling reduces the congestion characteristic, then ␾r < ␦r . Reducing distance driven to curb emissions is also desirable unless the emissions externality is fully internalised through the distance tax. Traffic reallocation: Traffic can be reallocated from more congested (M) routes to less congested (L) routes either by tolling M or, if M cannot be tolled, by reducing tolls on L in order to attract drivers from M to L. The potential to reallocate traffic from M to L depends on the cross-price elasticity of demand between auto and transit. The higher the elasticity, the less effective the toll reductions will be because they increasingly draw travellers away from transit rather than from untolled routes. Reallocation is also difficult when the cost of travel on the ring road (k) is high, which makes switching routes costly for drivers. Reallocation is also socially costly if it increases total distance driven, and if emissions are costly (high e).

Demand reduction requires a positive base toll, whereas optimal traffic reallocation may require a negative base toll. The results of the simulations in Section 5 show the potential conflict between the two goals. Revenue generation is a third goal of tolling if the agency responsible for managing roads is subject to a budget constraint. In this paper attention is limited to assessing whether toll revenues at least cover collection costs, and how much of the revenues might have to be rebated to travellers, or at least spent in ways that benefit them, in order to gain public support for road pricing.

114

ANDRE´ DE PALMA ET AL.

Fig. 2. Bi-Level Programming Structure of the Model.

Time-Varying Road Pricing and Choice of Toll Locations

115

4. PARAMETER VALUES The base case parameter values describe a medium-sized city. Arterial routes are of equal length: D r ≡ D = 2.5R km (about 1.5R miles). The route lengths are chosen to be proportional to the number of routes, R, in order to maintain a constant distance between the entrances around the circle, and therefore the degree of substitutability between routes. The ring road is circular, and has a circumference of 5␲R km (about 3␲R miles). Entrances to the arterials are equally spaced around the ring road: L r = 2␲D/R = 5␲ km (about 3␲ miles). Population density is set equal to unity (␳ = 1) so that total population is 5␲R. Route capacities are assumed to be equal, and chosen so that if all residents elected to drive, arrivals at the centre would be spread over a two hour period. This implies s r = 5␲/2 vehicles/hour. Unit costs of time are taken to be ␣ = $10/hour, ␤ = $6/hour, and ␥ = $25/hour. The ratios ␤/␣ and ␥/␣ follow approximately from Small’s (1982) estimates. Freeflow travel costs on the arterials are K r = $(0.75 + 5R/16), and the unit cost of travel on the ring road is k = $0.28/km (about $0.47 per mile).4 Estimates of the external cost of road transport emissions (European Conference of Ministers of Transport, 1998; Mayeres, 2001) range from $0.04/km to $0.10/km (about $0.06 to $0.16 per mile). In order to simplify the analysis a value of e = 0 is chosen as a base case in order to focus on congestion and congestion tolling rather than on emissions. A value of e = $0.10/km (about $0.16 per mile) is used for the sensitivity analysis. Finally, the parameters describing the cost of using transit are C- S =$6 and  = $20.5

5. TWO ROUTES The simulations for two routes focus on the questions of tolling priority and returns to scale of benefits. The choice between concentration and dispersal of toll roads is studied using a four route network in Section 6.

5.1. Base-Case Solution Table 1(a) provides summary statistics for the base-case equilibria. In the no toll equilibrium (NTE) each of the two routes carries half the total traffic of N ≡ N 1 + N 2 = 24.90. The auto share of trips is N/(10␲) = 0.79, and the generalised cost elasticity of auto usage is about −0.43. Partial tolling is assumed to be implemented on Route 1. Since the two routes are identical in the base case, tolling Route 2 would be equivalent. With flat tolls, partial tolling reduces N1 by

116

Table 1. Selected Results for Two Routes. ␶2 [$]

2.71 5.78 0.56 3.34 −2.89 0.00

0 5.78 0 3.34 0 0.00

(b) e = 0, distance tax implemented None ∅ Flat 1 0.47 R 5.78 Step 1 −2.13 R 3.33 Fine 1 −5.94 R 0.00 (c) e = 0.10, distance tax implemented None ∅ Flat 1 0.47 R 5.56 Step 1 −2.02 R 3.20 Fine 1 −5.65 R 0.00

T

(a) e = 0, no distance tax None ∅ Flat 1 R Step 1 R Fine 1 R

Source: Own calculations (see text).

Distance Tax [$/km]

Benefit Per Capita (Per Driver) [$]

gR T

N1

N2

N

0 0 0 0 0 0 0

12.45 10.03 9.39 12.23 10.83 15.10 12.45

12.45 13.46 9.39 12.54 10.83 11.35 12.45

24.90 23.49 18.78 24.77 21.67 26.46 24.90

0.17 (0.23) 0.75 (1.25) 0.82 (1.05) 1.83 (2.65) 1.65 (1.96) 3.04 (3.84)

0.23 1 0.45 1 0.54 1

0 5.78 0 3.33 0 0.00

0.60 0.58 0.00 0.58 0.00 0.56 0.00

9.60 9.38 9.39 11.45 10.83 14.17 12.45

9.60 9.79 9.39 9.31 10.83 8.72 12.45

19.19 19.17 18.78 20.76 21.67 22.89 24.90

0.70 (1.14) 0.00 (0.00) 0.05 (0.08) 0.57 (0.87) 1.13 (1.64) 1.30 (1.78) 2.34 (2.96)

0.04 0 0.51 1 0.55 1

0 5.56 0 3.20 0 0.00

0.68 0.66 0.10 0.65 0.10 0.64 0.10

9.23 9.03 9.03 10.95 10.42 13.45 11.98

9.23 9.42 9.03 9.01 10.42 8.51 11.98

18.46 18.44 18.07 19.96 20.84 21.96 23.96

0.89 (1.51) 0.00 (0.00) 0.05 (0.08) 0.53 (0.83) 1.05 (1.58) 1.19 (1.70) 2.17 (2.84)

0.04 1 0.50 1 0.55 1

ANDRE´ DE PALMA ET AL.

␶1 [$]

Tolling Regime

Time-Varying Road Pricing and Choice of Toll Locations

117

nearly 25%, but yields a benefit of only $0.17 per capita because of increased congestion on Route 2. With universal tolling, adverse traffic reallocation effects are avoided, the toll more than doubles, and benefits increase more than fourfold, making the benefits of tolling both routes strongly increasing. With step tolling, the partial and universal base tolls are both lower than with flat tolling because the step toll reduces the queuing externality. Benefits are considerably higher than those from flat tolling, and weakly increasing. Finally, under fine tolling queuing is eliminated and base tolls are even lower than base tolls under step tolling. As foreshadowed in Section 3, the partial base toll is negative in order to draw traffic off the untolled route. The universal base toll is zero because drivers who arrive either at the beginning or the end of the peak when the base toll is in effect impose neither a schedule delay nor a congestion externality (Arnott et al., 1993). Sensitivity of the results to various parameter values is now explored.6

5.2. Auto Usage Price Elasticity To vary the price elasticity of auto travel, parameters C- S and  were modified while keeping the auto share of trips in the NTE fixed at the base-case value. Figure 3 plots for flat, step and fine tolling the benefits from partial tolling of Route 1 as a

Fig. 3. Benefits from Partial Tolling as Fraction of Universal Tolling Benefits vs. Auto Elasticity (Two Routes).

ANDRE´ DE PALMA ET AL.

118

fraction of the benefits from universal tolling. This fraction is denoted g R 1 , where R = {1, 2}, and labelled “g1” in Fig. 3. Flat tolling: With inelastic auto demand, flat tolling yields little benefit because auto travel is relatively fixed and the 50-50 route split that prevails in the NTE is optimal given identical routes. Moreover, partial tolling has a low relative benefit because adverse traffic reallocation effects dominate gains from demand reduction. As the elasticity rises, the benefits from demand reduction increase. Usage of the two routes becomes less interdependent, g R 1 rises towards 0.5, and the benefits from flat tolling become less increasing. Step tolling: With inelastic auto demand, the partial base toll is negative because the gain from drawing traffic off the untolled route outweighs the gain from demand reduction. In contrast, the universal base toll is well above zero because no traffic reallocation occurs. Unlike the case with flat tolling, g R 1 starts above 0.5, drops below 0.5 as elasticity rises, and then rises back to 0.5. Hence benefits from partial tolling are initially decreasing, then increasing, and ultimately constant. g R 1 reaches a minimum at an elasticity near where demand reduction and traffic reallocation incentives offset. At this point, the base step toll is zero, step tolling has no effect on N1 or N2 , and is useful only in reducing queuing on Route 1. Fine tolling: Fine tolling differs from step tolling in two respects. First, the absolute gains from universal tolling are independent of the demand elasticity, and second, the relative gain from partial tolling remains above 0.5. The gains from fine tolling are therefore decreasing at all (finite) elasticities. In summary, the simulation results show that the returns to scale from time varying tolls are less strongly increasing (or more strongly decreasing) than the returns from flat tolls. The general implications for how much of a road network should be tolled are ambiguous. The greater absolute benefits derived from time varying tolls suggest that network coverage should be more extensive with time varying tolls than with flat tolls. The differences in returns to scale, on the other hand, suggest that tolling just part of the network may be cost effective for time varying tolls, whereas an all-or-nothing approach would be suitable for flat tolls. To the extent that infrastructure and operating costs are higher for time varying tolls than for flat tolls, this conclusion is reinforced. 5.3. Route Capacities Differences between route capacities were created by increasing s1 , holding s 1 + s 2 constant. Figure 4 plots the relative benefits of partial tolling on each route, g R 1 R R R and g R 2 , as well as their sum, g 1 + g 2 . In each tolling regime, as s1 increases g 1 increases towards unity and g R 2 decreases towards zero. Tolling the higher capacity

Time-Varying Road Pricing and Choice of Toll Locations

119

Fig. 4. Benefits from Partial Tolling as Fraction of Universal Tolling Benefits vs. Capacity Share (Two Routes).

route (Route 1 here) provides greater benefits because the higher capacity route R carries the majority of traffic. Throughout the range of variation, g R 1 + g 2 < 1 for flat and step tolling so that benefits in these regimes remain increasing. For fine R tolling, g R 1 + g 2 > 1, so that benefits remain decreasing.

5.4. Free-Flow Travel Costs To create differences between free-flow travel costs, K1 was reduced while holding K 1 + K 2 constant. This exercise can be interpreted as raising the speed limit on Route 1 relative to Route 2, and Route 1 will therefore be called the “quicker” route. R Figure 5 shows that in all tolling regimes, as K1 decreases g R 1 rises and g 2 falls. Tolling the quicker route is preferred. Although like the higher capacity route in Section 3 the quicker route draws more traffic, unlike the higher capacity route the quicker route is also more congested. Figure 5 reveals that tolling the quicker route is relatively more advantageous for flat tolling than for step or fine tolling. With flat tolling, imposing a positive toll on the quicker route is beneficial for both demand reduction and traffic reallocation because, as is well known, quick routes tend to attract too large a share of traffic. The problem is that with flat and fine tolling, the desire to shift traffic from the quicker to the slower route conflicts with the desire to reduce queuing on the quicker, tolled, route. The traffic reallocation

ANDRE´ DE PALMA ET AL.

120

Fig. 5. Benefits from Partial Tolling as Fraction of Universal Tolling Benefits vs. Free-flow Travel Cost (Two Routes).

incentive is therefore mixed, and this dilutes the advantage of tolling one route rather than the other.

5.5. Route Spacing The effect of route spacing was explored by increasing L2 while holding the length of the ring road, L 1 + L 2 , constant. Doing so has no effect on the relative benefits of partial tolling. The absolute gains from tolling fall slightly because average travel distance on the ring road rises, and this causes auto usage and congestion to drop.

5.6. Travel Cost on the Ring Road As the unit cost of travel time on the ring road (k) rises, switching to less accessible routes becomes more costly for drivers, and the boundaries of the attraction zones become less sensitive to differences in arterial travel costs. As a result, the relative benefits of partial tolling slowly tend towards 0.5 in all tolling regimes, and benefits approach constant returns to scale. The absolute benefits from universal tolling fall

Time-Varying Road Pricing and Choice of Toll Locations

121

at the same proportional rate in all regimes. As in the route spacing experiment, this happens because the auto share of trips declines. 5.7. A Distance Tax A distance tax is an imperfect tool for congestion management for three reasons: (a) it is paid to travel on the ring road even though the ring road is assumed to be congestion free; (b) the tax is the same on all routes; and (c) the tax does not vary over time. Nevertheless, a distance tax can play an auxiliary role in reducing congestion if tolling is either partial or non-existent. Table 1(b) shows the impacts of an optimal combined distance tax and tolling policy. With no tolling, the distance tax is set at a hefty $0.60/km (about $0.97 per mile), and yields a benefit of $0.70 per capita. Introducing either partial or universal flat tolling provides very little additional benefit because flat tolling is not time sensitive, and also because limitation (a) of the distance tax noted above turns out to be minor. The incremental benefits from step and fine tolling remain substantial, although they too are reduced appreciably compared to the case where there is no distance tax. 5.8. Environmental Externalities To explore whether environmental costs have an appreciable impact on optimal tolling, the emissions cost parameter was increased from zero to e = $0.10/km (about $0.16 per mile). Similar results are obtained when e is varied over the range $0.02–$0.14/km (i.e. about $0.03–$0.23 per mile). It is assumed that the distance tax is an ideal instrument to internalise the emissions externality. Tolls are not ideal because they do not vary with distance travelled on the ring road, and also because the step and fine tolls vary over time, whereas emissions per unit distance are assumed not to. A distance tax therefore has a primary role to play in reducing emissions. A comparison of Table 1(b) and (c) confirms that the benefits from a distance tax are increased when emissions are present. With no tolling or partial tolling, the distance tax is raised by somewhat less than the unit cost of emissions because auto usage is reduced. With universal tolling the distance tax can be targeted exclusively at the emissions externality, and is therefore set equal to e. The distance tax reduces auto usage, and the benefits from tolling are reduced further relative to the base case. The reductions however are fairly small. Inclusion of environmental externalities therefore does not materially alter the economics of congestion pricing, at least in the example.

122

ANDRE´ DE PALMA ET AL.

5.9. Tolling Costs and Acceptability The analysis so far has disregarded several practical considerations relevant to whether tolling should be or could be implemented. Two of these considerations are briefly addressed here: implementation costs and acceptability. Implementation costs of tolling depend on the technology used, the size of the network, the area or number of links or routes to be tolled, traffic levels, and other factors. Collection costs at manual toll booths for example, amount to roughly $0.5 per vehicle passage. Experience with electronic tolling is still limited to a few cities. Rough calculations using data from Singapore, Hong Kong, and a German project for tolling of trucks on motorways suggest a cost of about $0.2 per vehicle passage. The benefits per driver from tolling reported in Table 1 are gross of tolling costs. If $0.2 per vehicle passage is taken as a representative figure, the benefit per driver net of costs from flat tolling of one route in the base-case (Table 1(a)) is barely positive; i.e. $0.23 − $0.20 = $0.03. And the net benefits from flat tolling of either one or both routes are negative if a distance tax is also implemented (Table 1(b) and (c)). In contrast to that, the net benefits from step tolling and fine tolling are more robust, as the benefits are substantially higher. The sensitivity analysis presented earlier in this section, however, shows that the benefits from tolling vary with the demand elasticity and other network factors, and for that reason time varying tolling may not always pass the cost-benefit test. There is now abundant evidence that support for road pricing hinges not so much on the size of the efficiency gains as on whether individual travellers/voters end up better off after accounting for how toll (and any distance tax) revenues are spent. Although the model is simple, assessing the distributional effects is complicated by the fact that travellers differ in where they live and in whether they drive or take transit. It is therefore not sufficient to consider the change in travel costs for a particular driver. Instead, attention will be focused on the efficiency gains or benefits per capita derived from tolling as a fraction of revenues generated. The larger this fraction is, the better the prospects that a way of using revenues for rebates, infrastructure investment, and so on will be found so that it leaves the majority of travellers better off. Table 2 presents the efficiency gains and revenue per capita for the base-case solution. With flat tolling the efficiency gains amount to about 20% of toll revenue. Thus, at least 80% of the revenues would have to be returned to travellers in order to leave them in aggregate as well off as without tolling. For step tolling the results are more encouraging. Indeed, with partial tolling of one route the efficiency gains nearly match revenues so that only a modest fraction of the revenues would have to be returned. This result is attributable to the fact that the base toll is low

Time-Varying Road Pricing and Choice of Toll Locations

123

Table 2. Efficiency Gains and Toll Revenues for Two Routes (Base-Case Parameter Values). Tolling Regime

T

None Flat

∅ 1 R 1 R 1 R

Step Fine

Efficiency Gain Per Capita [$]

Toll Revenues Per Capita [$]

Efficiency Gain/ Toll Revenues

0.17 0.75 0.82 1.83 1.65 3.04

0.87 3.46 0.89 3.36 0.85 3.04

0.20 0.22 0.93 0.54 1.96 1.00

Source: Own calculations (see text).

and most of the toll revenue is derived from time variation of the tolls, which reduce queuing while leaving private travel costs unchanged. Finally, the prospects are even more attractive with fine tolling because efficiency gains equal or exceed the toll revenue transfer. This suggests that the potential for gaining acceptability is enhanced considerably by time varying tolls.

6. FOUR ROUTES With four routes there are 15 possible (non-empty) subsets of routes to toll, and it is impractical to describe results in as much detail as for two routes. Attention is focused on the returns to scale from partial tolling, and on the relative merits of concentration and dispersal when two of the four routes can be tolled.

6.1. Base-Case Solution In the base-case no toll equilibrium with four routes the auto share of trips is about 0.77, and the generalised cost elasticity of auto usage is about −0.46. Both values are very similar to the ones obtained for the two route equilibrium. The relative benefits from tolling in the base-case are listed in Table 3(a). Like in the two route case, benefits are strongly increasing with flat tolling, weakly increasing with step tolling, and weakly decreasing with fine tolling. The choice between concentration and dispersal depends on the tolling regime. With flat and step tolling, concentration is preferred. T = {1, 2} for example, yields higher benefits than T = {1, 3}. With fine tolling, dispersal is preferred. To see why, suppose Route 1 is tolled first. With flat and step tolling, ␶1 > 0. This

ANDRE´ DE PALMA ET AL.

124

Table 3. Relative Benefits from Tolling, Four Routes. Number of Routes Tolled

1

T

1 2 or 4 3 1, 2 or 1, 4 1, 3 2, 3 or 3, 4 2, 4 1, 2, 3 or 1, 3, 4 1, 2, 4 2, 3, 4 R

2

3

4

(a) Base Case

(b) Route 1 Low Capacity

Flat Tolling g R T

Step Tolling g R T

Fine Tolling g R T

Flat Tolling g R T

Step Tolling g R T

Fine Tolling g R T

0.103 0.103 0.103 0.275 0.234 0.275 0.234 0.521 0.521 0.521 1

0.225 0.225 0.225 0.454 0.451 0.454 0.451 0.699 0.699 0.699 1

0.282 0.282 0.282 0.529 0.543 0.529 0.543 0.766 0.766 0.766 1

0.098 0.114 0.094 0.272 0.213 0.284 0.264 0.503 0.525 0.574 1

0.173 0.248 0.231 0.425 0.405 0.484 0.498 0.670 0.693 0.757 1

0.217 0.307 0.291 0.495 0.492 0.560 0.591 0.741 0.760 0.819 1

(c) Flat Tolling, Route 1 Drops Out Number of Routes Tolled 1

2

T

gR T

1 2 3 4 1, 2 1, 3 1, 4 2, 3 2, 4 3, 4

0.214 0.170 0.205 0.187 0.441 0.436 0.437 0.406 0.370 0.453

Source: Own calculations (see text).

exacerbates congestion on Routes 2 and 4 more than on Route 3, so that more is gained by tolling Routes 2 or 4 next. Tolling Routes 2 or 4 also reduces the number of “boundaries” between the tolled and untolled portions of the network, which cause adverse traffic reallocation effects. In contrast to that, with fine tolling ␶1 < 0. This reduces congestion more on Routes 2 and 4 than on Route 3, and Route 3 is the best choice to toll next.

6.2. Auto Usage Elasticity The price elasticity of auto travel with respect to generalised cost was varied by changing parameters C- S and  while keeping the auto share of trips in the NTE fixed at the base-case value, like in the two route case. The effects are similar to the two route case as well. The benefits from flat tolling are increasing, the

Time-Varying Road Pricing and Choice of Toll Locations

125

benefits from fine tolling decreasing, and the benefits from step tolling are first decreasing, and then increasing. In all regimes the relative benefit from tolling T routes approaches 0.25T as the elasticity increases. Concentration is preferable for flat tolling, and dispersal is preferable for fine tolling. For step tolling the choice between concentration and dispersal makes little difference. In all regimes, preferences shift slightly towards concentration as the elasticity increases. The reason for this is that when the elasticity increases the base toll with T = 1 increases as well, and this exacerbates spillover of traffic onto adjacent routes.

6.3. Route Capacities To illustrate the effects of capacity differences, suppose Route 1 has half the capacity of the other routes, while total capacity and other parameters remain unchanged. (Letting Route 1 have twice the capacity of other routes conveys similar lessons.) As Table 3(b) shows, with T = 1 it is preferable to toll one of the routes adjacent to Route 1 (i.e. 2 or 4) because they carry the most traffic in the NTE. With T = 2, location preferences are the same as in the base case, except that with step tolling the dispersed location choice T = {2, 4} becomes the best on account of the heavy traffic carried on Routes 2 and 4.

Fig. 6. Benefits from Flat Tolling as Fraction of Universal Tolling Benefits vs. Auto Elasticity (Four Routes, Route 1 Low Capacity; One Route Tolled).

ANDRE´ DE PALMA ET AL.

126

Fig. 7. Benefits from Flat Tolling as Fraction of Universal Tolling Benefits vs. Auto Elasticity (Four Routes, Route 1 low capacity; Two routes tolled).

It should be pointed out that these location preferences are not robust. Figures 6 and 7 show the effects of varying auto elasticity on the flat tolling regime. It can be seen from Fig. 6 that at low elasticities the best choice for T = 1 is Route 1 because, even though Route 1 carries the least traffic, it is the most congested route and demand reduction is not a major goal. It can be seen from Fig. 7 that the best combination for T = 2 is Routes 1 and 3. As the elasticity rises, and demand reduction becomes more important, Routes 2 or 4 become the best choice for T = 1. Tolling Route 3 becomes worthless at an elasticity of about −0.05, where the demand reduction and traffic reallocation incentives offset each other. As elasticity rises, the best choice with T = 2 switches first from {1, 3} (dispersal) to {1, 2} or {1, 4} (concentration), and then to {2, 3} or {3, 4}. This example illustrates that optimal toll location points can be sensitive to parameter values. It also suggests that general rules for choosing locations may prove difficult to define, and that in practical applications simulations will be necessary on each network in order to make good choices.

6.4. Route Spacing Asymmetry in the spacing between routes was introduced by rotating Routes 2 and 4 clockwise to obtain a configuration similar to that shown in Fig. 1. Reducing

Time-Varying Road Pricing and Choice of Toll Locations

127

the spacing between Routes 1 and 2 makes them better substitutes for drivers on average; and the same applies to Routes 3 and 4. Perhaps surprisingly, this has little effect on the relative merits of concentration and dispersal. More interesting results are obtained if route capacities are adjusted as well as spacing. Let L denote the circumference of the ring road, and S denote total capacity. Holding L and S fixed at their base-case values, parameters were set as follows: L 1 = L 3 = (1/8)L, L 2 = L 4 = (3/8)L, s 1 = (12/41)S, s 2 = (8/41)S, s 3 = (11/41)S and s 4 = (10/41)S. The relative benefits from flat tolling one or two routes on this network are reported in Table 3(c). For T = 1, Route 1 is the best choice because it has the largest capacity and draws the most traffic. For T = 2, Routes 3 and 4 are chosen because they have a larger combined capacity than the combined capacity of Routes 1 and 2. Route 1 is therefore dropped from the set of tolled routes when T increases from 1 to 2. The potential for “dropouts” to occur is significant for two reasons. First, it has implications for how to compute best toll locations on sizeable networks. One approach is to proceed iteratively by solving optimal locations for T = 1, 2, 3 and so on, while retaining all links included in the tolled set for T when solving the T + 1 problem. As Verhoef (2002) and Shepherd and Sumalee (2004) point out, this overlooks the possibility for dropouts as T increases. Indeed, both Shepherd and Sumalee (2004) and Zhang and Yang (2004) present examples in which dropouts happen. Although neither study comments on the dropouts, they appear to occur when individual links get replaced by other links located either upstream or downstream; i.e. in a series configuration. The possibility of dropouts also has implications for how road pricing should be phased in over time on a road network. Because of financial and other constraints, tolling is likely to be initiated on a few links, and extended gradually to other links on the network as funds and political/social acceptability permit. Given the appreciable equipment costs per tolling point, it is important to choose initial locations with a view towards whether they will remain appropriate at later dates. A myopic strategy for selecting locations on the basis of the best incremental gains on real road networks may not work well if the dropout phenomenon turns out to be prevalent.

6.5. Tolling Costs and Acceptability An analysis of tolling costs and acceptability can be undertaken for four routes in a similar way to that used in Section 5.9 for two routes. The results are similar to the two route case, and it would be redundant to report them here.

ANDRE´ DE PALMA ET AL.

128

7. CONCLUSIONS For several decades the literature on road pricing was focused on deriving optimal tolls for a predetermined set of links on a road network. Attention has recently turned to analysing how many links or points should be tolled, and where. This paper concentrates on three particular questions: Priority: Which links yield the largest benefits from tolling? Returns to scale: Are there constant, increasing or decreasing marginal benefits from tolling successively more links? Concentration: Should tolling be concentrated on one part of the network, or spread out? The answers to these questions naturally depend on how tolling affects traveller behaviour. Static models capture some dimensions of behaviour such as mode choice and route choice, but they do not account for trip timing decisions and are therefore not well-suited to describe the impacts of time-varying tolls. To overcome this limitation the model used here is based on Vickrey’s (1969) bottleneck model in which trip-timing decisions are fully endogenous. Amongst the main findings of the analysis are the following:  Priority: The benefits from tolling a link or route depend on a number of factors including the spacing between routes, route capacities, free-flow travel times and the elasticity of auto travel demand. None of these factors alone is a reliable indicator of what part of the network should be tolled first, and endogenous measures such as equilibrium traffic volume or congestion delay on links are not satisfactory by themselves either.  Returns to scale and concentration: The degree of time variation in the tolling regime has an important influence on the returns to scale, as well as on the relative merits of concentrating toll roads on part of the network and dispersing them over the network. With time-independent (flat) tolling the marginal benefits of tolling successively more links are generally increasing, and tolling of adjacent routes is preferred. This is consistent with results derived with static models (e.g. Verhoef et al., 1996). But with fine tolling, the findings with respect to returns to scale and concentration are just the opposite. Therefore, if time varying tolling becomes the norm, it may be optimal at least in the early stages of road pricing to toll a few heavily travelled and dispersed routes while leaving the rest of the network toll-free.  Dropouts: As the number of links that can be tolled increases, some links may be dropped from the best set as other links are added. A foresighted strategy is thus necessary when choosing the initial toll points at the beginning of a long-term program for phasing in road pricing.

Time-Varying Road Pricing and Choice of Toll Locations

129

 Optimal degree of time variation of tolls: For at least two reasons time-variation of tolls appears to be worthwhile. First, the efficiency gains are greatly enhanced. For the base-case scenario with two routes considered here, the efficiency gains from fine tolling are about four times higher than the gains from flat tolling if both routes are tolled, and nearly ten times higher when tolling is confined to one route. The costs of collecting tolls, meanwhile, tend to be relatively insensitive to the degree of time variation because infrastructure costs (as well as most operating costs) are fixed. A second advantage of time-varying tolling is that the efficiency gains can amount to a substantial fraction of the toll revenues collected. Consequently, private travel costs rise by considerably less than the amount of toll revenue, and the prospects of finding ways to use the revenues that leave travellers better off with tolling are correspondingly improved. The analysis in this paper could be extended in a number of directions. Perhaps the most important extension would be to admit a richer road network topology. The simple hub-and-spoke configuration used here does not include such features as multiple destinations or links in series, and it therefore precludes the full array of network characteristics that may influence the optimal number and locations of toll points. Tolling successive links in series will yield sharply diminishing returns if the same travellers use each link because tolls will serve as substitutes for each other. But on networks with multiple origins and destinations, the utilisation pattern of links will be more varied, and the returns to scale correspondingly different too. A second interesting extension would be to account for traveller heterogeneity more fully. In addition to differences in origins and destinations, and preferences for using public transport, travellers differ with respect to their costs of travel time, their trip-timing preferences, their inclination to car pool and so on. Allowing for heterogeneity in these characteristics will typically alter the efficiency gains derived from tolling. For example, if tolling is introduced selectively the tolled routes or toll lanes will be used primarily by carpools or individuals with high costs of time who value travel time savings greatly. Including additional dimensions of traveller heterogeneity will also enrich the scope for assessing the welfare-distributional effects of tolling which have received much recent attention in the literature.

NOTES 1. Kraus (1989) adopts a similar model with a monocentric city, radial commuting routes, and two modes. Kraus’ model is quite detailed spatially, but he assumes a peak travel period of fixed length and steady-state traffic conditions within the peak.

ANDRE´ DE PALMA ET AL.

130

2. To the extent that fuel consumption is insensitive to speed, and is the same on the ring road and arterials, a fuel tax could function approximately as a distance tax. 3. In addition, with inelastic travel demand charging negative tolls on roads is equivalent to charging positive tolls or fares of the same magnitude on transit if all roads are similarly tolled. However, negative tolls show up in the solutions only in cases where some roads remain untolled, so that this argument does not apply here. 4. The value of K r is based on a value of time of ␣ = $10/hour, a fixed access time from the inner endpoint of the arterial to the centre of 4.5 mins, and a travel speed on the arterial of 80 km/hour (about 48 mph). The value of k is obtained by assuming a travel speed on the ring road of 50 km/hour (about 30 mph) and a vehicle operating cost of $0.08/km (about $0.13 per mile). 5. In principle the cost of travel by transit will vary with R. However, as noted in the following two sections, both the auto share and the generalised cost elasticity of auto usage are nearly the same in the base-case equilibria with R = 2 and R = 4. 6. Changing either the radius of the ring road or the population density has little or no effect on the results of interest, and is therefore not discussed.

ACKNOWLEDGMENTS The authors are grateful to Marvin Kraus for helpful comments on an earlier draft, to Faisal Khan for able research assistance, and to Georgina Santos as editor for catching errors and substantially improving the text. Thanks are also due to the European Commission for partial financial support under Project MC-ICAM (Implementation of Marginal Cost Pricing in Transport – Integrated Conceptual and Applied Model Analysis). Robin Lindsey would also like to thank the Social Sciences and Humanities Research Council of Canada for support under the project “Road pricing in urban areas.”

REFERENCES Arnott, R., de Palma, A., & Lindsey, R. (1990). Economics of a bottleneck. Journal of Urban Economics, 27, 111–130. Arnott, R., de Palma, A., & Lindsey, R. (1993). A structural model of peak-period congestion: A traffic bottleneck with elastic demand. American Economic Review, 83, 161–179. Braid, R. M. (1996). Peak-load pricing of a transport facility with an unpriced substitute. Journal of Urban Economics, 40, 179–197. de Palma, A., & Lindsey, R. (2000). Private toll roads: Competition under various ownership regimes. Annals of Regional Science, 34, 13–35. European Conference of Ministers of Transport (1998). Efficient transport for Europe: Policies for internalisation of external costs. Paris: OECD. Kraus, M. (1989). The welfare gains from pricing road congestion using automatic vehicle identification and on-vehicle meters. Journal of Urban Economics, 25, 261–281.

Time-Varying Road Pricing and Choice of Toll Locations

131

May, A. D., Liu, R., Shepherd, S. P., & Sumalee, A. (2002). The impact of cordon design on the performance of road pricing schemes. Transport Policy, 9, 209–220. Mayeres, I. (2001). Equity and transport policy reform. Center for economic studies. Katholieke Universiteit Leuven, Working Paper 2001-14. www.econ.kuleuven.ac.be/ew/ academic/energmil/downloads/ete-wp01-14.pdf. Shepherd, S. P., & Sumalee, A. (2004). A genetic algorithm based approach to optimal toll level and location problems. Networks and Spatial Economics, 4, 161–179. Small, K. A. (1982). The scheduling of consumer activities: Work trips. American Economic Review, 72, 467–479. Verhoef, E. T. (2002). Second best congestion pricing in general networks: Heuristic algorithms for finding second-best optimal toll levels and toll points. Transportation Research B, 36, 707–729. Verhoef, E. T., Nijkamp, P., & Rietveld, P. (1996). Second-best congestion pricing. Journal of Urban Economics, 40, 279–302. Vickrey, W. S. (1969). Congestion theory and transport investment. American Economic Review (Papers and Proceedings), 59, 251–260. Zhang, X., & Yang, H. (2004). The optimal cordon-based network congestion pricing problem. Transportation Research B, 38., 517–537.

This Page Intentionally Left Blank

6.

ROAD PRICING AND PUBLIC TRANSPORT

Kenneth A. Small 1. INTRODUCTION More than three decades ago, Thomas Lisco (1970) argued that urban public transport in the United States suffers from poor quality or more precisely, a poorly chosen package of fare and level of service. His prescription was to raise fares and aim for a higher-class service that would appeal to increasingly affluent urban Americans. Events have gone rather differently. Demand has sometimes been found to be quite sensitive to fares, for example in the comprehensive study of Boston’s public transport system by G´omez-Ib´an˜ ez (1996). The main source of greater public transport funding has therefore been through sharp increases in federal and local subsidies. This seems to have caused little increase in public transport usage, not enough even to counteract other trends adverse to public transport. It has also caused some severe and widely documented inefficiencies, from high wages to uneconomical capital investments (Pickrell, 1985; Winston & Shirley, 1998). Meanwhile, other approaches to reducing urban congestion have led to frustration. Demand for driving under peak conditions has outstripped the financial and environmental resources that communities are willing to devote to road building. Carpooling incentives, telecommuting, and other measures have had negligible effects on motor-vehicle use: the actual modal shifts have been extremely small and any improvements in congestion have been overwhelmed by “latent demand”

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 133–158 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09006-7

133

134

KENNETH A. SMALL

for peak travel, i.e. demand that was previously deterred by congestion itself (Giuliano & Small, 1995; Small & G´omez-Ib´an˜ ez, 1999). Road pricing is therefore a welcome newcomer to the menu of congestion-relief policies under serious consideration and trial. However, a potential benefit of road pricing that is not widely appreciated is the dramatic effects it could have on public transport operations. One reason for this neglect is that a crucial link in the chain of effects is little known or understood outside of the community of professional transportation economists. Roughly, the entire chain can be described as follows:  Raising the monetary price of car travel induces some modal shifting to public transport;  Reduced congestion makes on-street public transport (bus or streetcar) faster and cheaper to operate;  Increased route coverage and/or service frequency to handle the demand further enhances the service quality as perceived by the user;  Higher costs of automobile commuting cause land near major business centres to become more valuable, hence to be developed at higher residential and commercial densities. This further enhances the market potential of public transport by increasing its density of demand in just those areas where it is already most efficient. It is the third of these effects that is so little understood. Mohring (1972) showed that the dependence of service quality on frequency and route coverage is a form of economies of scale. This means that any modal shift toward public transport touches off a “virtuous circle” of further cost savings and/or service improvements, hence possibly further modal shifts. Viton (1983) and Kain (1994) suggest that the resulting equilibrium might involve a far higher modal share for public transport than the one prevailing before road pricing was undertaken, even not accounting for any changes in land use. The purpose of this paper is to explore the potential quantitative importance of this “virtuous circle” in the short run, i.e. ignoring land use changes. It does so within the context of congestion pricing covering a major downtown area such as Singapore or London. The focus is on the likely impacts of road pricing on costs and service quality of urban bus transport, and on the second-round effects of these changes on the behaviour of public transport operators and potential users. The paper then considers the contribution of these impacts to the overall benefits from congestion pricing. The analysis is restricted to bus transport, partly because that is where the greatest benefits are likely and partly because the impacts of road pricing on underground service depend heavily on capacity constraints that are highly city and route specific.

Road Pricing and Public Transport

135

It turns out that a reasonably sophisticated model of bus transport yields remarkably simple rules of thumb for making these predictions. These rules take as given the direct effects of road tolls on public transport ridership and average road speed. Using some preliminary numbers from London, it appears that the favourable impacts on public bus transport, and the associated contribution to net benefits of road pricing, are indeed considerable.

2. THEORETICAL MODEL 2.1. Model Structure The model, adapted from Nash (1988) and Jansson (1997), both simplifies and extends the basic framework described by Mohring (1972). It simplifies Mohring’s paper by ignoring the time taken by passenger boarding and alighting and the operator’s decision as to where to place bus stops. It extends Mohring’s paper, as do Nash and Jansson, by incorporating the spacing of routes as an operator decision. There are other, more minor, differences as well. The model is partial equilibrium, and so does not include the effects of changes in transport prices on other distortions in the economy. Bus service takes place within a well-defined service area, which for convenience can be thought of as a rectangle. It has length  (km) and a width described by the walk time 2w required to traverse it. Within this area, the operator can choose to operate N evenly spaced parallel routes, each of length . (Equivalently, the service area can be a circle of radius , whose perimeter could be walked in time 2w, with evenly spaced radial routes converging at the centre and with walk access to those routes taking place along concentric circles.) The bus operator can adjust the spacing of routes by choosing N, and the average frequency of service by choosing vehicle-kilometres of service per hour, M. Passengers make trips at rate Q per hour, originating at locations evenly dispersed throughout the service area. The average trip length is m km. These and other symbols are summarised for convenience in Table 1. Conditions are assumed to be constant throughout the time period in which road pricing is in effect. Operator cost C is assumed proportional to the number of bus-hours of service, with the proportionality constant a linear function of bus size. Bus-hours of service are simply bus-kilometres of service M (which is controlled by the operator) divided by average bus speed S (which is not); this speed incorporates vehicle stops and passenger boarding and alighting time, so is less than the average speed of cars in the same locations.1 Bus size must be adjusted in order to achieve a predetermined load factor (average percentage of passenger spaces filled); hence

136

KENNETH A. SMALL

Table 1. Symbols Used in the Theoretical Model. Bus service environment (exogenous)  Length of service area and of each route (km) w Time to walk half the width of service area if rectangular, or half its perimeter if circular (hr) h Duration of period of road pricing (hr/day) m Average trip length (km) S Average bus speed (km/hr) Service and price variables chosen by bus operator N Number of routes M Hourly bus-kilometres of service offered (km/hr) f Fare ($ per trip) Bus ridership and finance Q Number of passenger-trips per hour n Average number of passengers on a bus = mQ/M R Revenue ($ per hour of service period) Costs and cost parameters C Cost to operator ($ per hour of service period) U Cost of user time to all users ($ per hour of service period)  Subsidy ($ per hour of service period) ATC Average total cost: (C + U)/Q ($/trip) a1 Operator cost per bus-hour, independent of bus size a2 Increase in operator cost per bus-hour caused by increasing bus size to accommodate one additional passenger on average Value of walking time ($/hr) ␣w ␣x Value of waiting time ($/hr) ␣v Value of in-vehicle time ($/hr) Demand parameters ␧f Demand elasticity with respect to fare ␧M Demand elasticity with respect to bus-km of service

it is proportional to the average number of passengers on a bus at any point in time, n. Therefore agency cost is of the form: C = (a 1 + a 2 n)

M S

(1)

where a1 and a2 are proportionality constants. All aggregate cost and service variables, such as C and M, are expressed per hour of a well-defined service period, which lasts h hours per day. This period may be taken to be that during which road pricing is in effect. Equation (1) is consistent with empirical evidence that bus-miles are supplied with few if any scale economies or diseconomies (Small, 1992, p. 57).

Road Pricing and Public Transport

137

The average number of passengers per bus, n, is simply the total passenger-km travelled per hour, mQ, divided by the bus-km of service offered per hour, M. Hence (1) can be rewritten: C=

a 2 mQ a1M + ≡ C1 + C2 S S

(2)

User cost U arises from three kinds of time spent. Walk time for the average user is equal to half the maximum walk time from a home to a bus line, that distance being w/N. Waiting time for the average user is half the headway between buses.2 Headway is defined as the time interval between buses, which is route-km, N, divided by veh-km per hour, M. Finally, average in-vehicle time is equal to m/S, the average trip length divided by speed. The user’s perceived opportunity costs for these three types of time (in $/hour) are ␣w , ␣x , and ␣v , respectively. Therefore aggregate user costs are:  

␣ m  1 ␣x N 1 ␣w w  v Q+ Q U= Q+ 2 N 2 M S (3) ≡ Uw + Ux + Uv ≡ Uo + Uv Here the subscripts w, x, and v stand for “walk,” “transfer,” and “in-vehicle.” In the last equality in (3), U is divided into out-of-vehicle and in-vehicle cost. Ignoring boarding and alighting times causes the model to overstate scale economies somewhat and therefore to understate the incremental agency and user costs imposed by additional riders. This should not be too serious because of two other features of the model: the assumption of constant load factor, and the use of an assumed bus speed (and change in bus speed) that does incorporate those times along with all other reasons for buses being slower than surrounding traffic. As shall be seen, the model predicts that only one-third of any ridership increase is handled through increasing the number of passengers per bus, and therefore the effect of boarding and alighting on bus speed should be modest.

2.2. Optimal Operator Decisions and Cost Functions Amazingly, this is all that is needed to compute the optimal responses, in percentage terms, of service levels, operating costs, and user costs to specified percentage increases in speed S and aggregate ridership Q. This can be done by simply letting the operator minimise the total cost of serving Q passengers. As is shown later, the operating agency will follow such a procedure if it is maximising aggregate consumer surplus less aggregate total cost, even if subject to a specified total

138

KENNETH A. SMALL

subsidy. The following derivation is similar to that of Jansson (1997) except that he allows for boarding and alighting time. The quantity C + U is therefore minimised with respect to the service variables N and M. This yields two first-order conditions:

a  1  ␣ NQ  1 x =0 (4) − S 2 M2     1 ␣x Q 1 ␣w wQ + − =0 (5) 2 2 M N2 Solving (4) for M yields the well-known “square-root rule” for adjusting optimal service offerings to changes in demand, when route structure is fixed (Mohring, 1972):   ␣x SN 1/2 1/2 Q (6) M= 2a 1 However, route structure is not fixed. Solving (5) for optimal route spacing yields: 1/2  ␣w w ×M (7) N= ␣x  Solving (6) and (7) for N and M yields a result different from the square-root rule:  1/3 S Q 1/3 (8) N ∗ = (␣w w)2/3 2a 1 ␣x    S 2/3 2/3 Q (9) M ∗ = (␣x ␣w w)1/3 2a 1 Optimal service provision M∗ is seen to vary with output to the two-thirds power, not the one-half power; the reason for the difference is that when N can also be optimised, additional bus-kilometres can be used to reduce both walking and waiting time and thus are more beneficial than when N is held fixed. As shall be seen shortly, this is done at the expense of operating cost savings, average operating cost is still reduced as Q expands but less rapidly than when route structure is fixed. It should also be noted that bus size, which is proportional to n ≡ mQ/M, varies with Q1/3 ; thus two-thirds of a ridership increase is handled via additional buses and only one-third via larger buses. The total cost function can be computed by substituting these solutions for optimal N and M into (2) and (3). Dividing each component by output yields the following expressions for the five average cost components corresponding to the

Road Pricing and Public Transport

139

five separate additive terms in (2) and (3): a2m C2 = Q S ␣v m Uv = Q S

␣ 2/3 a 1/3  Q −1/3 Uw Ux C1 0 1 = = = Q Q Q 2 S w

(10)

where ␣0 ≡ (␣w ␣x )1/2 is the geometric mean of the values of walking and waiting time. Thus the first two components of average cost are independent of Q, indicating constant returns for these components. The other three components are equated to each other at the optimum; this generalises the result of Mohring (1972) that, under suitable simplifications, average operator cost is equated to average waiting cost. Furthermore, these latter three average cost components decline with passenger density, indicating scale economies for them. Average agency cost (C 1 + C 2 )/Q, also declines with Q, consistent with statistical evidence for scale economies in final output (Small, 1992, p. 57). Combining all five cost components, the average total cost is: m (11) ATC = (a 2 + αv ) + 3c 1 Q −1/3 S where c 1 = (␣0 /2)2/3 (a 1 w/S)1/3 .

2.3. Revenues, Subsidies, and Constrained Welfare Maximisation It is shown in this section why the agency might choose to minimise total cost even when operating under a fixed subsidy. Consider the broader problem of choosing service levels M and N and fare f in order to maximise social welfare, subject to a budget constraint involving a fixed aggregate subsidy . The inverse demand curve for trips, ␲(Q), is defined as the full price, including fare plus average user cost, at which ridership would be Q. Let V(Q) be some measure of the value of public transport travel to consumers, such as the area under ␲(Q). Subtracting total cost yields the following expressions for social welfare: W = V(Q) − C − U Revenues are equal to

   U Q = ␲(Q)Q − U R = f Q = ␲(Q) − Q

(12)

140

KENNETH A. SMALL

The agency’s budget constraint, assumed for simplicity to be binding, is C − R =  or C + U − ␲(Q)Q = 

(13)

The quantity C + U occurs as a unit in both (12) and (13), and furthermore the other terms in (12) and (13) do not contain N or M. Thus when a Lagrangian function is defined for this constrained maximisation problem, with multiplier ␭ on the constraint, the Lagrangian function divides into two parts; one part, −(1 + λ)(C + U), contains all the dependence on M and N. The optimality conditions for M and N are therefore identical to those derived previously for minimising C + U. Of course, the actual amount of service will be smaller if ridership Q is below the optimal level, but that amount will still minimise total cost given Q. A similar derivation shows that a profit-maximising private operator, also, will choose service characteristics to minimise total cost for given ridership level. Maximising revenue less agency cost means maximising fQ − C = ␲(Q)Q − (C + U), which requires minimising (C + U) for given Q. This result, however, depends on the assumed uniformity of values of time; if such values differ across the population, a monopolist private operator would provide service levels weighted more heavily by the desires of those market segments with the highest demand elasticities.3 What if public transport providers have other objectives and so do not choose service levels to minimise the sum of agency and user costs? Several studies in the U.S., Australia, and the U.K., including London, have found that too much service is provided, even when accounting for the benefits to road users of enticing people from private vehicles.4 In such a case, the model here does not apply precisely. Yet most of the benefits identified by the model are still measured accurately to first order in patronage changes. For example, if, for a given ridership, service levels are inefficiently high, then the ridership increase caused directly by pricing the competing auto mode can be handled at zero marginal cost (again abstracting from delays due to boarding and alighting of additional passengers). This produces a favourable fiscal effect on the operator, which is captured by the model here. What is not captured is the possibility that the cost savings might be dissipated in inefficient further service increases.

3. IMPLICATIONS OF THE MODEL FOR ROAD PRICING In this section, we consider the results of small changes in bus operating speed dS and in ridership dQ(1) that are directly caused by the reduction of automobile

Road Pricing and Public Transport

141

trips resulting from road pricing. Some of these trips may be suppressed entirely or diverted to time periods outside the analysis; it does not matter for the present analysis because it is assumed that dS and dQ(1) are determined by some other model. It is also assumed that the adjustments in the public transport system described by the model here have only negligible effects on traffic speeds, so that dS captures all the needed information. On the other hand, dQ(1) is just the first-round effect of the fare increase, and does not include subsequent changes in ridership calculated here. These assumptions are most appropriate when bus transport constitutes a small part of total trip-making and contributes only a small proportion of the traffic causing congestion. In other cases, of which London is surely one, they can be regarded as providing only a first approximation to the situation. A more complete model would need to describe the effect of public transport cost and quality on automobile traffic and the relationship between automobile traffic and travel speed. The model of the previous section can easily be used to describe the results of finite as well as infinitesimal changes in S and Q. However, the results of doing so are less illuminating. Furthermore it is not clear that the improved accuracy is worth the trouble, given the model’s aggregated and simplified nature.

3.1. Direct Effects of Improved Speed on Costs Even without any adjustments by the operator, the improved speeds brought about by road pricing have favourable effects on both the agency and users, due to reduction in operating costs and in-vehicle time. To put it differently, speed S affects both agency cost C and user cost U directly, even in the absence of any changes in N, M, or Q. It can be easily seen from (2) and (3) that the affected components are C1 , C2 , and U v , and that each responds to small changes in S with elasticity – 1. If it is further assumed that the savings in agency cost are passed on through lower fares, then the budget constraint f = (C − )/Q implies that the fare responds with elasticity equal to that of cost multiplied by C/(fQ) ≡ C/R. These results can be summarised as follows: ␩C,S = −1;

␩U v ,S = −1;

␩f,S = −

C R

(14)

It should be noted that the favourable effect on bus in-vehicle time is matched by a similar effect on automobile in-vehicle time; thus rising speed does not necessarily create a net inducement for modal shift. Rather, the effect described here counteracts what otherwise would be a positive effect of improved speed on automobile usage. Since the latter effect is normally already included in the full price

142

KENNETH A. SMALL

of automobile travel, it is important to recognise this competing improvement in the quality of bus service.

3.2. Direct Effects of Increased Bus Ridership on Costs The ridership boost dQ(1) also affects cost, through the scale economies described earlier. These effects could be calculated either with or without optimal adjustments by the operator; either method would give the same total change in cost if the initial operation were optimised (perhaps subject to a subsidy constraint). Here, the effects are calculated assuming optimal operator adjustment. Equation (9) shows that optimal bus service provision M∗ is adjusted proportionally to Q2/3 , i.e. the elasticity of M∗ with respect to Q is 2/3. Equation (10) shows (after multiplying by Q) that agency costs C1 and C2 respond to Q with elasticities 2/3 and 1, respectively. Therefore combined agency cost C responds with elasticity ␩C,Q = (2/3)(C 1 /C) + 1(C 2 /C) = (2/3) + (C 2 /3C). Equation (10) also shows that user costs Uo and Uv respond to Q with elasticities 2/3 and 1, respectively (recalling U o = U w + U x ). This can be summarised as follows: ␩M,Q =

2 ; 3

␩C,Q =

C2 2 + ; 3 3C

␩U o ,Q =

2 ; 3

␩U v ,Q = 1

(15)

3.3. Effects on Agency Budget Balance If fare were kept constant, then the bus subsidy could be reduced by an amount dR − dC, where dR/R = dQ (1) /Q and dC/C is determined by the elasticities of C in Eqs (14) and (15); combining these results,   dS 2 C2 R dQ (1) dR − dC = − + − (16) C S 3 3C C Q It can be seen that there are counteracting effects on the budget balance. Improved speed reduces total cost (first term), while increased ridership increases total cost but adds new revenue (second term). The overall balance is positive provided that dS/S C2 R 2 − . > + (1) 3 3C C dQ /Q This condition is very likely to be met. Size-related cost C2 is likely to be no greater than non-size-related cost (largely driver time), so that C2 /(3C) ≤ 1/6; and the cost-recovery ratio is likely to be not too much smaller than 1/3. Therefore, the condition requires only a quite modest speed improvement, around half the

Road Pricing and Public Transport

143

ridership increase, and it could even be met with no speed improvement if the cost-recovery ratio is more than 0.83. Alternatively, the argument behind (16) can be recast as the amount of fare reduction made possible if the total subsidy is unchanged. Again writing fare as (C − )/Q and using (14) and (15), the following expression is obtained:     dC C dQ (1) C C2 R dQ (1) 2 dS df (1) = − = + + − − . (17) f C R Q R S 3 3C C Q Henceforth it is assumed that budget changes are passed through to passengers, i.e. that (17) applies.

3.4. Second-Round Effects on Ridership Both the fare reduction computed in (17) and the service improvement computed in (15) will create a “second-round” increase in bus ridership, dQ(2) . It is convenient to write this in terms of the elasticities of bus ridership with respect to fare (␧f ) and with respect to vehicle-km of service (␧M ). Using (15) to compute change in M yields: 2 dQ (1) dQ (2) df (1) = ␧f + ␧M Q f 3 Q where df (1) /f is given by (17). Substituting (17) and grouping terms yields dQ (2) dQ (1) C dS = −␧f +␮ Q R S Q where ␮=

2 C ␧M + 3 R



2 C2 R + − 3 3C C

(18)

 ␧f

(19)

is a factor leading to a multiplier on ridership changes. It should be noted that the first term of (19) is positive, while the second term could be positive or negative but, as argued before, is probably positive. Subsequent rounds of fare and service changes result in further ridership changes following the iterative formula dQ (n) = ␮dQ (n−1) for n ≥ 3. Combining them, the ultimate change in ridership resulting from the direct changes in speed dS and ridership dQ(1) is   dQ SQ 1 dQ (1) C dS = + −␧f >0 (20) Q 1−␮ R S Q

144

KENNETH A. SMALL

provided that |␮| < 1. The corresponding fare and service changes are obtained analogously from (17) and the first of Eq. (15):     C C2 R dQ SQ 2 df SQ dS = + + − − (21) f R S 3 3C C Q and dM SQ 2 dQ SQ = M 3 Q

(22)

A more elegant way of deriving (20) is in fact to simultaneously solve (21), (22), and the equation expressing the sources of change in ridership.5 A crucial simplifying assumption made here is that the feedback effects of these second- and subsequent-round ridership increases on traffic speed can be ignored. Accounting for such changes would further magnify the favourable results found in the simulations presented later.

3.5. Effects of Using Road Pricing Revenues to Increase Bus Service In some cases, such as London in 2003, a road pricing policy may include a provision to use all or part of the revenues for improved public transport service. Suppose, then, that the bus transport operator receives an additional exogenous increment in its subsidy, d, designated for increased service M. Solving (2) for M, substituting  − R for C, and varying  while holding Q, S, and R constant gives an elasticity of bus-kilometres M with respect to subsidy  of ␩M, = /C 1 . (It is larger than /C because increasing service with Q constant permits use of smaller buses, which stretches the subsidy still further.) This induces a first-round increase in ridership governed by elasticity ␧M , and subsequent rounds governed by the multiplier ␮ as before. Combining all these rounds, the ultimate service and ridership increase arising from the increased subsidy are: 1 d dM  = M 1 − ␮ C1

(23)

␧M d dQ  = Q 1 − ␮ C1

(24)

and

Equations (20) and (24) are additive if all three first-round effects (i.e. the changes in speed, ridership, and subsidy) occur simultaneously, as for example is the case in London.

Road Pricing and Public Transport

145

4. BENEFITS FROM ROAD PRICING In this section, the model is used to assess the benefits from road pricing that occur due to the impacts on the public transport sector. That is, the results indicate what additional benefits occur beyond those computed just from the analysis of private vehicles alone.6 First, consider the benefits dBS from improved speed, dS. They are simply the savings in agency costs (assumed here, for simplicity, not to be used to decrease fares) and in user costs. Using the elasticities shown in Eq. (14), these benefits, to a first-order approximation in speed change, are:   dS S (25) dB = (C + U v ) S Next, consider the benefits dBQ from dQSQ , i.e. from the increase in bus ridership excluding that resulting from subsidy d. There are three components. First is increased revenues to the bus operating agency. To understand this component, it should be noted that some of the increase in ridership comes from mode switches from private vehicles. The lost consumer surplus to mode switchers, already included as part of the assessment of the costs and benefits to private vehicles, takes as given the full price for bus encountered by those switchers. Part of that full price is average user cost on the bus mode for the switchers, which is a real cost and so is already accounted for. Another part, however, is the bus fare paid, which is a transfer to the public transport operating agency and therefore is a benefit in the public transport sector. Given constant fare, this component of benefits may be written as:   dQ SQ dR = R Q The second component consists of the negative of the extra costs dC incurred by the bus operator to handle this increased ridership. From Eq. (15) and using C ≡ C 1 + C 2 , it is      SQ  dQ SQ dQ 1 −dC = −␩C,Q C = −C + C 1 Q 3 Q It is worth noting that if the optimised marginal agency cost, [C − (1/3)C1 ]/Q, exceeds the fare, then the sum of the first two components of benefits, dR − dC, is negative; this would indicate that the bus agency’s finances are harmed by the additional ridership.7

146

KENNETH A. SMALL

The third component of benefits consists of cost savings to existing users from improved service. This may be written, again using (15), as     dQ SQ U = −dU + U −Qd Q Q     dQ SQ dQ SQ = −(␩U o ,Q U o + ␩U v ,Q U v ) + (U o + U v ) Q Q     SQ   dQ SQ 2 2 dQ = C1 = − U o − 1U v + (U o + U v ) 3 Q 3 Q where the last equality follows because, from (10), U o ≡ U w + U x = 2C 1 . Combining the three components of dBQ and simplifying, it is found that     SQ dQ U = (R − C 2 ) (26) dB Q = dR − dC − Qd Q Q Equation (26) has a simple interpretation. Assume that the public transport operator was already choosing service level optimally, subject to a given fare. The envelope theorem then implies that the change in total costs from additional passengers dQSQ can be computed by assuming they are handled with no change in bus schedules or routes. In that case, the agency accommodates passengers just by increasing bus size. Agency profit then changes by the increased revenue (proportional to R) less the cost of increased bus size (proportional to C2 ). Under the assumptions made above, the welfare effects outside the agency are already captured in the analysis of car traffic (in the case of dQ(1) ) or may be assumed zero (in the case of dQ(2) , dQ(3) , etc.) provided the subsequent-round increases in ridership are drawn from activities in which price equals marginal social cost.8 Thus the increase in bus ridership may be treated, to first order in small changes, as a purely fiscal matter and so the theory can accommodate either positive or negative welfare effects from increasing bus ridership. Combining the benefits from speed and ridership increases, writing dBSQ for the total, and expressing the result as a fraction of agency cost gives     R dB SQ U v dS C 2 dQ SQ = 1+ + − (27) C C S C C Q Remarkably, only one new parameter, U v /C, is required to compute Eq. (27). As a very crude approximation, U v /C could be expected to be close to the average value of user time, multiplied by number of passengers per bus, divided by the

Road Pricing and Public Transport

147

average driver wage rate (inflated for breaks and other non-productive time); hence it is greater than one under most circumstances, and could be quite large when each bus carries many passengers. There is no comparable benefit from the use of road-toll revenues for public transport subsidies, because the revenues are already accounted for in the standard analysis of road pricing’s effects on road users. That analysis assumes revenues are used efficiently; to first order in the changes, prior optimisation implies that this assumption can be met by spending the road-toll revenues on service improvements. Here the assumption that existing service levels are efficient is obviously critical.

5. NUMERICAL EXAMPLE: LONDON It is interesting to calculate the consequences of road pricing for bus transport using the model described here and some realistic parameters. In this section, results are first computed using parameters representative of London’s dramatic implementation of road pricing in February 2003. They are then compared with results from some alternative sets of parameters likely to be more representative of other bus systems.

5.1. Background In the London road pricing scheme, all vehicles travelling within central London on weekdays between 7 AM and 6.30 PM are charged £5, or approximately US$8, once for the entire day. Taxis and motorcycles are exempt, and residents of the central area receive a large discount. Net revenues are targeted to improvements in public transport. In addition, prior to implementation, a major program of improvements to bus transport was undertaken.9

5.2. Parameters for the Numerical Example Table 2 presents the assumed parameters; justification is given in the Appendix. Available fiscal data refer to the entire London bus system, rather than just the central area; but since all parameters are ratios, this should be reasonably accurate. London is unusual in two respects, leading to the consideration of some alternative scenarios as well. First, it has a high cost-recovery ratio, R/C, estimated here to be 0.80. An alternative scenario takes this ratio to be 0.32.

148

KENNETH A. SMALL

Table 2. Parameters for Numerical Exercise. R/C C2 /C Uv /C ␧f ␧M

0.80 0.45 2.90 −0.25 0.30

dS/S dQ(1) /Q d/C

0.09 0.06 0.07

Source: See Appendix.

Second, the direct modal diversion dQ(1) /Q is rather small, estimated here at 6%. The reason is that public transport usage in central London is already very high, with only 12% of people entering the area during the morning peak travelling by car, according to Transport for London’s monitoring study (Transport for London, 2003c, p. 130, Fig. 1). Thus the impact on bus is modest, even though the system was designed so that bus would absorb proportionally much more of the mode shift than the underground.10 To represent cities with more typical initial mode share for bus transport, an alternative scenario is also considered in which dQ(1) /Q is five times as large. As shall be seen, the model predicts that the incentive of higher speeds, fare reductions, and service improvements brought about by the program will cause a much greater increase in ridership than this direct modal diversion. It appears, in fact, that Transport for London anticipated this outcome by putting in place a major program of bus service improvements, leading to a 10% ridership increase, before the start of congestion charging. Perhaps these improvements would have been undertaken regardless of the congestion charging, but it is equally reasonable to view them as part of a comprehensive package aimed at making congestion charging successful. The prior ridership increase therefore may be regarded, at least in part, as reflecting some of the later-round ridership increases dQ SQ + dQ  − dQ (1) predicted by the model, and the cost of providing those prior service increases may be regarded as being covered in part by the cost savings and revenue increases estimated in the model. (As it happens, the model does predict these later-round ridership increases to be almost exactly 10%.)

5.3. Results Table 3 presents results of the base case, as well as four alternative scenarios: the two mentioned in Section 5.2, a third in which none of the road pricing revenues are spent on public transport, and a fourth in which all three changes are made simultaneously.

Road Pricing and Public Transport

149

Table 3. Results of London Numerical Example. Change in

As Fraction of

Base Case

Bus-km of service

Bus-km of service Fare Ridership

0.231

Fare Ridership Average cost In-vehicle time cost Out-of-vehicle time cost Total average user cost Average agency cost Benefits From speed change From ridership increase Total benefits

R/C × 0.4 0.180

dQ(1) /Q d = 0 All Three ×5 Changes 0.430

0.073

0.208

−0.110 −0.276 −0.104 0.157 0.142 0.455

−0.110 −0.265 0.109 0.312

−0.326 −0.159 −0.485 −0.049

−0.326 −0.050 −0.376 −0.110

−0.816 −0.310 −1.125 −0.069

−0.326 −0.295 −0.622 −0.104

Agency cost Agency cost

0.351 0.351 0.038 −0.014

0.351 0.143

0.351 0.351 0.038 −0.041

Agency cost

0.389

0.494

0.389

Fare Fare Fare Avg agency cost

0.337

−0.816 −0.357 −1.173 −0.147

0.310

Source: Generated by author; see text.

Several results stand out. First, the increase in bus speed is very important. The reduction in user in-vehicle time is equivalent, from the user’s point of view, to a 33% reduction in the fare (base case, row 4). This in turn is responsible for the expansion of the assumed direct ridership increase (6%) into the 11% increase that would occur even without the increased subsidy (column labelled d = 0, row 3). Second, the cost savings to the agency from higher bus speed are a very significant offset to the extra costs incurred as ridership expands. In the absence of a change in subsidy (column d = 0), average agency cost falls by 11%, which happens to just offset the ridership increase. Holding aggregate subsidy constant, this enables the fare to be reduced by 11% (column labelled d = 0, row 2). Third, the complete package as envisioned in London produces a ridership increase much greater than that assumed from direct modal diversion. In the base case, the 6% direct increase becomes a 16% total increase; of the 10 percentage-point difference, half is due to the speed change and passed-through fare reductions, while the other half is due to the additional service funded by the road pricing revenues.11 Fourth, the net benefits are a significant fraction of initial aggregate agency costs (39% in the base case). Most of these net benefits arise from the speed change. They exclude any advantages from spending road pricing revenues because, as noted earlier, those revenues are assumed to be already accounted for in the evaluation of the road pricing scheme’s effects on private vehicles.

150

KENNETH A. SMALL

Fifth, the service improvements are quite substantial, amounting to a 23% increase in bus-km of service in the base case and 7% even if the road pricing revenues were spent elsewhere. These improvements have a favourable effect on average out-of-vehicle time costs (walking and waiting), which decline by 16% of the initial fare in the base case and 5% when subsidies are kept constant. The combination of in- and out-of-vehicle time changes is equivalent to a 48% fare reduction in the base case. Some of these results are even larger when the cost-recovery ratio is lower (second column), due mainly to the initial public transport revenue base being smaller. The reduction in average user cost is now 112% of the fare, which from the user’s perspective adds to the fare reduction itself of 28%. With fare so far below average cost, the ridership increase is actually a net disbenefit; yet it is a very small one because the extra agency cost incurred is mostly offset by cost savings to users. Meanwhile the benefits from increased bus speed remain enormous. Many results are larger when the direct modal diversion is larger (third column). In that case, service increases by 43%, ridership by 45%, average user cost declines by 62% of the fare, and total benefits are 49% of agency costs. Recent research has disclosed that the benefits of congestion pricing may be much reduced by adverse effects on pre-existing tax distortions when generalequilibrium effects are included, unless the revenues are used to reduce other taxes (Parry & Bento, 2001; Van Dender, 2003). The main reason is that the cost of living is raised, hence the real wage is lowered, adding to the distortion from taxes on labour income. Such effects, not incorporated in this model, would reduce the direct benefits of congestion pricing and magnify the benefits through public transport. They also constitute an argument for considering the third alternative scenario, in which road-pricing revenues are not spent on public transport. As already noted, the values for cost-recovery ratio and modal diversion shown in the alternative scenarios are more typical of bus transport in many large cities than are the base-case values. Arguably, then, the most useful scenario to consider for potential sites outside London is that shown in the last column, where all three alternative scenarios apply simultaneously. These results portray a system of public transport that is greatly transformed. Service is up by 21%, ridership by 31%, fares are down by 27%, average agency cost is reduced by 15%, and average user costs are reduced by 117% of the fare. Net benefits are nearly one-third of agency costs.

6. CONCLUSIONS Road pricing in a large city can dramatically change the role of public transport, at least of those modes that share the streets with private vehicles. Even without

Road Pricing and Public Transport

151

spending any of the road pricing revenues on public transport improvements, the combination of increased traffic speed and increased ridership permits large increases in service and ridership, reductions in user costs, and savings in average agency costs sufficient to pay for the increased service even while reducing fares. Net benefits may be very large, on the order of 30–40% of agency costs, if traffic speed can be increased by just 9%. These basic results can be derived from a reasonably simple aggregate model requiring knowledge of remarkably few parameters. Two primary mechanisms drive the results: cost savings to both agencies and users from faster traffic, and optimisation of service levels in response to increased ridership. Of course, results would be quite different if pricing were applied to radial expressways with unpriced parallel arterials, as is being considered elsewhere in the U.K. That case could be quite unfavourable for public transport if the bus routes become more congested as a result of diverted car traffic. A number of model limitations are worth bearing in mind. First, in some cases the shift in demand to public transport could require expensive capital investments, specially to rail transport. Second, the implied targeting of public transport service toward places where ridership is greater may upset the precarious balance of power between suburban and inner-city communities that sometimes holds together regional public transport systems in the U.S. (This could, however, turn out to be a blessing in disguise because these regional systems have diverted a lot of subsidies to locations where public transport service is not economical.) Third, operators may not adjust optimally to changes and may waste subsidy funds. Finally, it would be desirable to examine the public transport system at a disaggregated level in order to describe in greater detail what an optimal response would look like. The welfare benefits arising from the public transport sector should, in principle, affect the design of road pricing itself. The direction of this effect is uncertain: while the effects on public transport reduce the level of toll required to achieve a given reduction in congestion (Kain, 1994, p. 531), they also increase the benefits of that reduction. Attempts to take public transport into account in road pricing optimisation include Parry and Bento (2002) and Van Dender (2003). The model in this paper is somewhat richer than theirs in the way it predicts adaptation by the public transport agency to changes in its environment. There is certainly room for research which combines the model features considered here with the general-equilibrium effects considered by those authors. An important extension would be to consider two additional competing modes of urban travel: shared-ride taxi and single-destination shuttle services. Both have scale economies similar to those described here for public transport (Schroeter, 1983; Yang & Wong, 1998). Shared-ride taxi service may be a cheaper

152

KENNETH A. SMALL

alternative to conventional public transport in low-density areas, and thus may be part of a comprehensive proposal to enhance the financial viability of public transport. Shuttle services have been successful at carrying passengers to and from airports and could probably be expanded to other destinations. Both will be affected significantly by road pricing policies in much the same way as public transport.

NOTES 1. Kain (1994) gives evidence that bus operating speed is about 60% of automobile speed in U.S. cities. 2. This assumption involves two biases in opposite directions. If headways are large, users will learn the schedule in order to shorten their waiting times. On the other hand, if they are trying to coordinate their trip with some other scheduled events, as is common, they incur costs of schedule mismatch, which are omitted here and which probably grow more than linearly with headway. 3. For an analogous argument in the context of pricing by a monopolist road operator, see Edelson (1971). 4. See Glaister (1987) for the U.K., Dodgson (1986) for Australia, Winston and Shirley (1998) for a U.S. aggregate analysis, G´omez-Ib´an˜ ez (1996) for Boston, Savage and Schupp (1997) for Chicago. Glaister’s estimates for the London bus system suggest that an optimal allocation of the agency’s budget at that time would involve a 28% reduction in fare and 31% reduction in service levels (Savage & Schupp, 1997, Table 1). In contrast, Jansson (1980) and Larsen (1996) find that in the cases they study, more frequent service should be provided but with smaller vehicles, at higher fares if necessary to maintain budget balance. 5. That equation is dQSQ /Q = dQ(1) /Q + ␧f df SQ /f + ␧M dMSQ /M. The three equations are solved for the three unknowns dQSQ /Q, df SQ /f, and dMSQ /M. 6. The benefits from private vehicles alone consist of the cost savings for continuing users less the lost consumer surplus for prior users. Equivalently, they consist of toll revenues plus changes in consumer surplus (negative in most cases) for all original private vehicle users. 7. The optimised marginal agency cost can alternatively be computed from (10) as: d(C 1 + C 2 ) 2 C1 C2 1 = + = dQ 3 Q Q Q



 1 C − C1 . 3

8. The quantity changes are assumed to be small enough so that each new bus rider is marginal, i.e. he or she can be assumed to be indifferent between riding the bus and doing whatever he or she did previously. 9. For further information see the Transport for London web site on congestion charging: www.tfl.gov.uk/tfl/cc intro.shtml. 10. The underground was already operating very close to capacity. Part of the strategy, encouraged by public transport pricing policies, was to shift many existing shorter trips from rail to bus, so that rail could accommodate some of the longer trips that shifted to public transport due to road pricing. The author is indebted to Chris Nash for this observation.

Road Pricing and Public Transport

153

11. The first half, 4.9 percentage points, is the difference between the fourth entry in row 3 and dQ(1) /Q. The other half, 4.8 percentage points, is the difference between the first and fourth entries in row 3. 12. Bus speed in 2002 for the central area is reported to be 11 km/hr in TfL (2003c, p. 107). This number is being inflated here by 20%, which is the ratio of traffic speeds in Inner London to Central London (averaged over morning peak, daytime between peaks, and evening peak), from TfL (2001, p. 23, Table 12). 13. A similar result is reported by Nash (1988, Table 5.1), who compiles figures from Glaister (1986) for the parameters of the linear cost function (1); they imply that if an 86passenger bus is taken as representative of the current situation, C2 /C ≈ 0.50. (Glaister’s figures are based on cost per bus-km, not per bus-hr; this is appropriate here because any speed differential between a small and large bus is part of the cost advantage of the former.) Jansson (1980, p. 68) implies that C2 /C = 0.24 − 0.30. By contrast, Mohring (1983) calculates a ratio for Minneapolis that implies C2 /C ≈ 1.0; much of the cost savings from smaller buses in that calculation, however, are due to assuming non-union wages for minibus operators, which should not enter the calculation here. 14. It is interesting to check the equality of non-size-related agency cost, C1 , and waiting time costs, Ux , according to Eq. (10); if the theory applies, this should give the same result for (C2 /C) as the calculation in the text. The equality can be written as 1 − (C2 /C) = Ux /C = (Ux /Uv )·(Uv /C) = (␣x /␣v )·(Tx /Tv )·(Uv /C), where Tx and Tv are the average waiting and in-vehicle times, respectively. The ratio ␣x /␣v is estimated at 1.6 for Britain by Wardman (2001). Uv /C = 2.90 from the previous bullet. Ideally the ratio Tx /Tv should apply to the central area. Tv can be estimated as m/S =19 min, since m = 3.48 km (calculated from TfL, 2001, p. 37, Table 20a) and S = 11 km/hr (TfL, 2003c, p. 107). Thus the equality holds if T x = 2.25 min. This may be compared with than the actual average waiting time of 4.1 min reported by users for all public transport in the central area, including underground (TfL, 2001, p. 17, Table 8a). 15. See Goodwin (1992), Pratt et al. (2000, pp. 9–14), Savage and Schupp (1997, Table 2) and Kain and Liu (1999). 16. For bus, there were 193,000 entering and 162,000 leaving the charging zone in autumn 2002 (TfL, 2003c, p. 100); adding 100% for intra-zone trips yields 710,000. For underground, 547,000 passengers arrived within or just outside the charging zone in spring 2002 (TfL, 2003c, p. 113); assuming the fraction of these that exited later in the day is the same as for bus (162/193), and adding another 100% for mid-day trips, yields 2,012,000.

ACKNOWLEDGMENTS This work was supported by the University of California Energy Institute. The author is grateful to Helen Wei for research assistance, and to Herbert Mohring, Chris Nash, Ian Parry, Stef Proost, Martin Richards, Deborah Salon, Ian Savage, Kurt Van Dender, Clifford Winston, participants at the June 2003 STELLA conference in Santa Barbara, two anonymous referees, and the editor for helpful comments. All results and opinions, as well as any errors, are the author’s responsibility.

154

KENNETH A. SMALL

REFERENCES Department for Transport (2002). Transport statistics Great Britain: 2002 edition. London: Stationary Office. Dodgson, J. S. (1986). Benefits of changes in urban public transport subsidies in the major Australian cities. Economic Record, 62, 224–235. Edelson, N. M. (1971). Congestion tolls under monopoly. American Economic Review, 61, 873–882. Giuliano, G., & Small, K. A. (1995). Alternative strategies for coping with traffic congestion. In: H. Giersch (Ed.), Urban Agglomeration and Economic Growth (pp. 199–225). Heidelberg: Springer-Verlag. Glaister, S. (1986). Bus deregulation, competition and vehicle size. Journal of Transport Economics and Policy, 20, 217–244. Glaister, S. (1987). Allocation of urban public transport subsidy. In: S. Glaister (Ed.), Transport Subsidy (pp. 27–39). Newbury, UK: Policy Journals. G´omez-Ib´an˜ ez, J. A. (1996). Big city transit ridership, deficits and politics: Avoiding reality in Boston. Journal of the American Planning Association, 62, 30–50. Goodwin, P. B. (1992). A review of new demand elasticities with special reference to short and long run effects of price changes. Journal of Transport Economics and Policy, 26, 155–169. Jansson, J. O. (1980). A simple bus line model for optimisation of service frequency and bus size. Journal of Transport Economics and Policy, 14, 53–80. Jansson, J. O. (1997). Theory and practice of transport infrastructure and public transport pricing. In: G. G. de Rus & C. Nash (Eds), Recent Developments in Transport Economics (pp. 74–134). Aldershot, UK: Ashgate. Kain, J. F. (1994). Impacts of congestion pricing on transit and carpool demand and supply. In: National Research Council. Curbing Gridlock: Peak-Period Fees to Relieve Traffic Congestion, Transportation Research Board Special Report 242 (Vol. 2, pp. 502–553). Washington, DC: National Academy Press. Kain, J. F., & Liu, Z. (1999). Secrets of success: Assessing the large increases in transit ridership achieved by Houston and San Diego transit providers. Transportation Research A, 33, 601–624. Larsen, O. I. (1996). Efficient public transit subsidies. In: D. A. Hensher, J. King & T. A. Oum (Eds), Proceedings of the 7th World Conference on Transport Research (Vol. 3, pp. 225–236). Oxford: Pergamon Press. Lisco, T. E. (1970). Mass transportation: Cinderella in our cities. The Public Interest, 18, 52–74. Lago, A. M., Mayworm, P. D., & McEnroe, J. M. (1981). Further evidence on aggregate and disaggregate transit fare elasticities. Transportation Research Record, 799, 42–47. Mohring, H. (1972). Optimization and scale economies in urban bus transportation. American Economic Review, 62, 591–604. Mohring, H. (1983). Minibuses in urban transportation. Journal of Urban Economics, 14, 293–317. Nash, C. (1988). Integration of public transport: An economic assessment. In: J. S. Dodgson & N. Topham (Eds), Bus Deregulation and Privatisation (pp. 97–118). Aldershot, UK: Avebury. Parry, I., & Bento, A. (2001). Revenue recycling and the welfare effects of road pricing. Scandinavian Journal of Economics, 103, 645–671. Parry, I., & Bento, A. (2002). Estimating the welfare effect of congestion taxes: The critical importance of other distortions within the transport system. Journal of Urban Economics, 51, 339–365.

Road Pricing and Public Transport

155

Pickrell, D. (1985). Rising deficits and the uses of transit subsidies in the United States. Journal of Transport Economics and Policy, 19, 281–298. Pratt, R., Park, G., Texas Transportation Institute, Cambridge Systematics, Parsons Brinkerhoff Quade, & Douglas, S.G. Associates, & McCollom Management Consulting (2000). Traveler response to transportation system changes: Interim handbook. http://gulliver.trb.org/ publications/tcrp/tcrp webdoc 12.pdf, accessed April 2002. Savage, I., & Schupp, A. (1997). Evaluating transit subsidies in Chicago. Journal of Public Transportation, 1, 93–117. Schroeter, J. (1983). A model of taxi service under fare structure and fleet size regulation. Bell Journal of Economics, 14, 81–96. Small, K. A. (1992). Urban transportation economics, fundamentals of pure and applied economics, 51. Chur, Switzerland: Harwood Academic Publishers. Small, K. A., & G´omez-Ib´an˜ ez, J. A. (1999). Urban transportation. In: P. Cheshire & E. S. Mills (Eds), Handbook of Regional and Urban Economics: Applied Urban Economics (Vol. 3, pp. 1937–1999). Amsterdam: North-Holland. Transport for London (2001). Transport statistics for London 2001. London: Transport for London. Transport for London (2003a). Central London congestion charging scheme: Three months on. London: Transport for London. www.tfl.gov.uk/tfl/pdfdocs/congestion charging/cc-threemonth-report.pdf, accessed 1 September 2003. Transport for London (2003b). Congestion charging delivers better buses. London: Transport for London press release, 18 March. www.tfl.gov.uk/buses/press 666.shtml, accessed 25 March 2003. Transport for London (2003c). Impacts monitoring programme: First annual report. London: Transport for London. www.tfl.gov.uk/tfl/cc monitoring.shtml, accessed 1 September 2003. Transport for London (2003d). Public transport improvements – After. London: Transport for London. www.tfl.gov.uk/tfl/cc improve pt after.shtml, accessed 18 April 2003. Van Dender, K. (2003). Transport taxes with multiple trip purposes. Scandinavian Journal of Economics, 105, 295–310. Viton, P A. (1983). Pareto-optimal urban transportation equilibria. In: T. E. Keeler (Ed.), Research in Transportation Economics (Vol. 1, pp. 75–101). Greenwich, CT: JAI Press. Wardman, M. (2001). A review of British evidence on time and service quality valuations. Transportation Research E, 36, 107–128. White, P. (1995). Deregulation of local bus services in Great Britain: An introductory review. Transport Reviews, 15, 185–209. Winston, C., & Shirley, C. (1998). Alternate route: Toward efficient urban transportation. Washington DC: Brookings Institution Press. Yang, H., & Wong, S. C. (1998). A network model of urban taxi services. Transportation Research B, 32, 235–246.

APPENDIX Parameters for the Numerical Exercise The first set of parameters describes the base scenario for the London bus system in 2000–2001. Data are lacking for the central area specifically, so the ratios apply to the entire London region.

156

KENNETH A. SMALL

 Cost recovery ratio (R/C): Statistics from Transport for London (TfL) show that for 2000–2001, the London bus system’s receipts and costs were nearly identical, with receipts £656 million and costs £643 million (TfL, 2001, p. 38, Table 21). However these receipts include £118 million in “concessionary fare reimbursements” (Department for Transport, 2002, Table 5.6), which represent subsidies rather than revenues. Therefore R/C = (643–118)/656 = 0.80.  Ratio of user in-vehicle time costs to agency cost (U v /C): The value of in-vehicle time for bus users is taken to be 40% of the London area wage rate, which is estimated here at £530/40 from weekly earnings in year 2000 (TfL, 2001, p. 7, Table 2b). Thus ␣v = £ 5.30/hr. The number of passenger-hours in buses per year is calculated at 4,709 million passenger-km in 2000–2001 (TfL, 2001, p. 37, Table 20a) divided by average bus speed for the inner area of approximately 13.2 km/hr.12 This yields U v = £ 1,891 million for 2000–2001. Again using bus system costs for that year of £643 million (as in the previous paragraph), U v /C = 1,891/643 = 2.9.  Size-related agency cost as fraction of agency cost (C2 /C): White (1995, p. 197) quotes a 1990 study estimating that a minibus with 16–20 seats has operating cost per bus-km about 65% that of a full-size bus in the U.K. Assuming this applies to 20 seats and a full-size bus is 86 seats, extrapolation implies that C2 /C ≈ 0.45.13,14 The next parameters are demand elasticities:  Fare elasticity (␧ ): Some typical estimates include −0.5 for a selection of studies, f −0.35 for London bus, −0.25 for Chicago peak bus, and −0.32 for Houston and San Diego.15 The value measured for London bus is most appropriate here, but must be reduced further because the fare changes considered here apply only to the central area whereas many trips begin or end outside that area. The value chosen is therefore εf = −0.25.  Service elasticity (␧M ): some estimates include headway elasticities of 0.47 (Pratt et al., 2000, pp. 9–26, summarising Lago et al., 1981); and service elasticities of 0.32 and 0.65 for Dallas and San Diego (Pratt et al., 2000, pp. 9–14), 0.5 or higher in several U.S. cities (Kain, 1994, p. 543), and 0.71 for Houston and San Diego (Kain & Liu, 1999). With such a large initial share, London’s bus system could be expected to have a somewhat smaller elasticity than other cities with perhaps an elasticity of 0.40; this is reduced further, to the value ␧M = 0.30, to reflect the fact that many trips are only partly within the central area where the increased bus-miles are assumed to occur. The third set of parameters describes the very early results of the congestion pricing implementation in London which started in February 2003.

Road Pricing and Public Transport

157

 Change in bus speed (dS/S): According to TfL (2003b), central area bus speeds during the morning peak rose 15% (from 10.4 to 12 km/hr) by the end of the second week of implementation, while speeds outside the central area were barely changed. This figure may exaggerate the effect of congestion charging because bus speed had already risen to 11.6 km/hr just before congestion charging began, at least partly due to some street improvements (TfL, 2003c, p. 107); on the other hand, it ignores some advantages of reliability improvements that were also documented (TfL, 2003c, pp. 108–111). Here a more conservative figure is taken by assuming that the speed before congestion charging was 11 km/hr, the average figure for 2002 (TfL, 2003c, p. 107); i.e. the increase is from 11 to 12 km/hr, or 9%.  Modal diversion (dQ(1) /Q): During the first few months, modal diversion to all public transport for trips crossing the charging zone boundary was estimated at 110,000 passengers or approximately 3% of current public transport ridership (TfL, 2003a, p. 6). The breakdown between bus and rail is not reported, but a rough estimate is possible. Before congestion charging, daily bus and underground ridership to and within the charging zone during the charging period appears to have been about 710 thousand for bus and 2,012 thousand for the underground.16 Together these figures would imply that other public modes, mainly National Rail with no underground connection, accounted for 944 thousand trips or about one-fourth of all public transport trips, which is similar to its modal share for trips entering central London during the three-hour morning peak period in 2001 (TfL, 2003c, p. 130, Fig. 1). TfL estimates that underground ridership to the central area rose only 1% and that the change in National Rail ridership, while not yet measured, was “likely to be comparable” (TfL, 2003a, p. 7); thus 0.01 × (2,012 + 944) = 30 thousand diverted trips are to rail, leaving 80 thousand diverted to bus. This is an 11% increase in bus ridership caused by congestion charging, and may be compared to the 14% increase claimed by TfL (2003a, p. 7) for the single peak hour of 8–9 AM. However, to some extent these estimates are based on year-to-year comparisons, which could be misleading because substantial upgrades to bus service were implemented during 2002; thus some of the increased bus share is probably due to bus improvements (part of dQSQ or dQ in this paper) rather than to the direct effects of the road charge. For this reason, the figure assumed here for dQ(1) /Q is 6%, just over half the amount calculated above.  Subsidy increase from road pricing revenues (d/C): Expected net toll revenue available for public transport in 2003 is £130 per year (TfL, 2003d). It is assumed here to be allocated to bus and underground in proportion to passenger-km travelled on those two modes; thus a fraction 0.387 goes to bus service, based on TfL (2001, pp. 37, 39). These are then divided by estimated agency costs for 2003 of

158

KENNETH A. SMALL

£687.5 million; the latter is obtained by inflating the £643 million for 2000–2001 by two years growth at rate 3.4% per year (based on trends between 1998–1999 and 2000–2001, from TfL, 2001, p. 38, Table 21). The result is approximately d/C = 0.07.

Computational Formulae for Table 3 Bus-km of service (fractional change): (dM SQ + dM  )/M, from (22) and (23). Fare (fractional change): dfSQ /f, from (21). Ridership (fractional change): (dQ SQ + dQ  )/Q, from (20) and (24). Average in-vehicle time cost (change as fraction of fare): dS U v dS U v C d(U v /Q) = −1 =− . f S fQ S C R Average out-of-vehicle time cost (change as fraction of fare): Having already computed dM/M, the change in average cost Uo /Q can be computed from the first two terms in (3). These show that the two components of Uo /Q are proportional to (1/N) and (N/M), respectively; but substituting (7) for N shows that both terms are in fact proportional to M−1/2 . Thus:   d(U o /Q) 1 dM U o /Q 1 dM 2C 1 dM C2 C =− =− =− , 1− f 2 M f 2 M R M C R where the second equality depends on the equality of U w , Ux , and C1 as described by (10). (Equivalently, one may assume N does not change, in which case Ux is inversely proportional to M while U w is unchanged, leading to the same result.) Average agency cost (fractional change): Differentiating (2) with respect to M, S, and Q, we find dS C 1 dM C 2 dQ dC =− + + . C S C M C Q Then average agency cost satisfies:    dC dQ dS C2 dM dQ d(C/Q) = − =− + 1− − . C/Q C Q S C M Q Benefits: from (25), (26), and (27), divided by agency cost.

7.

MARGINAL SOCIAL COST PRICING FOR ALL TRANSPORT MODES AND THE EFFECTS OF MODAL BUDGET CONSTRAINTS

Stef Proost and Kurt Van Dender 1. INTRODUCTION This chapter studies the effects of more efficient pricing in the transport sector. The introduction of road pricing is a central component of social marginal cost pricing, but it is not the only one, as other modes, like rail and urban public transport, need important pricing corrections as well.1 Correcting the prices on one market and not others may decrease the overall welfare gain, or even decrease welfare. The first objective of this chapter is to estimate, for the transport sector as a whole, the magnitude of the required price corrections, as well as their effects on traffic flows, government revenues and welfare. Marginal social cost pricing is often considered as an exercise of academic nature with little political relevance. When proposals for substantial pricing reforms in the transport sector are discussed, one of the major counter-arguments used by interest groups is the risk of unbalanced budgets between transport modes. There is a general concern that taxes on car use would increase substantially, and the receipts would be used to fill the larger deficits in the public transport sectors. Car lobby groups will therefore typically require that all modes balance their budgets. This call for budget

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 159–177 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09007-9

159

160

STEF PROOST AND KURT VAN DENDER

equilibrium is motivated in several ways. First, there is the efficiency concern that prices below average costs lead to excessive use of a commodity. Second, there is the cost-efficiency concern that the absence of a rigid balanced budget requirement for a mode like rail acts like a blank check so that production operations will become less cost-efficient. Third, there is a fear of loss of transparency. Finally, there are concerns regarding unacceptable transfers between population groups. The second objective of this chapter is to assess the effects of balanced budget requirements. More specifically, the efficiency and transport sector effects of three alternative pricing schemes are discussed. Existing pricing practices are compared with marginal social cost pricing (MSC) and with two pricing rules that guarantee a balanced budget. The first is simple average cost pricing for each mode (ACM) and the second is Ramsey pricing with budget equilibrium for the whole transport sector (RMS). Both average cost pricing and Ramsey pricing are deviations from marginal social cost pricing. Revenue effects and potential efficiency losses of imposing budget constraints in the transport sector are gauged. This exercise could be considered as an input into a broader analysis that uses a political economy approach (e.g. Grossman & Helpman, 2001), where issues like earmarking of tax revenues2 and efficiency incentives for the transport ministries, in the sense of Tirole (1994), are studied. The computations are done for a set of European cities and non-urban areas, always using the same aggregate optimal pricing model. The chapter focuses on pricing reform for a given infrastructure, excluding the investment problem. Moreover, it is assumed that all transport is produced in a cost-efficient way, so that any cost-efficiency concern generated by budget balance requirements is ignored.3 Finally, income distribution concerns between population groups are ignored, as this requires a general equilibrium approach that keeps track of the use of transport revenue surpluses or deficits in other sectors of the economy (Mayeres & Proost, 1997, 2003). The pricing rules considered in this chapter are defined in Section 2. The model is briefly described in Section 3. The reference equilibrium and the budget constraints are discussed in Section 4. The effects of the different pricing rules are compared in Section 5. Section 6 concludes.

2. PRICING RULES Three pricing rules are tested. As shown in Table 1 they differ in two characteristics: (i) whether they need to balance the financial transport account or not; and (ii) whether they use marginal social cost information or not.

Marginal Social Cost Pricing

161

Table 1. Transport Pricing Rules. Balanced Modal Transport Account or Financial Cost Recovery Required by transport mode Required for transport sector Not required

Pricing Principle Average Cost

Marginal Cost

Average cost pricing (ACM) Ramsey social cost pricing (RMS) Marginal social cost pricing (MSC)

When the average cost pricing by mode (ACM) rule is used, prices are equal to the sum of financial costs of that mode divided by its total volume. This implies that no attention is paid to the structure of resource costs (fixed or not, sunk or not, etc.), no consideration of any external costs is made, and all transport services (freight, passengers, etc.) within that mode are treated identically. The main goal of average cost pricing is cost recovery. There are many forms of average cost pricing because both the numerator and the denominator are to some extent arbitrary. First, several volume indicators can be used, e.g. trips or vehicle-kilometre for passengers and ton-kilometre or vehicle-kilometre for freight. Second, accounting rules are not uniform (depreciation rules, etc.) and can lead to different total cost concepts (Jha, 1998). Only one of many definitions of average costs is used here, as what matters is to show the importance of cost recovery and the disregard of the resource cost and external cost structure (marginal costs by time of day, etc.) for economic efficiency. When the marginal social cost pricing (MSC) rule is used, prices are equal to the sum of the marginal resource cost (extra cost of driver time, fuel, wear and tear of vehicle, all before taxes) and the marginal external cost (including congestion, air pollution, noise, accidents and maintenance cost of the infrastructure), for a given infrastructure. With this pricing rule, the financial impact per mode is ignored. When the Ramsey social cost pricing for the transport sector (RMS) rule is used, prices are set as optimal deviations from marginal social costs. The deviations are required to meet cost recovery targets for the transport sector as a whole. If marginal social cost pricing generates insufficient revenue to cover financial costs, RMS pricing requires that the margins (price-marginal social cost) are increased in a way that is inversely proportional to the elasticity of demand in the relevant market. Moreover, in setting the margins, care has to be taken of the potential distortionary effects on other transport markets. This means that mark-ups on top of marginal social costs are differentiated between the different transport services (peak, off-peak, passengers, freight). This principle looks complicated but is well known in the theory of monopolistic pricing and is common practice in all businesses that exploit their market power (airlines, soft drinks,

162

STEF PROOST AND KURT VAN DENDER

telecom, computers, private education, supermarkets, etc.).4 The main difference with the application in this study is that private firms are not concerned with external costs. Many alternatives to the pricing rules defined above can be defined, and some may perform better than the ones discussed in this chapter. Two possibilities are worth mentioning. First, one could define the Ramsey pricing principle at the level of each mode and look for optimal differentiation between different services of the same mode (e.g. return or single ticket, etc.). This looks less interesting because the budget balance in this paper is not really imposed at the modal level; the model used here is multi-modal and rather aggregated. Second, two-part tariffs may outperform Ramsey pricing. The reason why Ramsey pricing is considered instead of two-part tariffs is technical: the aggregate simulation model that is used does not allow the computation of the full benefits of two-part tariffs. A micro-simulation model based on representative samples would be needed in order to tackle that issue.5

3. THE MODEL The pricing rules are assessed using TRENEN, a multi-modal partial equilibrium model.6 The model describes the market equilibrium for all surface transport markets simultaneously. It covers passengers and freight, and private and public transport modes. One of its main features is that it can optimise the pricing structure over several modes under any type of pricing constraint. There are different model versions for urban and for non-urban transport. The urban model version is briefly described below, with a focus on passenger transportation. Figure 1 shows the model structure.

3.1. Assumptions The TRENEN model rests on three important assumptions. The first assumption is that the travel conditions in the urban (non-urban) area can be represented by one aggregated speed-flow relationship.7 This means that area specific and route choice features cannot be studied. The second assumption is that demand is generated by a limited number of representative consumers, distinguished on the basis of transport cost differences, not income. Consequently, distributional issues cannot be studied. The third assumption is that the model is medium run and static, taking the level of all transport infrastructure as given, but allowing the car stock and public transport equipment (number of busses and rail carriages) to fully adapt to the level of demand.

Marginal Social Cost Pricing

163

Fig. 1. Model Structure.

3.2. Transport Demand The behaviour of the representative households8 is derived from the maximisation of an indirect utility function defined over generalised costs. The indirect utility function takes a nested constant-elasticity of substitution (CES) form, representing the choice between one aggregate non-transport good and twenty or so transport alternatives.9 The inclusion of the non-transport good allows total demand for transport to vary. The transport alternatives are combinations of mode, time of day and type of vehicle used, as shown in the utility tree in Fig. 2. The choices of the representative consumer are driven by generalised prices. Through the time-component of these generalised prices, the effect of congestion on travel time and modal choice is taken into account. The generalised costs for public transport depend on the frequency of service, thus taking the Mohring effect (Mohring, 1972) into account. The demand for freight transport services is modelled by assuming that each firm minimises its total production cost for a given output level. It can choose between non-transport inputs (labour, capital, other intermediate goods) and a wide range of transport inputs (different modes, times of day, etc.). The behaviour of the firms is represented by a nested CES cost function that is defined over generalised costs, again taking account of traffic congestion. In a typical case

164

STEF PROOST AND KURT VAN DENDER

Fig. 2. The Nested Utility Tree for the Behaviour of Urban Households. Note: Only the peak branch is shown.

study, a choice can be made between peak and off-peak freight, and in both periods different modes (rail, road, inland waterways) are available. Road freight can make use of light-duty and heavy-duty trucks.

3.3. Transport Supply The supply side of the model is simple. Supply is represented by cost functions. In each transport market, marginal resource costs of inputs other than time are constant per vehicle-kilometre, but the marginal costs may differ across transport markets (i.e. across modes, time periods, vehicle types, etc.). In passenger car and lorry markets there are no fixed costs10 and all inputs are supplied at the marginal resource cost plus tax. For collective transport markets a linear cost function is used. The constant term represents the fixed costs (e.g. administration costs, storage facilities, non-vehicle network costs). The variable term represents the rental costs of carriages and buses, as well as the costs of fuel and drivers. A constant occupancy rate (different for peak and off-peak) is assumed so that marginal resource costs per passenger kilometre can be defined. The frequency of service is taken to adapt in function of the total demand for public transport, so that the waiting time is variable. Finally, the policy maker can change environmental and other vehicle characteristics through regulations that affect the resource costs of vehicles.

Marginal Social Cost Pricing

165

For the non-transport market, the producer prices equal the constant marginal resource cost plus tax.

3.4. Calibration and Operation of the Model The model is calibrated to observed or forecasted transport volumes, prices and speeds in a given region. The model is completed by a speed-flow relationship (that is best derived from experiments with a network model) and by resource cost and external cost functions for all transport modes. Elasticities of substitution for the nested utility and cost functions were chosen so as to obtain price elasticities in line with the values reported in the literature. In particular, own price elasticities for peak period trips are around −0.3 for cars and −0.35 for public transport. Off-peak elasticities are higher: −0.6 for cars and −0.87 for public transport. Cross-price elasticities are positive, but usually fairly small. In simulation mode, the calibrated model computes the equilibrium for all the transport markets and for the non-transport commodity market, given a set of policy parameters (taxes, regulations on type of vehicles, etc.). The equilibrium is characterised by a set of speeds, volumes and generalised costs such that no transport user wants to alter his choice. For this equilibrium the model computes external costs (air pollution, congestion, accidents, noise, etc.) and a welfare indicator. The welfare indicator can loosely be described as the sum of total consumer surplus on the transport markets, producer surplus (the negative of the total production costs of firms and the deficit or surplus of the public transport firms) and total net tax revenue, minus external costs other than congestion (congestion is included in the consumer surplus defined over generalised prices).11 The model can also be used in optimisation mode, by maximising the welfare indicator under pricing constraints that represent the pricing instruments. Only if there are no constraints on the tax differentiation over different markets will the derived optimum correspond to marginal social cost pricing. In most applications it will only be possible to compute second best optima. It is well known that this can be numerically difficult and can generate counter-intuitive results. When every pricing solution is multi-dimensional (e.g. taxes for twenty transport markets), the best constrained solution can not be found using an ad hoc search procedure. The standard version of TRENEN contains no (balanced) budget requirement in any transport market or in any combination of transport markets. Instead, changes in tax revenues as caused by changes in transport policy are valued exogenously, and the exogenous value reflects an assumption on revenue use. For the purposes of this chapter, tax revenues get the same weight as consumer income. Here, the optimisation model will be used to compute the marginal

166

STEF PROOST AND KURT VAN DENDER

social cost pricing solution (no constraint on tax instruments), the Ramsey pricing solution (the total tax revenue of the transport sector is constrained so that taxes in the transport sector have to be adjusted away from the MSC value) and the average cost pricing solution. In the latter solution all taxes will be constrained to equal the average resource cost. Since average cost is not constant, this is not a straightforward computation.

4. AVERAGE COST POLICY SCENARIOS AND THE REFERENCE EQUILIBRIUM 4.1. Construction of the Policy Scenarios The data used for calibrating the models were originally collected for a study on the public revenue effects of optimal pricing (Roy, 2000). That study collected data sets for different countries, including traffic flows, prices, taxes and marginal external costs. The first step in the construction of a policy scenario for ACM and RMS is to determine the revenues that need to be raised through transport taxes. The data sets for Germany and the U.K. used in this study provide information on the cost side for collective transport modes and include urban metro, tram and bus, and nonurban bus and train, for 1995 and 2005. This information is sufficient to determine the revenue requirements for those modes, excluding road network costs. Information on road network costs is harder to collect. An estimate on the basis of the numbers for Germany and Switzerland, produced by Link and Suter (2001), was computed and analogous assumptions for the U.K. were made. The values derived for total road infrastructure costs are very uncertain, but they are not crucial for the results. Once the revenue requirements are determined, the scenarios can be designed. These are shown on Table 2.

4.2. Taxes and Marginal External Costs in the Reference Equilibrium Table 3 compares the reference taxes to total marginal external costs (TMEC). These are equal to the sum of marginal congestion costs (MCC) and other marginal external costs such as pollution, accidents and noise (MPAN). The values correspond to a small petrol car with one driver-occupant and a

Marginal Social Cost Pricing

167

Table 2. Policy Scenarios. Scenario

Details

Reference equilibrium (REF)

This scenario uses the expected reference prices for 2005 in all transport markets. It serves as the benchmark to which the remaining scenarios are compared.

Average cost pricing (ACM)

Here the modal budget is financed by a uniform tax per vehicle-kilometre for the private modes and by a uniform tax per passenger-kilometre or ton-kilometre for all public transport modes.

Ramsey social pricing (RMS)

The transport-sector-wide budget of total costs is financed through Ramsey taxes, allowing full differentiation across transport markets. The taxes maximise social welfare subject to the budget constraint. The social welfare function takes all external costs into account, so that optimal prices will be optimal deviations from marginal costs, where deviations are necessary to meet the revenue requirements.

Marginal social cost pricing (MSC)

This is the theoretical optimum obtained by maximisation of the welfare function, allowing full differentiation of taxes across transport markets, without any budget constraint.

representative bus at peak and off-peak times. The taxes considered include taxes on car ownership (annual license) and on car use (fuel taxes, existing tolls, and VAT). For public transport the tax equals the difference between the fare and the marginal resource cost. A negative tax is a subsidy. Marginal external congestion costs clearly dominate in peak periods.12 In order to assess the pricing inefficiencies the per passenger-kilometre total tax (“Tax” in the first column) can be compared with the total marginal external costs (TMEC in the second column). It can be seen from Table 3 that the total marginal external costs exceed taxes in all urban areas during peak hours (ratios are higher than 1 in all cases). It is evident from the table that current taxes more than cover the external costs in urban areas during off-peak periods and in interurban areas at all times including peak times. It can also be seen that there are inefficiencies for the public transport modes. With one exception, public transport prices are too low in the peak and may be too high or too low in the off peak. This may illustrate the second best policy of subsidising public transport to reduce the problems in the underpriced peak car market. Policy makers often do not appreciate that the introduction of road pricing may be an opportunity to correct the public transport prices as well.

168

STEF PROOST AND KURT VAN DENDER

Table 3. Marginal External Costs and Tax Levels in the Reference Situation – Partial Equilibrium Model (EURO/Passenger Kilometre, 2005). Peak Car

Off-Peak Car

Tax

TMEC

MCC

MPAN

Tax

TMEC

MCC

MPAN

Germany D¨usseldorf M¨unchen M¨unster Westphalen

0.097 0.097 0.097 0.089

0.185 0.308 0.537 0.028

0.171 0.252 0.523 0.022

0.014 0.056 0.014 0.006

0.088 0.088 0.088 0.106

0.064 0.128 0.113 0.014

0.051 0.073 0.100 0.009

0.013 0.055 0.013 0.005

U.K. London South-East

0.118 0.177

0.503 0.021

0.447 0.013

0.056 0.008

0.108 0.157

0.090 0.009

0.035 0.001

0.055 0.008

Peak Bus

Off-Peak Bus

Tax

TMEC

MCC

MPAN

Tax

TMEC

MCC

MPAN

Germany D¨usseldorf M¨unchen M¨unster Westphalen

0.016 −0.040 0.041 0.067

0.037 0.279 0.064 0.015

0.008 0.016 0.030 0.002

0.029 0.263 0.034 0.013

0.093 0.060 0.082 0.104

0.036 0.226 0.042 0.021

0.004 0.006 0.009 0.002

0.032 0.220 0.033 0.019

U.K. London South-East

−0.02 0

0.708 0.023

0.069 0.002

0.639 0.021

−0.010 0.003

0.398 0.055

0.005 0.001

0.393 0.054

Note: All prices are in EUROS per passenger kilometre. Tax: total amount of taxes paid. TMEC = MCC + MPAN = total marginal external cost. MCC = Marginal external congestion cost. MPAN = Marginal external costs of pollution, accidents and noise. Negative taxes are subsidies (= marginal resource cost − price). Source: Own calculations.

The difference between taxes and total marginal external costs is as such insufficient to compute optimal taxes. The marginal external cost (mainly the congestion cost) is a function of the volume of transportation and this is a function of the tax itself so that one tends to overstate the necessary increase in taxes when using the marginal external cost information in the reference situation as a guideline.

5. COMPARISON OF ALTERNATIVE PRICING RULES This section discusses the main effects of alternative pricing rules.

Marginal Social Cost Pricing

169

Table 4. Tax Levels for Different Pricing Scenarios, 2005. EURO/PKM

Peak Car REF

ACM

Germany D¨usseldorf M¨unchen M¨unster Westphalen

0.097 0.097 0.097 0.089

0.037 0.037 0.038 0.035

U.K. London South-East

0.118 0.177

0.105 0.060

RMS

Off-Peak Car MSC

REF

ACM

RMS

MSC

0.239 0.308 0.090 0.070

0.253 0.308 0.380 0.099

0.088 0.088 0.088 0.106

0.037 0.037 0.033 0.035

0.089 0.203 −0.070 0.070

0.156 0.203 0.184 0.084

0.589 0.034

1.000 0.114

0.108 0.157

0.105 0.060

0.505 0.021

0.840 0.100

Peak Bus

Off-Peak Bus

REF

ACM

RMS

MSC

REF

ACM

RMS

MSC

Germany D¨usseldorf M¨unchen M¨unster Westphalen

0.016 −0.040 0.041 0.067

0.311 0.343 0.100 0.100

0.143 0.382 −0.060 −0.040

0.104 0.382 0.041 0.027

0.060 0.060 0.082 0.104

0.522 0.450 0.146 0.190

0.089 0.349 −0.020 −0.030

0.104 0.349 0.082 0.029

U.K. London South-East

−0.02 0

0.234 0.082

0.150 0.015

0.785 0.043

−0.010 0.003

0.231 0.254

−0.060 0.043

0.517 0.073

Note: PKM: Passenger-kilometre. Source: Own calculations.

5.1. Optimal Transport Taxes in Different Pricing Schemes Table 4 presents an overview of the tax levels in car13 and public transport markets, for the various policy scenarios. First, compare the two pricing rules that meet the budget requirement, ACM and RMS, to the reference situation (REF). Clearly, achieving cost recovery by mode through average cost pricing (ACM) leads to substantial car tax reductions and (very) large bus tax increases in all cases. The resulting change in relative prices of private and public modes leads to an increase in the modal share of cars, as Table 5c shows. Combined with the small increase in peak period traffic volumes, shown in Table 5a and b, this exacerbates congestion and other external costs. Allowing the same budget to be raised by a Ramsey-rule (RMS) avoids these problems, as it always leads to lower bus taxes relative to car taxes. In most cases, the Ramsey bus taxes are lower than the ACM bus taxes in absolute terms as well. This price differentiation shows that, even in the presence of a cost recovery rule, price differentiation under a

170

STEF PROOST AND KURT VAN DENDER

Table 5. Under Different Pricing Scenarios, 2005 (a) Traffic Level Index (PCU), (b) Transport Demand Index (PKM), and (c) Modal Share of Car in Peak and Off-Peak (%). REF (a) Traffic level index (PCU) Germany D¨usseldorf M¨unchen M¨unster Westphalen region U.K. London South-East region

ACM

RMS

MSC

1 1 1 1

1.13 1.12 1.07 1.03

0.95 0.97 1.07 0.99

0.91 0.88 0.90 0.99

1 1

1.06 1.11

0.91 1.04

0.76 1.01

1.05 1.05 1.03 0.98

1.04 1.03 1.10 1.08

0.92 0.94 0.95 1.01

1.03 0.96

1.11 1.03

0.88 1.00

(b) Transport demand index (PKM) Germany D¨usseldorf 1 M¨unchen 1 M¨unster 1 Westphalen region 1 U.K. London South-East region

1 1

Peak Car Share REF

RMS

MSC

REF

ACM

RMS

MSC

(c) Modal share of car in peak and off-peak (%) Germany D¨usseldorf 67.1 76.0 65.7 M¨unchen 71.2 76.7 68.2 M¨unster 84.0 86.3 83.7 Westphalen 85.7 89.4 78.3

65.8 69.0 80.0 85.0

66.1 70.3 84.2 85.7

82.4 78.3 89.1 90.9

62.4 67.1 88.6 77.8

62.5 92.1 81.4 83.2

45.9 74.7

69.9 77.0

74.9 91.5

62.4 76.6

57.9 76.4

U.K. London South-East

53.4 70.1

ACM

Off-Peak Car Share

55.0 81.1

46.3 74.7

Source: Own calculations.

RMS scenario may generate substantial welfare gains. It should be noted that in the case of M¨unster the Ramsey rule even calls for subsidising off-peak car travel. This is an extreme case of a second-best policy geared towards reducing peak period car use. Although the TRENEN model allows for this theoretical possibility, such a pricing system may not be a realistic policy option.

Marginal Social Cost Pricing

171

Second, compare the reference situation to MSC pricing. Although the optimal taxes under marginal social cost pricing (MSC) usually are higher than the reference taxes for peak period car trips, there are exceptions, such as South-East England, which has low average congestion levels in the reference equilibrium. The impact on off-peak car taxes is more diverse: in both regions, Westphalen and South-East England, the off-peak car taxes decrease. In all cities except London, car taxes increase in comparison to the reference equilibrium. This result may be driven by the geographical scale of the case studies, as those cases covering the largest areas have decreasing off-peak taxes. Taking large networks into account may tend to spread out off-peak congestion levels.

5.2. Traffic Level and Composition Table 5a, b and c show the impact of the pricing mechanisms on traffic levels (passenger car units, PCU), on transport demand (passenger-kilometre, PKM), and on modal shares, respectively. Average cost pricing leads to an increase in traffic levels and transport demand, because the taxes are on average reduced with respect to the reference situation. The increase in PCU is larger than for PKM because of a modal shift towards car trips, as shown in Table 5c, away from collective modes, as a consequence of the relatively high revenue requirement (hence, relatively high taxes) for collective modes, and the relatively low revenue requirement in car markets. This illustrates that defining budget requirements in narrow sets of transport markets may have strong effects on modal split. The simple average cost pricing scheme performs badly both in terms of aggregate travel demand and in terms of modal split for a given level of demand.14 In both cases it causes an increase in PCUs and in demand for travel in general, thus exacerbating the problem of congestion and other transport externalities. Ramsey social pricing, on the other hand, manages to combine lower PCU levels (hence less congestion) with increased transport demand in a number of cases: D¨usseldorf, M¨unchen, London, and to a lesser extent, the region of Westphalen. The reason is that, despite the low revenue requirements, relative modal prices can be set to achieve a higher use of public transport. In practice this requires low or zero fares for collective modes. In other cases, such as M¨unster and the South-East region in the U.K., the revenue requirement is too low to allow for sufficient price differentiation, and increased PCU levels result. Marginal social cost pricing usually reduces travel demand (PKM) and traffic flows (PCU) in comparison to the reference situation, and leads to revenues in excess of the revenue requirements specified for average cost pricing and Ramsey social pricing. Under marginal social cost pricing there is no longer a justification

172

STEF PROOST AND KURT VAN DENDER

Table 6. Share of Peak Period PCU, 2005. % Under REF

% Under ACM

% Under RMS

% Under MSC

Germany D¨usseldorf M¨unchen M¨unster Westphalen region

63.3 58.7 59.1 64.0

61.0 56.9 58.3 64.0

62.1 57.5 55.0 63.7

62.2 58.0 56.5 63.3

U.K. London South-East region

68.1 69.5

66.9 69.8

68.0 69.3

69.1 69.3

for subsidising public transport beyond the level of fixed costs.15 The efficient modal split is obtained by pricing all modes at their marginal social cost. Table 6 shows the share of trips (measured in PCUs) that take place during peak hours. As can be seen this share is less sensitive to the pricing scheme than is the total traffic volume. Ramsey social pricing performs much like marginal social cost pricing in this respect, through a second best correction of prices. The effect of average cost pricing is to slightly decrease the share of peak hour trips in most cases. In the urban case studies, the impact of the different pricing schemes on freight transport is small, and the directions of change are similar to those of passenger car transport. In regional contexts, average cost pricing decreases the modal share of rail freight, in comparison to the reference situation. Ramsey social pricing does the opposite: it strongly pushes the share of rail up. In the Westphalen case, Ramsey social pricing leads to a much higher share of rail freight than in the marginal social cost pricing scenario. In the South-East U.K. case, the Ramsey share is approximately equal to the marginal social cost pricing share. Overall, the impact of the various schemes on aggregate freight demand is rather small.

5.3. Welfare Impacts Table 7 shows the welfare changes induced by the different pricing scenarios for the various cases. These welfare changes are expressed as a percentage of total generalised income, defined as the sum of money and time budgets, which is larger than national income. A welfare gain of 1% of generalised income realised on the transport market that only counts for 10% or less of total national income is therefore to be considered as important. Of course, marginal social cost pricing outperforms Ramsey social cost pricing, which in turn outperforms

Marginal Social Cost Pricing

173

Table 7. Welfare Impacts of Pricing Scenarios (% Change with Respect to REF, 2005). REF

ACM

RMS

MSC

Germany D¨usseldorf M¨unchen M¨unster Westphalen region

0 0 0 0

−0.79 −0.61 −2.45 −0.17

+0.09 +0.14 −2.15 −0.06

+0.14 +0.41 +2.45 +0.09

U.K. London South-East region

0 0

−0.76 −1.89

+1.28 +0.18

+2.70 +0.55

Source: Own calculations.

average cost pricing. First, the introduction of a budget constraint has a clear efficiency cost for the transport sector. Second, the way in which this constraint is met has further consequences for the welfare effects. Ramsey social pricing cannot be worse than average cost pricing and cannot be better than marginal social cost pricing. Interestingly, average cost pricing leads to a reduction of welfare with respect to the reference situation in all cases. While the size of the reduction varies substantially between cases, the two basic reasons for the welfare reductions are the same. First, the current transport prices go some way towards a second-best pricing structure. Under-priced passenger car transport (from the social point of view) is often combined with subsidised public transport, so that relative price distortions are reduced. Such a policy is not feasible under the modal budget requirements used in the average cost pricing simulations. Taxes for each mode are only determined by the modal revenue requirement, so that no account can be taken of prices in substitute modes. Second, the modal budget constraints require less revenue than is raised in the reference situation. This means that the revenues from current transport taxes are higher than what is required to balance the transport sector financially. Optimal commodity tax theory shows that, if transport demand is relatively inelastic, revenue-raising in that sector tends to limit the efficiency cost of collecting the required total amount of government revenue.16 The fact that the transport sector at present is “revenue positive” may then be justified from the optimal taxation point of view, although there is no guarantee that relative prices or the size of the surplus are anywhere near optimal.

174

STEF PROOST AND KURT VAN DENDER

If the revenue requirement were increased above the transport-related requirement, average cost pricing could, but need not, perform better than the reference price structure. As the peak-period taxes from average cost pricing approach the peak period external costs, the performance of ACM improves. This improvement will be counteracted to some extent by the growing deviation between off-peak taxes and off-peak external costs.17 However, since peak-period congestion costs are the dominant externality, a net improvement of welfare should be expected. Ramsey social pricing is better or worse than the reference situation depending on the case considered. The problem of reducing the amount of revenue to be raised, as compared to the reference situation, is less prominent here, as price differentiation is still possible. Second-best relative price structures are still a feasible policy option. It should be noted however that Ramsey social pricing performs considerably worse than marginal social cost pricing in all cases. This result suggests that the level of the revenue requirement is an important co-determinant of the welfare effects of transport pricing policies. Raising no more revenue than the one required for infrastructure maintenance (keeping the level of road infrastructure constant) may strongly reduce the welfare potential of pricing policies aimed at internalising externalities. The variation in results between cases depends on the degree of cost coverage of collective modes in the reference situation, and on the degree to which the new budget constraint allows sufficient differentiation of prices with respect to transport externalities. The high welfare gains for M¨unster have to do with the very steep speed flow relationship in the reference equilibrium. This implies that small reductions in volumes in the peak can generate important increases in speed and in welfare for the local population.

6. CONCLUSIONS In this chapter the potential of more efficient pricing for the transport sector as a whole and the impact of budget constraints has been analysed. Implementing marginal social cost pricing without budget constraints generates in principle the highest welfare gains. Marginal cost pricing would require important changes both in the private transport sector, such as road pricing, and in the public transport sector. For example, in many regions peak public transport fares would need to be increased. Overall mobility in some areas would need to decrease by 5–10% and car share in total transport would need to decrease at peak times. Two extreme scenarios have been studied to add a balanced budget constraint to the overall transport pricing reform. The first is simple average cost pricing per

Marginal Social Cost Pricing

175

mode. The second is Ramsey pricing with a budget constraint for the transport sector as a whole. The comparison of these two pricing approaches shows that Ramsey-type pricing rules perform significantly better in terms of welfare than average cost based rules, and that the absence of a modal budget constraint allows to better adapt prices to marginal social cost. The case studies suggest that the quantitative effects may be important. In all the cases studied requiring that modal budgets are met exactly through average cost pricing reduces welfare in comparison to the reference situation by 0.5% to more than 2.5% of national income. When the budget constraint is attained through Ramsey taxes, the results are most often welfare improving compared to the reference. The welfare cost of imposing the budget constraint, however, remains substantial in comparison to a marginal social cost pricing scheme. These results as such say nothing about the political and social feasibility of the various pricing rules. Furthermore, there are other ways of defining average cost pricing schemes, different from the ones analysed here, and such alternative definitions may produce better results. These alternative schemes will however become more complex, and they will still perform worse than marginal-cost-based pricing approaches.

NOTES 1. See Glaister and Lewis (1978) for an early application of optimal pricing with several modes. 2. See Newbery and Santos (1999) for a contribution on earmarking of road taxes. 3. It is well known in the theory of incentives that a regulation scheme that imposes a maximum deficit can generate production efficiency (Laffont & Tirole, 1993). 4. See any intermediate micro-economics textbook, e.g. Pindyck and Rubinfeld (2001, pp. 370–403). 5. The design of optimal two-part tariffs in the presence of externalities is discussed in De Borger (2001). 6. For a detailed description of the model see Proost and Van Dender (2001a, b). 7. This aggregate speed-flow relationship can be derived from a detailed network model (see O’Mahony et al., 1997). 8. In the urban case studies, the following representative consumer groups are distinguished: consumers with and without access to free parking, and consumers living inside and outside the urban area under study. The interregional case studies do not distinguish between consumer groups. 9. Labour supply is kept fixed in the model, as well as the total volume of production and the product price of general consumption goods. Endogenising these variables would lead to a full general equilibrium model. However, there is a trade-off here between the detail in the modelling of the transport sector and the coverage of the whole economy. 10. In line with the medium run horizon of the model, car ownership costs are expressed on a per vehicle-kilometre basis.

176

STEF PROOST AND KURT VAN DENDER

11. The welfare indicator in fact equals the sum of the indirect utility levels of the representative individuals who receive all profits in all firms as well as all net tax revenue, less the external costs other than congestion. This measure is more consistent than a simple sum of consumer surpluses on markets. 12. Note that the estimate for marginal external congestion costs for M¨unster is high. This follows from the small geographical scope of this case, and it explains the high welfare gains from MSC-pricing. 13. Car taxes slightly differ in the peak and off peak period even though the tax system does not distinguish between times of day. The difference is made by the TRENEN model, which takes into account the difference in fuel consumption between peak and off-peak. 14. Although not computed, it is likely that a Ramsey pricing scheme with modal budget constraints, as opposed to a sector-wide constraint, will suffer from the same problem, to a lesser – but still considerable – extent. 15. Economies of density in public transport, or the Mohring effect, is the exception. If the frequency of service increases due to additional passengers, there is a positive externality in public transport. The marginal social cost thus equals the sum of the marginal operating costs and the marginal external costs (congestion, air pollution, accidents) minus the external benefit of a more frequent service. Using data for Brussels and London, Van Dender and Proost (2001) find that taking account of this positive externality increases the welfare potential of pricing reforms by some 10%, and that it decreases fare revenues while public transport expenditures increase. Chapter 6 of this volume provides a detailed quantification of the effects of the introduction of congestion tolls on public transport, for the case of London. It finds that economies of density matter, but that cost savings due to increased speeds of operation, which are ignored in Van Dender and Proost (2001), are much larger. 16. The potential interactions with other distorted markets such as the labour market are ignored in this study. Mayeres and Proost (1997), Parry and Bento (2001) and Van Dender (2003) analyse this issue. 17. Sensitivity analysis for the D¨usseldorf case shows that increasing the revenue requirement to 150% of the central case actually decreases the performance of ACM in terms of welfare. Decreasing the revenue requirement to 50% of the central scenario improves the performance of ACM. These however, are not general results. In a second sensitivity analysis, the budgets were linked to changes in traffic levels. At the central scenario revenue requirement, this decreases the performance of ACM. At 50% of the central scenario requirement, this link increases the performance of ACM. The relationship between changes in the budget requirement and the performance of ACM is clearly non-monotonous. Interactions between budgets, the implied ratio of modal and time-of-day differentiation of taxes, and changes in the budget requirement, may produce counterintuitive results.

ACKNOWLEDGMENTS Support from the European Commission under the Fifth Transport Research and Technology Framework Program is gratefully acknowledged. The authors benefited from the comments on earlier versions by the editor of this volume, by two anonymous referees and by Inge Mayeres.

Marginal Social Cost Pricing

177

REFERENCES De Borger, B. (2001). Discrete choice models and optimal two-part tariffs in the presence of externalities: Optimal taxation of cars. Regional Science and Urban Economics, 31, 471–504. Glaister, S., & Lewis, D. (1978). An integrated fares policy for transport in London. Journal of Public Economics, 9, 341–355. Grossman, G., & Helpman, E. (2001). Special interest politics. Cambridge, London: MIT Press. Jha, R. (1998). Modern public economics. London, New York: Routledge. Laffont, J., & Tirole, J. (1993). A theory of incentives in procurement and regulation. Cambridge, London: MIT Press. Link, H., & Suter, S. (2001). Summary of the German and Swiss pilot accounts – Background paper – Draft version. UNITE Deliverable 5, Mimeo, presented at Ecole Nationale des Ponts et Chauss´ees, Paris, 17–18 September. Mayeres, I., & Proost, S. (1997). Optimal tax and public investment rules for congestion type of externalities. Scandinavian Journal of Economics, 99, 261–279. Mayeres, I., & Proost, S. (2003). Reforming transport pricing: An economist’s perspective on equity, efficiency and acceptability. In: J. Schade & B. Schlag (Eds), Acceptability of Transport Pricing Strategies. Oxford: Elsevier. Mohring, H. (1972). Optimization and scale economies in urban bus transportation. American Economic Review, 62, 591–604. Newbery, D., & Santos, G. (1999). Road taxes, road user charges and earmarking. Fiscal Studies, 20, 103–132. O’Mahony, M., Kirwan, K., & McGrath, S. (1997). Modelling the internalisation of external costs of transport. Transportation Research Record, 1576, 93–98. Pindyck, R., & Rubinfeld, D. (2001). Microeconomics (5th ed.). New Jersey: Prentice-Hall. Parry, I., & Bento, A. (2001). Revenue recycling and the welfare effects of road pricing. Scandinavian Journal of Economics, 103, 645–671. Proost, S., & Van Dender, K. (2001a). The welfare impacts of alternative policies to address atmospheric pollution in urban road transport. Regional Science and Urban Economics, 31, 383–412. Proost, S., & Van Dender, K. (2001b). Methodology and structure of the urban model. In: B. De Borger & S. Proost (Eds), Reforming Transport Pricing in the European Union: A Modelling Approach (pp. 65–92). Cheltenham, Northampton: Edward Elgar. Roy, R. (Ed.) (2000). Revenues from efficient pricing: Evidence from the member states. UIC/CER/European Commission DG-TREN Study. Tirole, J. (1994). The internal organization of government. Oxford Economic Papers, 46, 1–29. Van Dender, K. (2003). Transport taxes with multiple trip purposes. Scandinavian Journal of Economics, 105, 295–310. Van Dender, K., & Proost, S. (2001). Optimal urban transport pricing with congestion and economies of density. Center for Economic Studies, Katholieke Universiteit Leuven, ETE Working Paper 2001-19. www.econ.kuleuven.ac.be/ew/academic/energmil/downloads/ete-wp01-19.pdf.

This Page Intentionally Left Blank

8.

WELFARE AND DISTRIBUTIONAL EFFECTS OF ROAD PRICING SCHEMES FOR METROPOLITAN WASHINGTON DC

Elena Safirova, Kenneth Gillingham, Ian Parry, Peter Nelson, Winston Harrington and David Mason 1. INTRODUCTION Traffic congestion imposes substantial costs on society: the Texas Transportation Institute (2001) estimates that travel delays and the resulting extra fuel combustion cost the United States $68 billion in 2000. Congestion is likely to worsen in the future with continued growth in vehicle miles; for example, in the Metropolitan Washington DC area, vehicle miles are projected to increase by more than 40% over the next 20 years (National Capital Region Transportation Planning Board, 2002). Meanwhile environmental constraints, neighbourhood opposition, and budgetary limitations are making it ever more difficult to build new roads. Consequently, transport planners and policy analysts are looking to novel approaches for bringing expanding demand for driving in line with available road capacity. One such approach is time-of-day congestion pricing, which economists have long advocated as an effective way of allocating scarce roadway capacity to the highest valued users (Pigou, 1920; Vickrey, 1969; Walters, 1961).

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 179–206 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09008-0

179

180

ELENA SAFIROVA ET AL.

Yet, with a few exceptions, congestion tolls have been vigorously opposed by most local elected officials in the United States and by the public.1 Amongst the major public concerns about congestion pricing are high implementation costs, privacy issues, and distributional effects. The development of electronic debiting from smart cards has made implementation easier, and it may help alleviate fears about privacy, as there is no need to record the license plates of vehicles with smart cards. Distributional effects of congestion tolls, however, remain a major obstacle. Since everyone pays the same charge regardless of income, there are concerns that low-income motorists will suffer disproportionately. Compensation of potential losers, by appropriate spending or other recycling of toll revenues, may crucially determine the overall political feasibility of congestion pricing. This has been confirmed by various surveys, where support for congestion pricing increased with explicit proposals for using the revenues in, for example, other tax reductions or investment in roads or public transport (Harrington et al., 2001; Ison, 2000; Jones, 1991; RAC Foundation for Motoring, 2003). Designing such compensation packages requires information about the initial impact of tolls on different groups. Prior literature has emphasised that the effect of congestion pricing on vertical equity is governed by two factors working in opposite directions (Small, 1983): higher income individuals have greater propensity to use automobiles and therefore will pay more in tax dollars, but they also benefit disproportionately from congestion relief, not least because they have higher values of travel time. Indeed under certain conditions higher income groups may on balance be better off (i.e. they value the time savings more than the tax payments) while lower income groups are made worse off (Cohen, 1987; Small, 1983). Nonetheless, a sufficiently progressive recycling of toll revenues could ensure that all income groups benefit overall (Small, 1992a). Previous studies of the distributional effects of congestion pricing have mainly been confined to highly aggregate models where agents might choose among different modes, but travel on a single route (Anderson & Mohring’s 1997 model of the Twin Cities is a notable exception). In contrast, this paper analyses distributional effects using a detailed transport network model, representing metropolitan Washington DC. The model, which disaggregates forty travel zones and four income groups, offers several advantages over more aggregated models. First, congestion pricing on a subset of links within a network will induce an array of substitution effects as people switch to unpriced routes, public transport, travel off-peak, etc. Consequently, the costs of travelling elsewhere can change; for example, congestion can increase on alternative roads but may fall in the

Welfare and Distributional Effects of Road Pricing Schemes

181

downtown core if fewer people are driving in. The model used in this study is ideally suited to capturing these indirect effects, which can be important, yet have been neglected from prior incidence studies.2 Second, the model reveals the crucial importance of pre-existing congestion policies for the welfare and distributional effects of congestion pricing. Major freeways in the DC area currently have restricted high-occupancy vehicle (HOV) lanes during peak periods; allowing single occupancy vehicles to use these lanes in exchange for a fee amounts to opening up scarce road capacity. Many motorists benefit from this policy, including those who willingly pay the fee and others who benefit from reduced congestion on adjacent freeway lanes. Policies to convert HOV lanes into HOT (high occupancy/toll) lanes are gaining in political acceptability, yet there has been little formal analysis of their welfare and distributional effects.3 Third, the model can be used to analyse horizontal distributional effects, particularly the winners and losers across different regional zones. Horizontal equity effects have been absent from previous studies, but they are important for political feasibility as the greater the number of localities that stand to benefit from congestion pricing, the greater the likelihood of assembling a coalition of winners large enough to enact the measure. The focus of this chapter is on three main policies: (i) the conversion of existing HOV lanes into HOT lanes; (ii) a “limited pricing” policy where charges are applied to all lanes of freeway segments that currently have HOV lanes; and (iii) a “comprehensive pricing” policy where all lanes of all segments on all major freeways are covered by pricing. In principle a more encompassing policy would also cover city streets, though this is not yet practical with current technology. The results can be summarised as follows. First, the aggregate annual social welfare gain from comprehensive freeway pricing is estimated at $220 million; however, 77 and 83% of this gain are achieved under the HOT lane and limited freeway pricing policies respectively. Thus, just allowing single occupant vehicles to pay to use HOV lanes achieves more than three quarters of the gains from more comprehensive pricing. Pricing of additional lanes and freeway segments reduces congestion on those links further, but the gains are partly offset by added congestion elsewhere in the network, primarily on unpriced arterials. Second, even when toll revenues are excluded, all household income quartiles are better off under the HOT lane policy, while for the most part all income groups are made worse off under other pricing policies. People only pay HOT lane tolls if they are more than compensated by the value of travel time savings, while drivers remaining on adjacent unpriced lanes benefit from reduced congestion. Under other policies drivers must pay tolls, regardless of whether they are compensated

182

ELENA SAFIROVA ET AL.

by time savings or not, unless they make (costly) adjustments in travel behaviour, such as driving on other routes or at off-peak hours. Across jurisdictions or zones within the region, the coalition of winners is much broader under the HOT lane policy; to build as wide a geographical coalition of winners under the other policies would require a complex system of inter-jurisdictional transfers. These considerations suggest that political opposition should be less of an obstacle for HOT lanes than for more comprehensive pricing. Third, in relative terms the regressive tendencies of congestion pricing are less severe under HOT lanes: all income groups benefit under this policy (prior to revenue recycling), and the top-income quartile gains less as a fraction of income than households in the third-income quartile, who have longer average commute times. Under other pricing policies, welfare losses as a proportion of income are significantly larger for lower income groups. In short, the results of this chapter suggest a strong case on the grounds of efficiency, equity, and political feasibility, for converting existing HOV lanes into HOT lanes as a first step in addressing the region’s congestion problems. An initial jump towards more comprehensive pricing raises some troublesome issues on equity and practical grounds, unless toll revenues can be credibly used in appropriate compensation schemes, for relatively modest additional welfare gains. It should be emphasised, however, that much of the benefit from HOT lanes is due to their undoing a large inefficiency created by pre-existing HOV lanes; that is, under-use of the premium lane and over-use of adjacent, unrestricted lanes. Single lane tolls imposed on freeways with no prior HOV restrictions are far less attractive on welfare and distributional grounds, not least because they increase rather than reduce congestion on adjacent lanes (Small & Yan, 2001). Moreover, policy results should be taken with caution until a number of issues discussed below have been explored in future modelling efforts. These include fine-tuning prices to reflect real-time traffic flows within peak periods and across different links, incorporating non-recurrent congestion (e.g. from accidents), analysing how pricing policies interact with the broader fiscal system under different scenarios for the use of toll revenues and taking into account changes in land use pattern caused by road pricing. The rest of the chapter is organised as follows. Section 2 provides a heuristic framework for understanding some basic determinants of the distributional effects of road pricing. Section 3 briefly describes the structure and calibration of the computational model. Section 4 discusses who currently bears the burden of congestion in the Washington DC region. Sections 5–7 present the results from the policy simulations. The final section concludes and discusses model limitations.

Welfare and Distributional Effects of Road Pricing Schemes

183

2. CONCEPTUAL FRAMEWORK This section discusses welfare and vertical equity effects of road pricing policies in single and multiple route settings, and comments on the significance of pre-existing HOV lanes and horizontal equity. 2.1. Single Route Framework Consider a highly simplified model of automobile travel on a single route where individuals differ according to whether they have low (L) or high (H) income. Denote household income by Ii where i = L, H and I H > I L . Individual mileage is Mi , where M H > M L as auto ownership and use tend to increase with income.4 The initial “full” cost per mile driven (in dollars) for household i is f i = c + ␲vi . Here c denotes gasoline and other money costs per mile; ␲ is the average time it takes to drive a mile and vi is the value of travel time, therefore ␲vi is the per-mile time cost in dollars. High-income agents have greater willingness to pay for time savings (Mackie et al., 2001; Wardman, 2001); therefore vH > vL . Individuals perceive ␲ as exogenous, though it increases with the total volume of traffic as congestion slows travel speeds. Individuals as a group drive up to the point where the marginal (private) benefit, or height of their group demand curve for vehicle miles, equals the full cost per mile; initial mileage for group i is therefore M i0 in Fig. 1. Consumer surplus from driving, i.e. benefits of travel minus full travel costs, is triangle abc. The change in consumer surplus in response to pricing polices measures the group’s change in welfare. Suppose a toll of ␶ is levied for every mile driven by households and that the government keeps the revenues rather than recycling them. The demand for driving falls to M i1 in Fig. 1, as individuals car pool, switch to public transport, reduce trip frequency, etc. and congestion will be reduced; ␲ < 0 denotes the change in time cost per mile. Thus, the change in the full cost of driving is ␶ + ␲vi per mile. The welfare loss for group i is the shaded trapezoid debc in Fig. 1; it consists of a first-order income loss from the full cost increase (rectangle dhbc), less a second order gain from reducing driving to its new (privately) optimal level (triangle ehb, equal to cost savings net of forgone driving benefits). As a proportion of income, the welfare loss (assuming linear demand) can be described by Eq. (1):   (␶ + ␲vi )M i0 1 M i 1− (1) 2 M i0 Ii where M i = M i0 − M i1 .

184

ELENA SAFIROVA ET AL.

Fig. 1. Welfare Effect of Congestion Tax.

If the proportionate reduction in driving M i /M i0 is small, two main factors determine congestion toll incidence. The first factor is the increase in the full cost of driving per mile, ␶ + ␲vi , which is smaller for high-income agents as they have a greater value of time. Indeed the full travel cost could fall for high-income groups, while that for low-income groups necessarily increases.5 The second is the ratio of initial mileage to income, M i0 /I i . Normally a policy is thought of as progressive/regressive if it imposes a burden that is a larger/smaller fraction of income for higher income groups. In this example, the welfare loss as a proportion of income is larger for higher income households only under stringent conditions. These are that the change in the full cost for the high income group is positive and that mileage relative to income is not only larger for the high-income group, but by enough to outweigh the effect of their smaller full cost per mile increase. Opponents of congestion pricing often claim that it will drive poor people off the roads, that is, M i /M i0 is substantial for low-income groups. From Fig. 1 it is easy to see that, for a given increase in the full cost of driving and a given initial mileage, the greater the reduction in mileage by the low-income group, the smaller must be their welfare loss. This is because the group demand curve must be flatter, implying a larger second order gain from reduced driving (triangle ehb). In other words, if a large substitution off tolled roads by low-income drivers occurs this is not evidence of a large reduction in their welfare per se; instead it

Welfare and Distributional Effects of Road Pricing Schemes

185

likely reflects indifference on their part between driving on the tolled road and other travel options, or not travelling.

2.2. Multiple Routes Now suppose that individuals travel on two routes, denoted by superscript A and B, both of which are congested but only one of which, route A, is subject to a toll. A and B might be substitute roads, such as different lanes on the same freeway, or spatially separated but competing routes into town. They could also be “complements”; for example B might be roads feeding into and out of the freeway. Suppose that, as motorists are diverted off A in response to the toll, congestion increases on B (␲B > 0). This reduces consumer surplus for group i by the shaded trapezoid in Fig. 2 (if congestion falls there would be a gain in surplus). Combining effects from routes A and B (i.e. the two trapezoids in Figs 1 and 2) the welfare loss to income ratio for group i is:6     (␶ + ␲A vi )M iA ␲B vi M iB 1 M iA 1 M iB 0 0 1− + 1− (2) 2 M iA 2 M iB Ii Ii 0 0

Fig. 2. Welfare Effect of Change in Congestion on Related Route.

186

ELENA SAFIROVA ET AL.

The second component in Eq. (2) may well be larger for low than high-income groups, as they are more likely to avoid driving on toll roads compared with wealthy motorists. If so, an analysis that neglected welfare effects arising from changes in congestion on non-tolled routes might seriously understate the overall burden on lower-income groups. It should also be noted that the previous point about the welfare loss for the low-income group diminishing the greater their willingness to substitute away from the tolled road is less clear when congestion costs increase on alternative routes. The formula in (2) readily generalises to the case of multiple tolled roads, denoted j = 1 . . . N, and multiple non-tolled roads, denoted k = 1 . . . M:     ij N M   (␶j + ␲j vi )M 0 ␲k vi M ik 1 M ij 1 M ik 0 1− + 1− (3) 2 M ij 2 M ik Ii Ii 0 j=1

0

k=1

It is difficult to judge how important indirect effects on other routes will be for the welfare effects of road pricing (i.e. the relative magnitude of the second summation term in Eq. (3)) without detailed modelling of the particular policy and the transport network in question. It will depend on how comprehensive congestion pricing is (a broader policy reduces substitution possibilities onto other roads), the extent to which people re-schedule trips to off-peak period rather than continuing to drive at peak-period, and the extent of initial congestion on other routes.

2.3. Importance of Pre-Existing HOV Lanes Currently, major stretches of freeway lanes going into the District of Columbia are restricted to high-occupancy vehicles. As they are off-limits to single occupancy vehicles (SOVs), these lanes are underused relative to adjacent freeway lanes. Allowing SOVs to use existing HOV lanes in exchange for a fee effectively would open up underutilised road capacity. It should be noted that every SOV driver that switches to the premium lane must value the travel time savings by more than the fee; hence their welfare increases, even with no compensation for the tolls that they pay. Moreover, the diversion of drivers onto (rather than away from) the priced lane reduces rather than increases congestion on competing routes.

2.4. Horizontal Distribution Horizontal distribution refers to the welfare impacts of policies across groups distinguished by some attribute other than income. The focus here is mostly on the

Welfare and Distributional Effects of Road Pricing Schemes

187

geographical dimension; that is, how the welfare of households living in different regional zones may be differentially affected by pricing policies. One zone may have a congested freeway running through it, while in another the only links might be arterial roads. Thus, a policy that significantly reduced freeway congestion may benefit the first zone, while the resulting traffic diversion may harm the second zone. On the other hand, zones in the downtown core might benefit from reduced traffic coming in using priced suburban freeways. Identifying which zones gain and lose is important for designing inter-jurisdictional compensation packages that might widen the coalition of support for road pricing.7

3. DESCRIPTION OF THE WASHINGTON-START MODEL The Washington-START model distinguishes forty travel zones. Each zone contains three stylised links (inbound, outbound, and circumferential) that aggregate arterials and side streets; the model also incorporates various “special links” which represent freeway segments and bridges.8 Six main corridors, I–270, I–95, and US–50 in Maryland and I–66, I–95 and US–267 in Northern Virginia, connect the outer suburbs to the central region within the circular I–495/I–95, known as the Beltway, and shown in Fig. 3. Existing HOV lanes on these freeways at peak periods are taken into account.9 The rail network combines the Washington Metro-rail system and the two suburban light rail systems, MARC and VRE. Bus travel is represented by a stylised route network, with bus accessibility in any zone determined by the density of stops, frequency of service, and reported bus travel times. Public transport crowding costs and parking search costs are included in the model; these are obtained from functions relating time penalties to excess demand for public transport and parking capacity. The value of time spent waiting for bus/rail and searching for parking is greater than the value of in-vehicle time. On the demand side households are aggregated into four income groups. Five trip purposes, in addition to freight, are distinguished: home-based trips either originate or terminate at home and are classified as commuting to work, shopping, or other (e.g. recreation) while non-home-based trips are distinguished between work-related and non-work related. There are four travel modes, including single-occupancy vehicle (SOV), high-occupancy vehicle (HOV), bus/rail and walk/bike. And there are three times of day – morning peak, afternoon peak, and off-peak (weekend travel is excluded). START takes the distribution of households by demographic segment and residential location as given. Travel decision-making is modelled as a nested

188

ELENA SAFIROVA ET AL.

Fig. 3. START Modelling Region with all Special Links.

logit tree; in successive nests, households choose whether to take a trip, then destination, mode, time of day and route. Utility functions at each nest are linear in full travel costs, which combine time and money costs. The value of time is a fraction of the driver’s wage rate, with a higher rate attributed to unpleasant tasks of waiting or travelling in crowded conditions. It should be emphasised that the model is well-suited to tackle issues related to HOV lanes since solo driving and carpooling are choice variables for an individual traveller and therefore aggregate carpooling is endogenous as well. The model was calibrated to the year 2000 by aggregating trip data (including freight) from the Metropolitan Washington Council of Governments (COG) Version 1 transport planning model, which disaggregates over 2100 travel zones.

Welfare and Distributional Effects of Road Pricing Schemes

189

The number of households from different income groups living and working in different zones was estimated using data from the Census Transportation Planning Package and 1994 Travel Survey; from this total trips between any given origindestination pair were allocated to different household groups. Data on wages and price indices were obtained from the Census and Bureau of Labor Statistics. Trip times on each link were computed from speed/flow curves generated from runs of the COG model and validated against estimates of rush-hour speeds developed from analysis of aerial photography (Council of Governments, 1999). Travel demand response parameters were chosen to satisfy the hierarchical structure of the logit model and to be largely consistent with empirical literature. In particular, computed fuel price elasticities of vehicle-miles travelled varied across trip purposes between 0.013 for non-home-based non-work-related and 0.055 for home-based shopping. It should be noted that those elasticity values are not model parameters, but the results obtained in model runs. Therefore, they reflect not only the direct effect of increase in fuel price, but also the secondary effects related to reduced traffic congestion. Although those values seem to be on the low end of the reported spectrum (de Jong & Gunn, 2001; Harvey & Deakin, 1998; Johansson & Schipper, 1997), the underlying model parameters are in line with empirical evidence. Washington-START offers an attractive compromise between the highly disaggregated planning models typically used by metropolitan planning organisations, such as the COG model for Washington DC, and economic models which do not disaggregate different travel zones, such as the Transport, Energy, and Environment (TRENEN) model that has been developed for Brussels and other European cities (De Borger & Proost, 2001). Unlike the COG model, Washington-START is rigorously grounded in household optimisation, computes welfare measures that take into account behavioural responses to policy changes, and has relatively quick run times enabling a wide range of policy simulations and sensitivity analysis.10 In addition, unlike the TRENEN model, the Washington-START model can be used to examine the impact of policies across different zones within the region.

4. CURRENT DISTRIBUTIONAL BURDEN OF CONGESTION The metropolitan Washington DC area is consistently ranked as having some of the worst traffic congestion in the United States, behind only Los Angeles and San Francisco. Daily backups have now spread to vast stretches of the Washington area highway system such as the I–95 in Northern Virginia and the I–495 Beltway, where ten years ago traffic was relatively free flowing (Council of Governments,

190

ELENA SAFIROVA ET AL.

Table 1. Trips by Mode. Trip Purpose

SOV

HOV

Public Transport

Walk/Bicycle

All 000s per day %

10,300 45

10,337 45

591 3

1,683 7

Work only 000s per day %

3,722 71

896 17

439 8

173 3

Ending in DC 000s per day %

247 38

210 32

142 22

49 8

1999). This section briefly comments on how the costs of congestion are currently borne by different household groups and regional zones.

4.1. Characteristics of the Transport System Washington DC has a fairly well developed public transport system, including a metro-rail and a bus network. However, it is predominantly located in DC proper, with little public transport in the outer suburbs. As shown in Table 1, public transport accounts for 22% of trips that terminate in the DC core and only 3% of total trips in the metropolitan region as a whole. It can also be seen in the table that 90% of trips in the region are made in either SOVs or HOVs; moreover, 71% of work trips, which occur at peak periods, are in SOVs. This preponderance of SOV (and HOV) trips, combined with limited capacity on the region’s road network, leads to considerable congestion. Table 2 provides some sense of how congestion currently slows travel speeds on selected highway segments, and an aggregate of arterials in North-West DC. Here morning peak refers to the average travel speed between 6.30 and 9.30 AM, the evening peak between 3.30 and 6.30 PM, and off-peak the average at all other times during weekdays. Peak-period speeds are anywhere between 10 and 44% slower than off-peak speeds.

4.2. Burden of Congestion As shown in Table 3, mean income across all households is $69,349, and average income for households in the bottom and top quartiles is $25,998 and $135,591

Welfare and Distributional Effects of Road Pricing Schemes

191

Table 2. Average Speeds on Selected Congested Links (Miles Per Hour). Link

Direction

Morning Peak

Evening Peak

Off-Peak

Wilson Bridge on I–495

W-bound E-bound

38.9 49.5

41.5 38.5

56.4 56.6

I–95 in Prince William Co.

N-bound S-bound

39.1 37.1

32.7 43.4

58.7 58.7

I–395 in Alexandria

N-bound S-bound

36.2 45.7

42.7 37.2

50.4 51.0

28.6

28.2

34.3

DC North-West arterials

respectively.11 Following Small (1992b), the value of travel time is set at 50% of the hourly market wage; values of time vary from $3.2 per hour for quartile 1 to $20.4 per hour for quartile 4. Hours of travel delay are computed by dividing mileage at peak period by the peak and off-peak travel speeds and aggregating the difference over links and households. In absolute terms hours of travel delay increase with income as higher income households own more vehicles and make more trips: annual travel delays are 42.6 and 24.4 hours for average households in the top and bottom income quartile, respectively. Multiplying the time delay by the value of time yields annual delay costs in dollars. These increase by more than in proportion to income: delay costs for the top quartile are $869, about 11 times those for the bottom quartile, while their income is about five times as high. Total delay costs aggregated over all households in the region are $1.17 billion per year; of these costs 55.7% are borne by the top income quartile and 4.1% by the bottom quartile. These estimates are for recurrent congestion only; they exclude delays due to accidents, breakdowns, road maintenance, and bad weather. A widely-cited study Table 3. Distribution of Burden of Congestion (in 2000 Dollars). Quartile

1 2 3 4 All

Average Household Income ($/Year)

Value of Travel Time ($/Hour)

Time Delay (Hours/Year)

Average Delay Cost ($/Year)

Total Delay Cost ($Million /Year)

25,998 45,196 70,620 135,591

3.2 6.7 11.3 20.4

24.4 28.2 39.7 42.6

78 189 449 869

48 116 355 653

69,349

10.4

38.1

396

1,172

Share in Aggregate Delay Cost (%) 4.1 9.9 30.3 55.7 100

192

ELENA SAFIROVA ET AL.

Table 4. Average Delay by Region and Road Class (in Minutes/Mile).

Entire road network Freeways Arterials

DC Core

Inner Suburbs

Outer Suburbs

0.21 0.17 0.24

0.18 0.14 0.21

0.13 0.11 0.13

by the Texas Transportation Institute (2001) puts total congestion costs for the Washington Metropolitan Area in 2000 at $2.3 billion; roughly half of this is due to non-recurrent congestion, hence their estimates for recurrent congestion costs are similar to the ones presented here.12

4.3. Geographic Distribution of Current Congestion Congestion is predominantly concentrated in the city streets of the inner DC core and surrounding urban highways, with generally decreasing costs of congestion as distance from the downtown increases. As shown in Table 4, the inner core, represented by the area inside the Beltway with around 666,000 households, has an average delay time across peak and off-peak periods of 0.21 minutes per mile. The inner suburbs, with around 728,000 households, and outer suburbs, with around 710,000 households, have average delay times of 0.18 and 0.13 minutes per mile.13 It can also be seen from the table that arterials have higher delay costs per mile than freeways, underscoring the potential costs of policies that divert traffic onto them from freeways.14

5. HOT LANE POLICY 5.1. Policy Overview The first policy considered converts all existing HOV lanes into HOT lanes; a fee of 20 cents per mile is applied to SOVs using the HOT lane during rush hours, while HOVs continue to use restricted lanes free of charge.15 The toll level was chosen by roughly optimising the comprehensive pricing policy described below and, to make policies comparable, applying (approximately) the same rate to other policies. In the sensitivity analysis it is shown that optimised toll rates for the HOT lane and limited pricing policies only moderately increase social welfare. Table 5 illustrates the policy’s effect for a typical special link, a 3.5-mile segment of I–395 in Alexandria (Northern Virginia) during evening rush hour; this stretch

Welfare and Distributional Effects of Road Pricing Schemes

193

Table 5. Effect of HOT Policy on Selected Link. HOV/HOT Lane

Average peak speed (miles/hour) Flow (passenger-car-units/mile/hour)a Average peak trip time (minutes)

Adjacent GPLs

Before

After

Before

After

52.8 1,909 4.0

51.9 3,215 4.1

36.2 10,903 5.9

37.0 10,638 5.7

Notes: Results are for a 3.5-mile stretch of I–395 in Alexandria during evening peak. a Flows are aggregated across two HOV/HOT lanes and across three GPLs.

has two HOV lanes and three general-purpose lanes (GPLs). Initial travel speeds are 52.8 mi/hour on the HOV lanes and 36.2 on the GPLs. By opening up lane capacity to SOV drivers, the HOT lane policy increases traffic flow on the restricted lanes by 68%. Due to the diversion of fee-paying SOV drivers, speed on the GPLs increases from 36.2 to 37 mi/hour; this is a small increase but it applies to a large number of drivers. A typical SOV commuter from zone 31 (Prince William County) to zone 1 (DC) who uses the I–395 freeway and other freeway segments with HOT lanes, saves 3.6 minutes per day (16 hours per year) if she continues to use GPLs. But if she switches to the new HOT lanes she saves 15.2 minutes a day (63 hours per year), in exchange for toll payments of $7.40 per day. A similar commuter who was initially driving in HOV lanes faces an increase in trip time of one minute per day (4 hours per year); thus carpoolers are worse off, but only moderately so.

5.2. Welfare and Distributional Effects As shown in Table 6, the HOT lane policy produces a social welfare gain, aggregated over all households, of $105.9 million per year, before counting the value of toll revenues; thus, it easily passes a cost-benefit test on pure economic efficiency grounds.16 This is not surprising, as people are not coerced to pay the toll; they only pay it if they are more than compensated by the value of reduced travel time. Most of those not paying the toll are either indirectly made better off through reduced congestion, or not affected. Adding the $65 million of annual toll revenues yields an after-revenue social welfare gain of $170.9 million.17 The wealthiest quartile pays 50.3% of total toll revenues collected, and receives 54.7% of the before-revenue social welfare gains, while the bottom quartile pays only 5.2% of the tolls, and receives 2.9% of the before-revenue welfare gains. In absolute terms, households in the wealthiest quartile are easily the biggest winners, gaining $57.9 million annually as a group, while the bottom quartile gains

194

ELENA SAFIROVA ET AL.

Table 6. Welfare Changes by Income Group Under HOT Policy. Quartile

Tolls Paid by Income Group, $000/Year

% of Tolls Paid by Income Group

Welfare Change, $000/Yeara

% of Welfare Change Accruing to Quartile

Welfare Change, % of Income

1 2 3 4

3,412 7,822 21,073 32,728

5.2 12.0 32.4 50.3

3,047 12,172 32,717 57,935

2.9 11.5 30.9 54.7

0.028 0.037 0.050 0.042

Total

65,035

100.0

105,870

100.0

0.045

a Before

counting the value of toll revenues.

$3 million. However, relative to income, the third income quartile benefits the most as they are most likely to live in the outer suburbs and have long commutes; their before-revenue welfare gain is 0.05% of income, while that for the top and bottom quartiles is 0.042 and 0.028% respectively. The most important point from Table 6

Fig. 4. Before-Revenue Welfare Change by Zone Under HOT Policy ($ per Trip).

Welfare and Distributional Effects of Road Pricing Schemes

195

however is that all income groups benefit from HOT lanes prior to any recycling of toll payments. Figure 4 shows before-revenue welfare changes for households living in different regional zones. Zones with or near HOT lanes received the greatest benefit; for example, North-East Prince George’s County (zone 12) benefits the most (7–13 cents per trip), mainly because the US–50 corridor there is highly congested and contains a 24-hour HOV lane that would be converted into a 24-hour HOT lane. Zones that show losses, for example the DC core, suffer increased congestion and parking search costs as the total number of people driving into town increases; however, the losses are small in magnitude (less than 1 cent per trip). Other losing groups include those initially using HOV lanes such as carpoolers and family members travelling together. In a typical example, a commuter using the I–66 HOV lanes to travel from North-West Fairfax County to downtown DC loses by just over two minutes a day, or 8.7 hours per year. Thus, two commuters who are identical except that one carpools on the HOV lane and the other does not, fare quite differently: the carpooler faces increased congestion while the single driver benefits from reduced congestion.

6. LIMITED PRICING 6.1. Policy Overview The second policy considered involves a toll of 22 cents per mile for all users of current HOV lanes, and a toll of seven cents per mile on all GPL segments adjacent to existing HOV lanes.18 The policy combines elements of traditional congestion pricing, i.e. charging people for road use that was previously free, with the opening up of capacity on HOV lanes to new users. It is also “limited” in the sense that it only applies to highway segments that currently have HOV lanes. Table 7 illustrates impacts on travel speeds and flows for the same stretch of I–395 at evening peak as before. Traffic flows on the GPLs fall by around 10%, compared with a 3% reduction under the HOT lane policy, as the new charges on these lanes divert drivers onto other routes; travel speeds on these lanes increase from 36.2 mi/hour to 38.6 mi/hour. On the premium lane traffic flows increase by 44%, noticeably less than the 68% increase under the HOT lane policy: some HOVs no longer use this lane, as they now have to pay. Despite freeway pricing, there is a very slight reduction in traffic on nearby side roads: although some drivers are tolled off the freeway onto streets, this is offset by the diversion of drivers taking advantage of new capacity on the premium lane.

196

ELENA SAFIROVA ET AL.

Table 7. Effect of Limited Pricing on Selected Link. Premium Lane

Average peak speed (miles/hour) Flow (passenger-carunits/mile/hour) Average peak trip time (minutes)

Adjacent GPL

Nearby Side Roads

Before

After

Before

After

Before

After

52.8

52.1

36.2

38.6

28.0

28.1

1,909

2,752

10,903

9,780

18,012

17,778

4.01

4.07

5.85

5.49

7.71

7.69

Note: Results are for a 3.5-mile stretch of I–395 in Alexandria and slightly longer adjacent side roads during evening peak.

An average commuter using GPLs to go from zone 31 (Prince William County) to zone 1 (DC) will save over 4.4 minutes (18 hours per year); if she originally drove on GPL and then switched to the new premium lane, she would save over 20.2 minutes a day (84 hours per year). However, if the toll induced her to drive on side roads rather than the GPL, trip time would increase by 3.9 minutes (16.3 hours per year).

6.2. Welfare and Distributional Effects As shown in Table 8, the before-revenue social welfare change under the limited pricing policy is negative $70 million. Households are worse off in aggregate, mainly because for many drivers on GPL lanes and HOVs on premium lanes, the value of reduced travel time is insufficient to fully compensate them for the tolls. The after-revenue welfare change is plus $182 million, only 7% greater than that under the HOT lane policy. Annual revenues raised are $253 million, almost four times the amount raised under HOT lanes, as this policy charges for additional lanes and HOVs using premium lanes. These findings differ sharply from those in some earlier studies of individual freeways where single lane tolls capture only a minor portion of the welfare gains from pricing that covers all freeway lanes (Liu & MacDonald, 1998; Parry, 2002; Small & Yan, 2001). In these models, which exclude pre-existing HOV lanes, a single lane toll compounds congestion on adjacent GPLs; this problem however is avoided as more freeway lanes are priced. Converting an HOV lane to a HOT lane reduces rather than increases congestion on adjacent GPLs; moreover, unlike in single-freeway models, welfare gains from more comprehensive pricing are offset to some extent in this case by added congestion elsewhere in the network.

Welfare and Distributional Effects of Road Pricing Schemes

197

Table 8. Welfare Changes by Income Group under Limited Pricing. Quartile

Tolls Paid by Income Group, $000/Year

% of Tolls Paid by Income Group

Welfare Change, $000/Yeara

1 2 3 4

23,563 38,233 81,927 108,859

9.3 15.1 32.4 43.1

−26,555 −23,474 −24,301 4,085

37.8 33.4 34.6 −5.8

–0.217 −0.112 −0.057 0.005

Total

252,583

100.0

−70,244

100.0

−0.048

a Before

% of Total Welfare Loss Borne by Quartile

Welfare Change, % of Income

counting the value of toll revenues.

Although the share of the toll burden increases with income, in terms of before-revenue welfare effects, limited pricing is regressive throughout the income distribution. As shown in Table 8, the top quartile gains by $4 million per year, while the other three income quartiles lose, by $23–$26 million per year.

Fig. 5. Before-Revenue Welfare Change by Zone Under Limited Pricing ($ per Trip).

198

ELENA SAFIROVA ET AL.

Welfare losses as a portion of income are greatest for the bottom quartile, mainly because they value time savings the least. The spatial dispersion of before-revenue welfare impacts is more striking under limited pricing than under HOT lanes, as Fig. 5 shows. The few zones that gain under this policy are highly congested and reap the largest benefits from the conversion of the HOV lane to a premium lane, such as the North-East of Prince George’s County (zone 12) which contains US–50. Most other zones suffer a loss due to the burden of the toll and the diversion of traffic onto already congested side roads. Most negatively affected zones are located in the outer suburbs where residents have relatively long commutes.

7. COMPREHENSIVE PRICING 7.1. Policy Overview The final policy considered extends the seven cents per mile toll to all freeway segments throughout the region; tolls for any vehicle using prior HOV lanes remain at 22 cents per mile. This policy covers considerably more of the road network than limited pricing, though it is not fully comprehensive because side streets and arterials remain unpriced. The policy has essentially the same effect on traffic flows and speeds on freeway segments with HOV lanes as limited congestion pricing. On other freeway segments the main effect is some diversion of vehicles onto nearby side roads. Table 9 illustrates this for a four-mile stretch of the Beltway between I–66 and US–267 in Fairfax County during the evening peak. Traffic flows on this Beltway segment decrease 9% while speeds increase from 36.4 to 40.6 mi/hour; travel flows on adjacent side roads increase by 4%, while average speeds decline by 0.4 mi/hour. Drivers who pay the toll benefit from an 11% reduction in travel time on this segment, while those switching to side roads to avoid the toll suffer Table 9. Effect of Comprehensive Pricing on Selected Link. Beltway

Average peak speed (miles/hour) Flow (passenger-car-units/mile/hour) Average peak trip time (minutes)

Nearby Side Roads

Before

After

Before

After

36.4 8,811 6.6

40.6 8,052 5.9

36.3 8,226 9.9

35.9 8,528 10

Note: Results are for the approximately 4-mile stretch of the Beltway between VA–267 and I–66 and 6 miles of adjacent side roads during the evening peak.

Welfare and Distributional Effects of Road Pricing Schemes

199

an increase in travel time of 51%. Due to increased speeds on all special links, an average commuter using a GPL from Prince William County to downtown DC saves nearly five minutes a day (21 hours per year); however, if that commuter switched to side roads to avoid tolls, she would lose 4.3 minutes a day.

7.2. Welfare and Distributional Effects Comprehensive pricing produces an annual after-revenue social welfare gain of $220 million, 29 and 21% larger than under HOT lanes and limited pricing, respectively. As shown in Table 10, toll revenues are $446 million, 76% higher than under limited pricing, and almost seven times revenue under the HOT policy. Comprehensive pricing produces a before-revenue social welfare loss of $225.6 million, more than three times the loss under limited pricing. It can also be seen from the table that all four quartiles suffer before-revenue welfare losses under this policy, but relative to income losses are almost 12 times as large for the bottom quartile as for the top quartile. Thus, the policy is quite regressive, even though the top quartile pays 40% of the total toll revenues. Geographical disparities in before-revenue welfare effects are more pronounced for some zones under comprehensive pricing than under limited pricing, while other zones fare about the same. For example, several zones including non-HOV freeways, such as zone 13, now suffer higher losses as residents must choose between paying a toll on the Beltway or driving on more congested side roads. Several zones on the city outskirts that were affected the most under limited pricing (e.g. zones 30, 31, 29 and 14) show similar levels of per trip welfare losses, but are now affected less relative to other zones. This is because residents of these suburbs now enjoy reduced congestion on the Beltway when they drive in, as Fig. 6 shows. Table 10. Welfare Changes by Income Group Under Comprehensive Pricing. Quartile

Tolls Paid by Income Group, $000/Year

% of Tolls Paid by Income Group

Welfare Change, $000/Yeara

% of Welfare Loss Borne by Quartile

Welfare Change, % of Income

1 2 3 4

47,849 71,771 147,580 178,824

10.7 16.1 33.1 40.1

−55,815 −60,510 −79,009 −30,312

24.7 26.8 35.0 13.4

−0.456 −0.288 −0.187 −0.039

Total

446,026

100.0

−225,646

100.0

−0.155

a Before

counting the value of toll revenues.

200

ELENA SAFIROVA ET AL.

Fig. 6. Before-Revenue Welfare Change by Zone Under Comprehensive Pricing ($ per Trip).

7.3. Sensitivity Analysis Loosely speaking, the distributional impacts described above change in proportion as all tolls are varied by the same proportion. Here aggregate social welfare effects are briefly discussed. As shown in the middle two rows of Table 11, increasing and decreasing all tolls by 50% moderately reduces welfare gains relative to those in the benchmark. The one exception is the HOT lane where higher tolls increase social welfare, but only moderately. As shown in the bottom row, doubling the toll on GPLs keeping the premium lane toll fixed is welfare-reducing under comprehensive pricing, but moderately increases welfare under limited pricing. This is because average congestion on GPLs affected by the limited pricing policy is greater than average congestion across the (much wider) range of GPLs affected by comprehensive pricing.

Welfare and Distributional Effects of Road Pricing Schemes

201

Table 11. Sensitivity Analysis. Policy Scenario

HOT Lane

Limited Pricing

Comprehensive Pricing

Benchmark

Premium lane toll (cents/mile) Toll on other lanes (cents/mile) Social welfare gain ($ million/year)

20 0 171

22 7 182

22 7 220

High tolls

Premium lane toll (cents/mile) Toll on other lanes (cents/mile) Social welfare gain ($ million/year)

30 0 181

33 10.5 180

33 10.5 216

Low tolls

Premium lane toll (cents/mile) Toll on other lanes (cents/mile) Social welfare gain ($ million/year)

10 0 149

11 3.5 162

11 3.5 193

Reduced toll differential

Premium lane toll (cents/mile) Toll on other lanes (cents/mile) Social welfare gain ($ million/year)

22 15 191

22 15 188

8. CONCLUSIONS The inability of enhanced road and public transport capacity to keep pace with relentless growth in vehicle miles has led to urban centres becoming ever more congested. Recent experiments with high-occupancy vehicle (HOV) lanes have failed to seriously dent congestion by encouraging the hoped-for expansion of carpooling (Poole & Orski, 1999). Whatever the economic merits of road pricing, in the past policy stakeholders have regarded it as impractical, because of the apparent unwillingness of motorists to pay for something that they have previously used for free. However HOV lanes, which have disappointed their many advocates, may end up being a Trojan horse for congestion tolls, at least in a limited form. There is no coercion involved in opening HOV lanes up to single occupant vehicles in exchange for a fee: motorists can continue to use adjacent freeway lanes for free if the value of time savings is insufficient to compensate them for the toll. The policy creates a broad coalition of winners, both across different income groups, and across local jurisdictions, even prior to recycling of toll revenues. And according to the results from the model presented in this chapter for Washington DC, the social welfare gain from HOT lanes can amount to around three-quarters of the welfare gains from substantially more comprehensive road pricing. A major caveat is that the results reflect specific features of the Washington DC metropolitan area such as geography of income distribution, relative importance of public transport, level of carpooling and degree of utilisation of HOV lanes.

202

ELENA SAFIROVA ET AL.

It would be interesting to study whether the quantitative results reported here apply broadly to other urban centres or not. Nonetheless the qualitative result that HOT lanes offer travellers more choices and therefore can produce better distributional outcomes compared with more general congestion pricing should be robust. A number of other limitations to the analysis deserve mention. For computational reasons, the tolls are exogenously specified, they are constant across entire peak periods and they are the same for different freeways; more precisely targeted tolls that varied across freeways and with real-time traffic flows within the peak period would yield greater welfare gains. Only recurrent congestion is considered; welfare gains from improving travel flows would be significantly larger if reduced non-recurrent congestion (due to accidents, bad weather, breakdowns, etc.) were also taken into account. Welfare gains would also be larger if additional heterogeneity in values of time due to factors beyond differences in wages were also accounted for (Small & Yan, 2001; Small et al., 2002). The interactions between congestion charges and other motor vehicle externalities, such as pollution and accidents, are beyond the scope of this paper. Parry and Bento (2002) discuss the welfare effects of congestion charges in the framework of other distortions within the transport system. The interactions between policies and the broader fiscal system are also ignored. If policy revenues are used to improve economic efficiency, for example by reducing income taxes that distort labour and capital markets, welfare gains could be substantially higher than computed here. But there is an offsetting effect that should be included in a more general analysis: by raising the costs of commuting to work and discouraging labour force participation, congestion pricing can also reduce efficiency in the labour market, which is badly distorted at the margin by the tax system (Mayeres, 2001; Parry & Bento, 2001). Finally, only the direct short-term effect of congestion policies is considered; over the long haul distributional incidence will change as people respond to road charges by changing residential location or place of work and demanding higher wages or travel allowances from employers to compensate for tolls (Boyd, 1976).

NOTES 1. To date, there are few examples of congestion pricing schemes in the U.S. Some of these operate on Route 91 linking Riverside to Orange County in Southern California, I–15 in San Diego, I–10 in Houston, and two bridges in Lee County, Florida. Pricing schemes however are under discussion in a number of other urban centres. Chapter 12 of this volume discusses congestion pricing projects in the U.S.

Welfare and Distributional Effects of Road Pricing Schemes

203

2. Previous studies have shown that substitution into alternative unpriced (and congested) routes can considerably diminish the efficiency gains from, and optimal levels of, congestion tolls applying to a single freeway, or lane on a freeway (Parry, 2002; Small & Yan, 2001; Van Dender, 2001; Verhoef, 2002). However, the distributional implications of these substitution effects have not been explored, nor has there been much attention paid to reduced congestion on complementary roads that feed into or out of the priced freeway. 3. HOT lanes have recently been endorsed by the Federal Highway Administration under its Value Pricing Pilot Program, discussed in Chapter 12 of this volume. They have been formally modelled by Dahlgren (1999) and Kim (2000); however their focus was on travel delays modelled within a one-road bottleneck framework, rather than on welfare and distributional effects. HOT lanes are also discussed in Fielding and Klein (1993) and Poole and Orski (1999, 2002). 4. Estimates of the elasticity of vehicle miles with respect to income are typically around 0.35−0.8, or higher (Pickrell & Schimek, 1997). 5. The full cost cannot fall for all income groups. Congestion tolls reduce aggregate vehicle miles and lower congestion; this implies that at least some individuals must face an increase in full driving costs and that the reduction in their mileage must more than offset any increased mileage for individuals whose full driving cost falls. 6. For clarity, shifts in demand curves in Figs 1 and 2 stemming from cross-price effects are not shown; nonetheless, the formula in (2) is still correct, so long as the Ms are interpreted as quantity changes taking into account price changes on both routes. 7. Horizontal distribution has received very little attention in prior literature. Spatially disaggregated models usually assume that agents with identical preferences and income would relocate until their welfare is equated across locations; differential welfare impacts do not therefore persist in these models over the very long run. And theoretical models often focus on comprehensive congestion pricing of all links within a network; in practice it is very difficult to price streets and arterials with frequent intersections, hence pricing will create differential impacts on zones with different mixes of freeways, arterials, and streets. 8. The START (STrategic and Regional Transport) modelling suite was developed by MVA Consultancy and has been applied to a range of urban centres in the United Kingdom. 9. The only exception is US–50 operating 24-hour HOV lane which is modelled off-peak as well. 10. Traditional welfare measures in the literature incorporate only changes in full costs associated with travel choices (e.g. Jara-D´ıaz & Farah, 1988, Section 3.2). The measure used here is more comprehensive as it also includes utility changes from modal, time of the day and route adjustments. 11. Household income exceeds annual wages per worker due to non-labour income and secondary workers in many households. It should also be noted that the income quartiles are approximate – the percentage of households in quartiles 1–4 is: 22, 23, 28, and 27% respectively. 12. However there are some methodological differences. The Texas Transportation Institute study uses the difference between observed speeds and a free-flow estimate of potential speeds, and they include extra fuel consumption costs; both these factors raise estimated congestion costs relative to the estimates reported in this chapter. On the other hand, they compute costs for a smaller geographical region with 3.2 million people whereas the START modelling region encompasses around 5 million people.

204

ELENA SAFIROVA ET AL.

13. It should be noted that those delay times are experienced by all travellers using a particular part of the transport network and not necessarily by residents of a particular regional zone. 14. Marginal congestion costs for individual links are not reported, as they are difficult to compute in a network model. This is because adding an extra vehicle on one link affects congestion and travel costs on that link relative to other links, leading to a reallocation of driving that affects congestion on other links. However, as noted in the sensitivity analysis below, the exogenously specified tolls considered here do not appear to be too far off their second-best optimal levels. 15. The policy is implemented by lowering the effective price on SOVs in HOV lanes from a very high level used to deter most solo drivers (currently, a small number of SOVs illegally use HOV lanes and these drivers are included in the baseline calibration). 16. Welfare gains reflect the excess value of time savings over toll payments, and account for a range of behavioural responses to take advantage of reduced travel costs (e.g. trip re-scheduling, route substitution, reduced carpooling). Losses to people originally using the HOV lane, and from added congestion in the downtown core as more people are encouraged to drive, are also included in the welfare measure. 17. The HOT lane policy discourages use of public transportation and therefore also decreases fare collection. However, lost revenue ($1.2 million) is much smaller than the road toll collection and is ignored below. In principle, an alternative measure of aggregate social welfare could be obtained using distributional weights derived from a social welfare function (e.g. Mayeres & Proost, 1997). This issue is left aside however, given the difficulty of assessing society’s preference for redistribution. For example, it is possible to infer a set of distributional weights by exploring how much economic efficiency the government is willing to sacrifice to have a progressive, distortionary income tax system. At the same time, the tax system is at least partly determined by the interplay of interest groups, rather than purely benevolent government behaviour, implying that such estimates may be an unreliable indicator of society’s true preferences. 18. Previous analytical work shows that optimal tolls differ across freeway lanes in order to sort out drivers with high and low time costs across faster and slower lanes (Small & Yan, 2001).

ACKNOWLEDGMENTS The authors are very grateful to Stef Proost, Georgina Santos, Ken Small and two referees for excellent comments and suggestions.

REFERENCES Anderson, D., & Mohring, H. (1997). Congestion costs and congestion pricing. In: D. L. Greene, D. W. Jones & M. A. Delucchi (Eds), The Full Costs and Benefits of Transportation Contributions, Theory and Measurement. New York: Springer.

Welfare and Distributional Effects of Road Pricing Schemes

205

Boyd, J. H. (1976). Benefits and costs of urban transportation: He who is inelastic receiveth and other parables. Transportation Research Forum Proceedings, 17, 290–297. Cohen, Y. (1987). Commuter welfare under peak period congestion: Who gains and who loses? International Journal of Transport Economics, 14, 239–266. Council of Governments (1999). Traffic quality on the metropolitan Washington area freeway system. Prepared by Skycomp, Inc, Columbia, MD, for the Washington Metropolitan Council of Governments. Dahlgren, J. W. (1999, June). High occupancy vehicle/toll lanes: How do they operate and where do they make sense? California Partners for Advanced Transit and Highways, UCB-ITSPWP-99-8. De Borger B., & Proost, S. (Eds) (2001). Reforming transport pricing in the European Union – A modelling approach. In: Transport Economics, Management and Policy Series. Cheltenham, UK: Edward Elgar. de Jong, G., & Gunn, H. (2001). Recent evidence on car cost and time elasticities of travel demand in Europe. Journal of Transport Economics and Policy, 35, 137–160. Fielding, G. J., & Klein, D. B. (1993). High occupancy/toll lanes. Policy Study No. 170, Los Angeles, Reason Foundation. Harrington, W., Krupnick, A. J., & Alberini, A. (2001). Overcoming public aversion to congestion pricing. Transportation Research, A, 35, 87–105. Harvey, G., & Deakin, E. (1998). The STEP analysis package: Description and application examples. In: USEPA. Technical Methods for Analyzing Pricing Measures to Reduce Transportation Emissions, USEPA Report #231-R-98-006. Ison, S. (2000). Local authority and academic attitudes to urban road pricing: A U.K. perspective. Transport Policy, 7, 269–277. Jara-D´ıaz, S. R., & Farah, M. (1988). Valuation of user’s benefits in transportation systems. Transport Reviews, l8, 197–218. Johansson, O., & Schipper, L. (1997). Measuring the long-run fuel demand for cars. Journal of Transport Economics and Policy, 31, 277–292. Jones, P. (1991). Gaining public support for road pricing through a package approach. Traffic Engineering and Control, 32, 194–196. Kim, J. E. (2000). HOT Lanes: A Comparative evaluation of costs, benefits and performance. Unpublished doctoral dissertation, University of California at Los Angeles. Liu, L. N., & MacDonald, J. F. (1998). Efficient congestion tolls in the presence of unpriced congestion: A peak and off-peak simulation model. Journal of Urban Economics, 44, 352–366. Mackie, P. J., Jara-D´ıaz, S., & Fowkes, A. S. (2001). The value of travel time savings in evaluation. Transportation Research, E, 37, 91–106. Mayeres, I. (2001). Equity and transport policy reform. ETE Working Paper 2001-14, Center for Economic Studies, Katholieke Universiteit Leuven. Mayeres, I., & Proost, S. (1997). Optimal tax and public investment rules for congestion type of externalities. Scandinavian Journal of Economics, 99, 261–279. National Capital Region Transportation Planning Board (2002). Update to the financially constrained long range plan for the national capital region. Washington, DC: Washington Metropolitan Council of Governments. Parry, I. W. H. (2002). Comparing the efficiency of alternative policies for reducing traffic congestion. Journal of Public Economics, 85, 333–362. Parry, I. W. H., & Bento, A. M. (2001). Revenue recycling and the welfare effects of road pricing. Scandinavian Journal of Economics, 103, 645–671.

206

ELENA SAFIROVA ET AL.

Parry, I. W. H., & Bento, A. M. (2002). Estimating the welfare effect of congestion taxes: The critical importance of other distortions within the transport system. Journal of Urban Economics, 51, 339–365. Pickrell, D., & Schimek, P. (1997). Trends in personal motor vehicle ownership and use: Evidence from the nationwide personal transportation survey. In: U.S. Federal Highway Administration, Proceedings from the Nationwide Personal Transportation Survey Symposium October 29–31, No. 17 of Searching for Solutions: A Policy Discussion Series (pp. 85–127). Washington, DC: U.S. FHWA. Pigou, A. C. (1920). Wealth and welfare. London: Macmillan. Poole, R. W., & Orski, C. K. (1999). Building a case for HOT lanes. Policy Study No. 257. Los Angeles: Reason Public Policy Institute. Poole, R. W., & Orski, C. K. (2002). HOT networks: A new plan for congestion relief and better transit. Policy Study No. 305. Los Angeles: Reason Public Policy Institute. RAC Foundation for Motoring (2003). RAC report on motoring. Making the Most of Britain’s Roads. London. Small, K. A. (1983). The incidence of congestion tolls on urban highways. Journal of Urban Economics, 13, 90–111. Small, K. A. (1992a). Using the revenues from congestion pricing. Transportation, 19, 359–381. Small, K. A. (1992b). Urban Transport Economics, Fundamentals of Pure and Applied Economics, 51. United Kingdom: Harwood Academic Press. Small, K. A., & Yan, J. (2001). The value of value pricing of roads: Second-best pricing and product differentiation. Journal of Urban Economics, 49, 310–336. Small, K. A., Winston, C., & Yan, J. (2002). Uncovering the distribution of motorists’ preferences for travel time and reliability: Implications for road pricing. University of California Transportation Center, Working Paper 546. http://www.uctc.net/papers/546.pdf. Texas Transportation Institute (2001). Urban roadway congestion: Annual report. Texas A&M University, TX. Van Dender, K. (2001). Aspects of congestion pricing for urban transport. Unpublished doctoral dissertation, Katholieke Universiteit Leuven. Verhoef, E. T. (2002). Second-best congestion pricing in general networks: Heuristic algorithms for finding second-best optimal toll levels and toll points. Transportation Research, B, 36, 707–730. Vickrey, W. S. (1969). Congestion theory and transport investment. American Economic Review, 59, 251–260. Walters, A. A. (1961). The theory and measurement of private and social cost of highway congestion. Econometrica, 29, 676–699. Wardman, M. (2001). A review of British evidence on time and service quality valuations. Transportation Research, E, 37, 107–128.

PART II: EVIDENCE

This Page Intentionally Left Blank

9.

TRANSPORT POLICIES IN SINGAPORE

Georgina Santos, Wai Wing Li and Winston T. H. Koh 1. INTRODUCTION Being an island city-state that measures 42 km (26.1 mi) East to West and 23 km (14.3 mi) North to South, Singapore has 3,149 km of roads (1,955 mi) for a population of about 4.2 million people and 707,000 registered motor vehicles in the year 2002. Since gaining independence in 1965, Singapore has enjoyed sustained and rapid economic growth, with an annual average growth rate of about 8% until the mid-1990s. Rising affluence has brought an increase in demand for personal motor transportation. In 1960, the level of motorisation in Singapore was 50 cars per 1000 people, similar to levels in the U.K. and France then, but was substantially higher than other developing countries and Japan, where the figure was 10 cars per 1000 people. By 1995, with a per-capita GDP of S$33,520 at 1995 prices (or US$19,154, at a 2003 exchange rate of US$1 = S$1.79), comparable to the levels in U.K., France and Japan, the level of motorisation in Singapore, at 105 cars per 1000 people, was about a quarter of the levels experienced in these three countries (Willoughby, 2000). Successfully managing congestion was considered an important part of the country’s efforts to attract multinational corporations to locate their operations in South-East Asia in Singapore. Like in other cities, urban congestion not only creates air and noise pollution, but also increases the cost of doing business. This was a serious problem for Singapore in the early 1970s, when traffic during the

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 209–235 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09009-2

209

210

GEORGINA SANTOS ET AL.

morning and evening rush hours in the central business district (CBD) slowed to a crawl of about 20 km per hour (12.5 mi per hour). Recognising the repercussions on foreign direct investments into Singapore, the government tackled the problem by formulating a transport policy, which has continued to evolve to this day. Besides a series of road pricing schemes designed to control traffic congestion, efforts were also made to develop and upgrade the transport infrastructure (e.g. the construction of a mass rapid rail system) to support the country’s economic development. Singapore has adopted an integrated approach to traffic management, with a set of “carrot and stick” policies. The sticks raise cost of car ownership and motoring through the imposition of various “ownership taxes” and “usage taxes.” As of 2003, the ownership taxes include a quota licence fee (levied through an auction scheme), import duties, a basic registration fee and an additional vehicle registration fee (calculated as percentage of car value). The usage taxes include tolls for entering the CBD or using expressways, petrol taxes and parking fees. The carrots to encourage the use of public transport include the development of reliable and convenient bus, rail and taxi services. The policies that Singapore has undertaken in the last thirty years can be summarised as follows:  coordinated traffic management with ownership taxes and usage taxes on vehicles in order to achieve an optimal usage of the road network;  expansion of the road network in order to provide access to every part of the country;  development of a mass transit railway network;  improvement of the quality of buses and bus services;  coordination and integration of bus, rail and taxi services.1 As well as directly reducing congestion, Chapter 6 of this volume points out that road pricing produces a virtuous cycle in shifting motorists to public transport. Shifting people from cars to buses reduces congestion and allows buses to travel faster; with more customers for both buses and trains, the frequency of services increases, resulting in a better service and a further increase in demand, taking more cars off the road. This chapter has three objectives:  to describe the transport policies in place in Singapore with special attention to road pricing;  to analyse their successes and shortcomings;  to draw lessons for other cities around the world.

Transport Policies in Singapore

211

2. OVERVIEW OF POLICIES From 1960 to 1970, the private vehicle population in Singapore roughly doubled, from 70,100 to 142,500. The key factors behind the increase in car ownership were rising household incomes, a housing programme to develop residential districts in the suburban areas away from the CBD, and unreliable public transport. During this period, public buses and taxis increased by about 64%, while the total length of public roads increased by only about 35%. In 1970, the length of major arterial roads was 240 km (150 mi), and had only increased 12% since 1960. With the number of private cars per km of arterial roads rising from 328 in 1961 to 594 in 1970, traffic conditions were deteriorating fast. The railway system was not built until the 1980s, so the only modes of public transport were taxis and buses. Although there were ten bus companies, services were infrequent and schedules were uncoordinated. In 1970 the government directed the bus companies to merge into four and to substantially expand their fleets; in 1973 these four companies were nationalised and merged to form a new bus company, the Singapore Bus Services. This move resulted in swift and significant improvements in both the quality and quantity of public bus services. The national company’s shares were floated on the Singapore stock exchange in 1978 but the company remained strongly regulated. After ten years of studies and cost-benefit analysis of different options, the Singapore government approved the construction of the Mass Rapid Transit (MRT) railway system in 1982. World Bank consultants and a team of economists from Harvard found that the rail construction costs in the cost-benefit analysis carried out by Wilbur Smith and Associates had been underestimated (Phang, 2003). Mohring (1983) estimated that the measurable real rate of return would be less than 2%. Notwithstanding all this, the project went ahead, mainly because the government wanted to attract foreign investments and talent to Singapore. A consistent policy of the Singapore government in the last three decades has been to ensure that Singapore’s urban infrastructure (transport and housing) and public services provide a conducive environment for businesses. It is this consideration that swayed the government to embark on the MRT project. The MRT network has been continuously expanded since it came into operation in 1987, with a system of buses and the Light Rail Transit (LRT)2 serving as feeders to the MRT network. Beginning in the mid-1970s, there was substantial expansion in the strategic road network of arterials and motorways, with a six-fold increase in investments in road construction and upgrades between 1975 and 1980. Road density doubled between 1975 and 1985, and tripled by 1997 (Willoughby, 2001).

212

GEORGINA SANTOS ET AL.

An ownership tax on vehicle ownership, the Additional Registration Fee (ARF) has been in place since the late 1950s, when Singapore was still under British colonial rule. Intended originally as a revenue-raising measure, the ARF is an ad valorem duty on a vehicle’s open market value (OMV)3 payable by buyers of new motor vehicles, in addition to an administrative fee, referred to as the Basic Registration Fee. The publication of a study by Wilbur Smith and Associates (1974), highlighting the dangers of uncontrolled private vehicle ownership (Foo, 2000), alerted the Singapore government of the urgency of reforming traffic management policies. The ARF was expanded in scope to control car ownership, and its rate was steadily raised through the 1970s, reaching 125% in 1978 and 150% in 1980. However, as the ARF rate rose, it also discouraged existing vehicle owners from replacing their cars and encouraged new car buyers to buy used cars. Concerned with a stock of aging vehicles, the government introduced a Preferential Additional Registration Fee (PARF) to counterbalance the disincentives on vehicle renewal, when the applicable ARF rate was raised to 100% in 1975. The purchaser of a new vehicle paid a substantially lower PARF rate if he de-registered an old vehicle (i.e. by exporting or scrapping it) of the same engine category at the time of his new purchase. PARF rates varied according to the engine capacity; for example between December 1975 and October 1983, they ranged from 35% (expressed as a percentage of the required ARF payment) for the smallest engine category (<1000 cc) to 55% for the largest engine category (>3000 cc). Since 1997, the PARF has been amended to a system where the applicable discount is a function of the age of the vehicle to be de-registered. As of 2003, de-registering a vehicle under 5 years old qualifies for a 25% PARF rate and de-registering a vehicle between 9 and 10 years old qualifies for a 50% PARF rate. Vehicles over 10 years old no longer qualify for PARF treatment.4 The Area Licensing Scheme (ALS), essentially a usage charge, was formulated in 1973 and implemented in 1975, and it was the first of its kind in the world. Vehicles entering the 7 km2 (2.8 mi2 ) restricted zone (RZ), which included the CBD, were required to purchase and display a paper area licence on their windscreen. Enforcement was done manually by enforcement officers standing at the boundaries of the RZ. Offending vehicles were not stopped but issued a summons. Drivers could either appeal to the Traffic Police Department or pay the fine. When the scheme was initially implemented, the restricted hours were from 7.30 to 9.30 AM daily, except on Sundays and public holidays. Three weeks later, the restricted hours were extended to 10.15 AM in order to reduce the excess traffic occurring immediately after 9.30 AM (Chin, 2002), as motorists rescheduled their trips to just before and after the restricted hours and businesses delayed their opening hours to avoid paying the area licence fee. In June 1989,

Transport Policies in Singapore

213

the scheme was further extended to the evening peak, from 4.30 PM to 7 PM on weekdays. The evening period was later cut back by half an hour to 6.30 PM to accommodate requests from residents who lived inside the charging area but worked outside, although this was subsequently extended back to 7 PM, because of increased traffic congestion (Chin, 2002). In January 1994, the operating hours of the scheme were further extended to cover the inter-peak period, from 10.15 AM to 4.30 PM on weekdays and the post-peak period of 10.15 AM to 3 PM on Saturdays. The Saturday charging hours were later cut back to 2 PM. A vehicle displaying the licence could enter and leave the RZ an unlimited number of times during the day. Police cars, ambulances, fire engines and public transport buses were all exempt. In addition to that, in the beginning, taxis, goods vehicles, motorcycles, and passenger cars carrying three or more passengers apart from the driver were all exempt. Two months after the implementation of the ALS, taxis were required to purchase the permit as well. In 1989, motorcycles and goods vehicles, were also required to purchase the permit, together with car-pools (Chin, 2002). No discounts or exemptions were given for residents living inside the RZ. Driving inside the zone without crossing the boundary could be done free of charge. In this sense, the scheme was a cordon system and not an area licence system. “Area Licence” was not a good name, it should have probably been called “Entry Licence” or “Cordon Licence.” The ALS was effective in reducing urban congestion during the morning and evening peak hours. The charges however might have been set too high initially and might have reduced traffic by more than was necessary, leading to under-utilisation of the road network and in the process, shifting congestion to the expressways and non-restricted times. Holland and Watson (1978) for example, find that upon introduction of the ALS, the volume of cars entering the RZ during the restricted period fell by 73% whilst the volume of cars entering the RZ outside the restricted times rose by 23%. Average speeds went up from 19 to 36 km per hour (11.8–22.4 mi per hour), exceeding the government’s optimal flow target (Phang & Toh, 1997). McCarthy and Tay (1993) conclude that the initial price of an ALS licence, which the government set based on its judgement rather than any calculation of the congestion externality, was about 50% above the optimal level. The government appeared to be less concerned with short-term optimisation of road use than with long-term reduction of urban congestion, and raised the daily licence fee for private cars from S$3 to S$4 six months after the introduction of the ALS, and raised it again to S$5 in 1980, where it stayed until it was brought back to S$3 in 1989. This reduction in rates in 1989 was mainly because more vehicles were required to purchase licences (Chin, 2002). Li (1999) estimates the optimal toll using traffic count data and concludes that S$3 was the correct fee for 1990.

214

GEORGINA SANTOS ET AL.

In January 1994, an alternative “part-day” licence was introduced. Thus, vehicles could purchase a licence to enter the RZ throughout the day, or a licence that would only allow them to enter the RZ during the inter-peak period (Chin, 2002). The inter-peak ALS charges were two thirds of the whole day ones. For cars, the fees were S$3 and S$2 respectively. Part-day licences were valid for use between 10.15 AM and 4.30 PM on weekdays, and from 10.15 AM to 3 PM on Saturdays (Foo, 1997a). In 1989, the Parliament of Singapore appointed a Select Committee to examine the transport policy at the time. Several public hearings were held with the majority of views in favour of usage restraint rather than ownership restraint. However, the Select Committee recommended the Vehicle Quota System (VQS), to control vehicle growth5 (Olszewski & Turner, 1993). One of the concerns about vehicle ownership must have been the limited parking provision, mainly the result of high housing density, as providing housing had been a key government policy. The VQS was thus introduced in May 1990 and remains in place today. It is basically an ownership tax applied in addition to the ARF,6 and another policy that was also first of its kind in the world. Prospective vehicle owners are required to purchase a ten-year quota licence called a Certificate of Entitlement (COE). The quota on COEs for each category is announced a year in advance, in May each year. Initially the COEs were allocated in a monthly sealed-bid uniform price auction. Each individual was allowed to bid only once in one of the categories and was required to pay a deposit equal to half his bid. All the successful bidders paid the lowest bid that cleared the quota allocation, which is referred to as the quota premium, and had three months to purchase a new vehicle using the COE, failing which the deposit was forfeited.7 The deposits from unsuccessful bidders were returned. When it was implemented in 1990, the monthly quota was separated into seven categories:  Category 1: Small cars with engine capacity of 1,000 cc and below;  Category 2: Medium-sized cars with engine capacity of 1,001–1,600 cc, and     

taxis; Category 3: Large cars with engine capacity of 1,601–2,000 cc; Category 4: Luxury cars with engine capacity of 2,001 cc and above; Category 5: Goods vehicles and buses; Category 6: Motorcycles and scooters; Category 7: “Open.” A Category 7 COE may be used to purchase any type of motor vehicle.8

In May 1999, after a government review of the quota system, Categories 1 and 2 were merged into one category, as were Categories 3 and 4. The categories were re-named as follows:

Transport Policies in Singapore

    

215

Category A: Cars, 1600 cc and below, and taxis; Category B: Cars, above 2000 cc; Category C: Goods vehicles and buses; Category D: Motorcycles; Category E: “Open,” for registration of all types of vehicles.

Under the VQS, when a vehicle reaches ten years of age, the owner is required to renew the COE for another five or ten years at the prevailing quota premium. No further renewals are allowed thereafter. Since November 1998, the prevailing quota premium is computed as a three-month moving average of the quota premium for the category. Before November 1998, a twelve-month average was used. If the motor vehicle is sold before the expiry of the quota licence, the quota licence will be transferred to the new owner together with the vehicle. A pro-rated rebate of the remaining validity period is paid if the vehicle is deregistered, scrapped or exported. Owners of vehicles over eight years old at the time of introduction of the VQS were given a grace period of two years, after which they were required to purchase a COE at the prevailing quota premium. A discussion on the successive changes to the VQS until today follows in the next section. In June 1995, a paper-based Road Pricing Scheme (RPS), operating in the same way as the ALS, was introduced on an expressway (East Coast Parkway). This was later extended to other expressways. The aim of the RPS was to reduce congestion on the expressways during the office rush hours, and to familiarise Singaporeans with both linear passage tolls and road charging outside the CBD (Goh, 2002). In 1998, Electronic Road Pricing (ERP) replaced both the ALS and RPS. Gantries were installed at all the approach roads to the ERP zone and on the expressways. In contrast to the ALS and RPS, the ERP scheme charges vehicles each time they cross a gantry. The system uses a dedicated short-range radio communication system. Vehicles wishing to enter the ERP zone when charges apply need to install an In-vehicle Unit (IU), which is a radio transponder in which a stored-value smart card is inserted. The fees are deducted from the smart card when a vehicle passes under a gantry at the restricted times, and the IU displays the remaining balance on the smart card. ERP charges vary between different gantries and different times of the day, depending on the level of congestion.9 The charges are reviewed quarterly and for the June and December school holidays to achieve an optimal flow of traffic. Optimal speeds in Singapore have been found to be between 20 and 30 km per hour (12–19 mi per hour) for major roads and 45–65 km per hour (28–40 mi per hour) on expressways.10 The basic principle behind the government’s transport management policies regarding the ARF, petrol taxes, import duties and parking charges has always been to maintain them at as high a level that is politically acceptable. The VQS

216

GEORGINA SANTOS ET AL.

and the ERP, still the only ones of their kinds in the world today, deserve a detailed analysis; hence, the rest of this chapter will concentrate on these two schemes, within the wider framework of Singapore’s integrated transport policy.

3. THE VEHICLE QUOTA SYSTEM Various problems encountered with the use of the ARF to control vehicle ownership led to the introduction of the Vehicle Quota System (VQS), which was based, as explained above, on the recommendations made by the Select Committee appointed by the Parliament. Phang et al. (1996) note that the PARF made used and scrap car prices dependent on the prices of new cars. For example, the Japanese yen appreciated 52% against the Singaporean dollar between 1984 and 1989, raising the import prices of new cars and raising the absolute amount of the PARF discount from scrapping a car. In turn, this directly increased the value of old cars. Furthermore, increases in the ARF rate without concomitant changes in the PARF also raised the price of old cars, since the ARF was set as a percentage of a vehicle’s OMV whilst the PARF was set according to categories with engine sizes. Phang et al. (1996) point out that an inferior car within one of the top engine size categories, bought in the late 1970s, could have been sold for higher than the purchase price ten years later. This was because the old car could be sold to a potential car buyer, who could scrap it when buying an expensive new car in the same engine capacity category, and thus qualify for the PARF. This “appreciating asset” argument was used by car dealers to stimulate sales, and many car buyers bought cheaper makes from Russia and Eastern Europe, expecting the scrap prices to be close to their initial purchase price. Although the ARF increased the cost of vehicle ownership significantly, it proved to be relatively ineffective in controlling the growth of the vehicle population. Phang and Chin (1990) estimate the income elasticity of demand for cars in Singapore to be around 1 and the price elasticity to be around −0.45, indicating that the demand for ownership is relatively inelastic. Between 1980 and 1989, rapid economic growth had produced a 65% increase in the motor vehicle population in Singapore.

3.1. VQS in Theory and Practice In principle, usage taxes (such as ERP charges) can fully internalise congestion externalities, as these taxes directly affect the price of road journeys, which at the margin are the source of the externality. Ownership taxes (such as the quota

Transport Policies in Singapore

217

premiums under VQS and the ARF), on the other hand, control the size of the vehicle population, and aim to indirectly reduce the amount of potential urban congestion. The main justification for ownership taxes, however, must have been that given the investment in MRT and housing, and the scarcity of parking spaces, the government felt necessary to contain vehicle ownership. It is with this consideration in mind that the government decided to augment its plethora of usage taxes with a new ownership tax that was directly set by the market. Furthermore, as Toh (1992) points out, one-time ownership charges are likely to be less painful to consumers than recurrent usage charges. The success of the VQS should therefore be judged by its ability to control the growth rate of the motor vehicle population. Between 1975 and 1989, the average annual motor vehicle population growth rate was 4.4% with a standard deviation of 4.24%. Between 1990 and 2002, after the implementation of the VQS in 1990, the average annual motor vehicle population growth rate went down to 2.83%, with a standard deviation of 2.24%.11 Thus, the VQS was successful in lowering both the rate of vehicle population growth and its volatility. Bhagwati and Srinivasan (1969) show that the social utility-maximising policy to restrict consumption below a certain level is a consumption tax on the good. Since Singapore has no domestic motor vehicle manufacturing industry, and assuming perfect competition in the motor vehicle market, an auction quota is equivalent to an import tariff, hence it is an efficient consumption tax. However, Singapore’s VQS led to some unexpected and undesirable consequences.

3.2. The Open Category The rationale for the open category licence was to allow for flexibility in the mix of the motor vehicle population, according to the demands of the market, so that, for example, if consumer preferences shifted towards larger cars, the open category COEs would be used to purchase large cars. As long as the open category is not too large,12 one would expect that its quota premium would be close to the highestpriced category, otherwise buyers of vehicles in this highest-priced category would be better off bidding in the open category rather than in their own category. Hence the open category COE would tend to be used in the highest-priced Categories 3 and 4, and one would expect the shift in the vehicle mix to be towards luxury cars (Tan, 2001). Using the latest available data (May 1990 to June 2003), Fig. 1 and Table 1 show that the prices of an open category COE were most closely correlated with Categories 3 and 4 prices, the luxury car categories with the highest COE

218

GEORGINA SANTOS ET AL.

Fig. 1. Quota Premiums Under VQS: 1990–2003. Note: In 1999, Categories 1 and 2 were combined to form Category A, while Categories 3 and 4 were combined to form Category B. Category 7 was renamed Category E. The quota premiums for Category 7 (E) were often close to those of Categories 3 and 4 (B). Source: Land Transport Authority, Singapore. www.lta.gov.sg.

prices. Data on the use of the open category COEs are not published, but a May 1990 to April 199913 average shows that the ratios of new registrations to quota level for Categories 1–4 were 95, 113, 195 and 260% respectively. This was possible only through the use of the open category COEs on the higher categories (Tan, 2001). Phang et al. (1996, p. 148) observe that “by 1995, the Mercedes Benz had overtaken Toyota as the most popular make of car registered in Singapore.” The increasing population of luxury cars gradually led to larger quotas for these cars. The shares of cars between Categories 1–4 in 1990 were 15, 67, 14 and 4% respectively. By 1999 these shares had changed to 12, 60, 20 and 8%. This trend continued even after the change to open bidding (which is discussed below). In Table 1. Correlation Coefficient With the Open Category. Cat 1

Cat 2

Cat 3

Cat 4

Cat 5

Cat 6

0.50

0.88

0.96

0.97

0.84

0.64

Source: Land Transport Authority, Singapore. http://www.lta.gov.sg.

Transport Policies in Singapore

219

2002 the share of the combined Category 1 and 2 was 68% and the share of the combined Category 3 and 4 was 32%. Although sub-categorisation created the incentive to purchase larger cars, there was also a shift in consumer preference to larger cars, particularly in light of rising income levels.

3.3. Sub-Categorisation The rationale for having four categories for passenger cars, grouped by engine capacity, was that low-income car buyers would not have to bid against wealthy buyers for a COE, and pay more as a result. Tan (2001) uses a two-category model of high and low-income buyers, to show that low-income buyers will not necessarily be forced out of the market, because even though their inverse demand function is everywhere below that of high income buyers, the supply price of their cars is also lower. Falvey (1979) analyses the case where there is substitution between the categories, i.e. buyers of cheap cars switch their purchase to luxury cars because of the fall in the price of luxury cars relative to cheap cars. In this case, the absence of categorisation would not force low-income buyers out of the market, because they would voluntarily upgrade to luxury cars. One problem with sub-categorisation is allocative efficiency. It is very difficult to determine the correct quota for each category, possibly resulting in excess demand in some categories and excess supply in others, and leading to high prices in the former case and an under-utilisation of the total quota in the latter. This has been the experience of the VQS, which produced wild swings in COEs’ prices. In December 1997 for example, the price of a Category 3 COE was S$64,100; in January 1998 the price was S$50, as the number of bids was lower than the number of quotas. In late 1994, a Category 4 COE topped S$110,000. At the same time, there was a collapse in demand for motorcycles, due to the imposition of strict emissions standards in October 1991, when not many motorcycles in the market at the time met the standards, and improved models took some time to arrive. As a result, for 27 months between December 1991 and February 1994, the price of a motorcycle COE was S$1. The evidence clearly suggests that sub-categorisation failed to ensure equity. Of the 106 auctions held between May 1990 and April 1999, Category 1 premiums ranked the lowest of the four car categories for 81% of the time, Category 2 premiums ranked second lowest for 59% of the time, Category 3 ranked third lowest for 49% of the time, and Category 4 ranked highest for 53% of the time. The fair outcome with the ranking of quota premiums in the order of the categories 1 < 2 < 3 < 4, on the other hand, occurred only 43% of the time. In other words, well over half of the auctions produced an outcome where it was cheaper to buy

220

GEORGINA SANTOS ET AL.

a COE for a larger car than a small car. Furthermore, in 13% of the auctions, Category 1 premiums were higher than Category 4 premiums. The VQS reduced the difference in the relative prices between luxury and cheap cars. The relative price of luxury cars fell and there may have been a substitution effect raising demand for luxury cars. With similar COE prices, it was preferable to purchase an expensive car to preserve the resale value. This also explains, at least in part, the improvement in the general quality and price of the Singaporean vehicle fleet. As can be seen from Fig. 1, the quota premiums of the four quota categories were very close between May 1990 and mid-1993, and have been close again since early 1998. Even when the quota premiums for smaller cars were lower than those for larger cars (the fair outcome of category premium 1 < 2 < 3 < 4), for example in January 1992, the implicit tax rates for Categories 1–4 cars were 119, 123, 75 and 28% respectively, producing a highly regressive system (Tan, 2001). Sub-categorisation is attractive in theory, but in practice, it is difficult to determine the shape and position of the demand curves (Tan, 2001). The quota limits were inappropriate most of the time, the quotas for Categories 1 and 2 should have been higher and the quotas for Categories 3 and 4 should have been lower. In May 1999, the government merged Categories 1 and 2 into Category A and Categories 3 and 4 into Category B14 to alleviate these problems. Although the situation improved, COE prices were still regressive. In 40% of the auctions held between May 1999 and June 2003 for example, the price of a Category A COE was higher than that of a Category B COE. Even after the introduction of open bidding in 2001, as explained in the next section, 36% of the auctions produced outcomes where it was cheaper to buy a COE for a larger car than to buy one for a small car. In reality, the Singaporean government essentially set the quota limit for the very first auction. The auctions that followed were determined by a pre-announced target growth rate common to all categories, the number of de-registrations in each category and the use of the open category COEs. The open category has, over time, led to larger quotas for luxury cars.

3.4. Transferability and Speculation When the VQS started in May 1990, COEs were transferable once before being used to purchase a vehicle. The COE had to be obtained at least one month before the actual purchase of the vehicle. Given the uncertainty surrounding the value of the quota licence at the time of the auction, the secondary market for COEs acted as a mechanism to smooth out demand. As discussed in Krishna and Tan (1997, 1999) and Koh (2003), transferability is an important consideration for social welfare as it provides flexibility in the final allocation of COEs.

Transport Policies in Singapore

221

Licence transferability meant that unsuccessful bidders with a high post-auction valuation could purchase a COE from a successful bidder with a low post-auction valuation. Car dealers who had been unsuccessful bidders could also purchase COEs from the secondary market and honour their sales contracts. Transferability improved liquidity, specially for smaller car dealers with liquidity constraints, who may had decided not to tie up their funds in the deposit for a bid but purchase the COE from the secondary market instead (Koh & Lee, 1994). A transferable COE possesses an option value, giving the holder the choice of using it to buy a car or selling it. Although in principle transferable COEs have a premium over non-transferable COEs, this is not necessarily always the case. The transferability premium depends on how restrictive the quota is relative to demand; a restrictive quota gives a positive transferability premium but a not-so-restrictive quota may give a negative transferability premium. However, the public blamed rapidly raising quota premiums on the secondary COE market, encouraged by reports in local newspapers of the high profits being made by some speculators (the biggest group of speculators were the used-car traders). The government, having initially maintained that transferability was desirable, bowed to public pressure and in October 1991 made Categories 1–4 and Category 6 COEs non-transferable. At the same time the period of validity was increased from three months to six months. Category 5 and 7 COEs remained transferable and valid for three months. Transferability is unlikely to have been the cause of escalating quota premiums. As Fig. 1 shows, COE prices rose even more dramatically after the introduction of non-transferability, though this could be due to demand shifts such as rising income and increased highway capacity. Koh and Lee (1993) regress the quota premium on a dummy variable and other variables such as the range of bids and the ratio of submitted bids to successful bids. They find that the change to non-transferability led to lower quota premiums for Category 1 and higher quota premiums for Categories 3 and 4, and had no significant effect on premiums for Category 2. Tan (2001) uses as regression variables the quota premiums for Categories 1–4 relative to Category 5, which was not made non-transferable. The regression also includes a dummy variable, the quota level relative to Category 5 and the number of bids relative to Category 5. The paper finds that non-transferability lowered the quota premium for Categories 1–4. However, transferability provides for flexibility in the final allocation of licences, and therefore carried economic benefits for auction participants. Most of the participants in the COE auctions have always typically been either car buyers who need a COE or car dealers who are already committed to a sales contract, and neither of these would have any intention of reselling their COEs. When transferability was still possible, they might have changed their mind occasionally if prices in the secondary COE market

222

GEORGINA SANTOS ET AL.

were attractive, but this was unlikely as they would have had to bid again in the following auction. The main reasons for the escalating prices since the COEs were introduced have been rising incomes and speculation. A major speculator could easily corner the market, since in each auction there were only 100–400 quotas for Categories 1 and 3 and less than 100 for Category 4 (these are pre-1994 figures). Speculators then had an incentive to submit higher bids to keep pushing the quota premium up. Since the transferable COEs were valid for three months, the speculators were guaranteed a profit as long as the quota premium kept rising (Koh & Lee, 1994). When non-transferability was imposed in October 1991, car dealers and traders were able to continue trading COEs, although in a round-about manner, by using the COE to purchase a vehicle (specified by the buyer) and immediately selling the vehicle with its COE. This was known as “double transfer.” A double transfer was subject to a levy of 2% of the vehicle’s OMV. In April 1995, transfers of ownership of a vehicle registered with a non-transferable Categories 1–4 COE were banned in the first three months after purchase, and an additional levy was imposed on transfers of ownership made within four to six months (Phang et al., 1996). Phang et al. (1996) find that the measures introduced in 1995 were effective in reducing the quota premium. However, Fig. 1 shows that by late-1997 the quota premiums had returned to the peak levels of 1995. Many bidders were risk-averse or had an urgent need for a COE. These included people who had de-registered their vehicles and needed a new car, dealers who had to honour sales contracts, dealers who wanted to increase sales, and people who simply did not want to risk being unsuccessful and having to bid in the following auction, incurring interest, transaction costs and frustration (Koh & Lee, 1994). With the sealed-bid uniform auction format, these people submitted higher bids than their true valuation in order to secure their COE, knowing that the clearing price would be lower. Competition was intense between car dealerships. Once a customer gave a deposit to purchase a car to be delivered by a certain date, the dealer had to obtain a COE on their behalf. Luxury car dealers, with their higher profit margins, could afford to put in higher bids, which forced dealers of cheap cars to follow suit. Failure to obtain a COE not only meant the loss of an order, the dealer also incurred storage and capital costs for the unsold cars (Koh & Lee, 1994). Furthermore, dealers of cheap cars could not turn to the secondary market in transferable open category COEs since they were typically priced at Category 4 quota premium levels. Many car dealerships went out of business after the VQS was introduced. In some cases the luxury car dealerships replaced the cheap car dealerships. Within a dealership, the more expensive cars were promoted more vigorously due to their higher profit margins. These factors pushed the quota premium up and helped produce the shift in the vehicle mix towards luxury cars (Koh & Lee, 1994).

Transport Policies in Singapore

223

3.5. Auction Format When it was initially implemented, the VQS was a uniform-price sealed-bid auction. Milgrom and Weber (1982) show that in a uniform-price auction where successful bidders pay the highest losing bid,15 bidders bid according to their true valuation. This is because a bidder that bids higher than his true valuation may end up having to pay a price higher than his valuation, whilst bidding lower than his valuation reduces the probability of winning but does not increase his consumer surplus when he wins. In July 2001, the auction format was changed to open online bidding format. Potential bidders can now see the current market-clearing quota premium before deciding to whether participate in the auction or revise their bids. A bidder may change his bid as often as desired once he is in the auction. Bids can be made through the Internet, phone or Automatic Teller Machines. Once submitted, a bid cannot be withdrawn, and a bidder may only adjust his bid upwards, which reduces volatility in the clearing price. When there is a common-value element in the valuations of quota licences, an open auction improves transparency and offers an efficient price-discovery process. Milgrom and Weber (1982) show that an open auction format is indeed preferable when bidders’ valuations are affected by the information held by other bidders. An open ascending auction provides a reliable process of price discovery, so that the allocation of licences goes to the bidders that value the COEs the most. However, more information may also facilitate collusive outcomes given the small number of quotas in each category in each auction. Table 2 shows that the change to open bidding in July 2001 was accompanied by lower average quota premiums and volatility in all categories. Table 2. Quota Premiums Before and After Change of Auction Format in July 2001. Cat A

Cat B

Cat C

Cat D

Cat E

Sealed bidding Mean (S$) Standard deviation

36,685 9,208

38,345 10,939

22,140 7,321

1,084 361

38,738 8,305

Open bidding Mean (S$) Standard deviation

29,876 2,432

30,434 3,787

18,263 5,241

324 296

30,650 3,786

Note: To give a fair comparison, only the period after the combining of categories in May 1999 is considered. Sealed bidding covers the period May 1999 to June 2001; open bidding covers the period July 2001 to June 2003. Source: Land Transport Authority. www.lta.gov.sg.

224

GEORGINA SANTOS ET AL.

Although the introduction of open bidding has improved the transparency of the auction process, the system continues to suffer from the problems caused by sub-categorisation, the open category and non-transferability. In the first auction of June 2003 for example, the price of a Category A COE was S$31,799 and the price of a Category B COE was S$31,501. As a percentage of the car price, a Category B licence continues to cost less than a Category A licence.

3.6. Summary of the Shortcomings The need for sub-categorisation depends on the ultimate aim of the VQS. If the VQS is intended to ultimately reduce road usage, then since all cars occupy roughly the same amount of road space and parking, an auction system that placed a similar cost of owning a car on all cars, regardless of size, would be reasonable. Since cars of different engine sizes generate different amounts of emissions, from an environmental point of view, there is an argument for sub-categorisation, in addition to the argument of social equity. However, sub-categorisation to bring about social equity (so that buyers of small cars do not pay the same premium as buyers of large cars) clearly conflicts with open category to bring about flexibility. Sub-categorisation is aimed at segregating buyers, so that the quota premiums for smaller cars are lower than the premiums for larger cars. On the other hand, the open category provides flexibility in the vehicle fleet composition and has the effect of helping to equalise the premiums of all the categories. The relatively small quota of the open category has mostly been used to purchase larger cars (Categories 3 and 4), leading to an increase in the supply of quotas for larger cars over time and lowering their quota prices. The outcome has been a highly regressive system. Buyers of cheap cars have had to pay more for their quotas relative to the car’s price, and pay more in absolute terms too at times. There have been calls to scrap the open category and adjust the size of quotas for each category in order to make the quota premiums for Category A COEs cheaper than Category B COEs. It is doubtful however that equity will be easily restored even if the open category disappears. The government would need to determine the demand for vehicles in each category and the volatility in quota premiums would probably not be eliminated. Alternatively, the government could consider raising the ARF and the annual road tax rate. As the ARF is proportional to the vehicle’s OMV and the road tax is fully proportional to its engine capacity,16 raising both the road tax and the ARF would go some way to restoring the equity in the system. As this would prompt the COE premiums to fall, it would also help to reduce the distortionary effects of the VQS.

Transport Policies in Singapore

225

Koh and Lee (1994) propose an ad valorem bidding system for the VQS, whereby bidders bid a percentage duty on the OMV of their desired vehicle, rather than nominal amounts. In a VQS ad valorem auction, only three categories for cars, motorcycles and “goods vehicles and buses” would be required, and COEs could be made transferable again. This was considered by the VQS review committee but was rejected on the basis that it would make the system “unnecessarily complex” (Tan, 2001). However, ad valorem bidding should simplify the system and correct the problems created by sub-categorisation, the open category and transferability. The current system is regressive with buyers of cheap cars paying a higher percentage of their car price for their COE than buyers of expensive cars. Under ad valorem bidding, wealthy people buying a more expensive vehicle would pay more for their COE, and the quota licence as a percentage of each successful bidder’s vehicle’s OMV would be the same for everyone. Scrapping the open category would reduce the bias towards buying more expensive cars, and transferable COEs would improve efficiency given the uncertainty in the future value of a successful bidder’s COE.

4. ELECTRONIC ROAD PRICING Despite the crudeness of the system and the labour-intensive manual enforcement, the ALS introduced in 1975 was effective and succeeded in reducing traffic congestion in the central area of Singapore, which contained the CBD, during the morning and evening peak hours. As Chin (2002) explains, the main problems however were precisely linked to the lack of any automation. Being manual the system was prone to error. There were different vehicle types, plus the option of part-day licences, leaving room for mistakes. Enforcement was made at the entry points to the RZ and extending the ALS would have meant hiring more personnel to enforce the system. It was also difficult to fine-tune it, as introducing shoulder-peak charges would have lead to even more options and an even higher probability of mistakes by the enforcement officers. Thus, there was always a rush to enter just before or after the restricted hours that could not be smoothed by a shoulder-peak charge. Also, since a licence offered a vehicle an unlimited number of passages through the ALS control points17 there must have been some temptation for drivers to transfer their licences to other drivers, even though this was not legal. Introducing a manually controlled chargeper-pass would have delayed traffic as queues would have built at toll-booths. There was also a perception that a paper based ALS scheme was becoming out-of-place in a city-state that was becoming high-tech and aspired to be regarded as such.

226

GEORGINA SANTOS ET AL.

Thus, in 1989 the government approved the use of ERP for Singapore roads (Chin, 2002). The main contract to implement the ERP system was a “Design and Build” project awarded to a Consortium in October 1995 at about S$197 million (S$208 million at 2003 prices and US$116 million at 2003 exchange rates).18 From those S$197 million, about S$100 million covered the initial supply of IUs given to motorists for free during a ten-month grace period that was given at the time. As of 2003, motorists who wish to install an IU have to purchase it at S$150 (US$84) inclusive of the installation cost. The IU has several parts including a one-inch liquid crystal display, a sealed battery, a card acceptor, and electronic components that can operate reliably in the Singaporean tropical heat and humidity, with a five-year warranty. Foreign visiting vehicles can either rent IUs (at S$5 per day) and pay the variable ERP charges or buy an Autopass card (at S$10, comprising a pre-loaded cash value of S$4 and a non-refundable card cost of S$6) and pay a fixed ERP fee of S$10 for each day they drive on priced roads.19 The remaining amount of S$97 million was for the design, development, supply, installation and one-year warranty of the ERP equipment, including the gantries and central computer system. This computer system includes a powerful back-end system with three computer servers (with backups), monitoring systems and a master-clock to ensure that the timing at all the ERP gantries are synchronised. All the transactions and violation images are processed there. The system is designed to support up to a maximum of 100 ERP points. However, as of 2003, there are only 44 gantries on the road. A separate five-year maintenance contract of about S$39 million (S$41 at 2003 prices and US$23 million at 2003 exchange rates) was also awarded together with the main contract for the maintenance of the system. The ERP system is managed by the Authority at an annual operating cost of about S$8 million (US$4.5 million). This however varies from year to year and comprises mainly the salary of the staff and administrative expenses to manage the system. As explained above, the ERP system in Singapore charges motorists each time they pass through a gantry when the system is in operation. The ERP charges are published on the Land Transport Authority (LTA) website.20 The current scheme is a rather complex system with different charges according to vehicle type, time of the day, and location of the gantry. Police cars, ambulances and fire engines are all exempt. Charges for passenger cars, taxis and light goods vehicles vary between S$0.50 and S$3, charges for motorcycles vary between S$0.25 and S$1.50, charges for heavy goods vehicles and light buses vary between S$0.75 and S$4.50, and charges for very heavy goods vehicles and big buses vary between S$1 and S$6. In February 2003 a graduated ERP rate was introduced in the first five minutes of the time slot with a higher charge in order to

Transport Policies in Singapore

227

discourage motorists from speeding up or slowing down to avoid higher charges. For example, where the charge for passenger cars was S$2 between 8 and 8.30 AM and S$3 between 8.30 and 9 AM, it is now S$2 between 8.05 and 8.30 AM, S$2.50 between 8.30 and 8.35 AM, and S$3 between 8.35 and 8.55 AM, when it changes to S$2. The charging area is divided into central business districts, where charging applies from 7.30 AM to 7 PM, and expressways/outer ring roads, where charging applies from 7.30 AM to 9.30 AM. In principle, the ERP is able to charge the difference between marginal social and marginal private costs of road use, or at least, approximate the difference more accurately than a flat charge. The per-entry charge of the ERP has enabled the use of lower charges in comparison to the per-day charges under the ALS. Even though the ERP charges were lower than ALS charges, Menon (2000) finds that a year after its introduction, traffic volumes in the restricted areas had fallen by 15% during the whole day and by 16% during the morning peak. There were only minor increases in traffic at the times just outside the evening peak restricted period. He attributes this reduction to the sharp fall in multiple trips, as could be expected with the change to per-entry charging. Traffic speeds in the CBD remained in the optimum of 20–30 km per hour (12.4–18.6 mi per hour) range.

5. INTEGRATED TRANSPORT POLICY Apart from discouraging vehicle ownership and usage, Singaporean transport policy has also included the provision of a wide range of attractive alternatives to driving. The development of the MRT rail system was a strategic decision, linked to the objective of building a world-class urban infrastructure to make Singapore a destination city to invest, work and live in. Phang (2003) argues that in other cities, the increase in land use density and land value along the rail corridor are usually accompanied by decreases in land use density and land value elsewhere. In Singapore, things are different. It is a small island where land is scarce and the government effectively controls its use. The railway network expansion in Singapore has increased the intensity of land use not only in the CBD but also around many suburban railway stations. The Singaporean government provided the financing for the construction of the MRT rail system. During the 1980s and early 1990s, the government paid for the long-term infrastructure and initial operating assets of the MRT project, including rails, tunnels, stations, trains and signalling systems (Phang, 2003). The operator was required to generate enough revenue to cover operating and maintenance

228

GEORGINA SANTOS ET AL.

costs, including the replacement cost of operating assets (Phang, 2003). In 1996 with the publication of the LTA (1996) White Paper, which recognised that the current rules would require the imposition of higher fares, the government decided to bear the burden of asset inflation. Existing and future railway projects were only required to generate enough revenues to cover operating and maintenance costs, depreciation, and the historical replacement cost of operating assets (Phang, 2003). Hence in 2017, of the S$6.9 billion projected cost of replacing the current set of operating assets, the operator will only pay S$1.6 billion (cost of the assets in 1987) and the government will pay S$5.3 billion. In 2047 when the second set of operating assets are due for replacement, the operator will pay S$6.9 billion and the government will pay the asset inflation costs (Phang, 2003). This change paved the way for the subsequent approval of seven new MRT/LRT projects, more than doubling the existing rail network length (Phang, 2003). In other areas, the government has also been proactive to make public transport a palatable alternative to driving. For instance, in 1996, the government encouraged all three taxi operators to introduce Global Positioning System (GPS) technology for booking and dispatching taxis (LTA, 1996). Fares are allowed to vary according to time of day and demand, and can be set much higher during peak periods. Similarly, in 1997, Singapore Bus Services introduced an “on time” GPS system to provide customers with accurate information about bus arrival times. In 2001 it launched a pilot project to install electronic display panels at major bus stops to inform waiting passengers the expected arrival time of the next bus. Commuters could also telephone to find out the expected arrival time of the bus at their stop (Goh, 2002). However, the project was soon halted in light of the soft macroeconomic conditions, and lukewarm response. Two other developments in the 1990s deserve some mention. First introduced in 1975 but abandoned several months later, a Park-and-Ride scheme was re-introduced in 1990 to encourage drivers to park their vehicles at suburban MRT stations and take the MRT to the central business district. After changes in 1992 and 1995, the revived scheme is still successful (Foo, 1997b), in contrast with the poor response it had in 1975. Then, it was a bus-based system, where drivers parked their cars at designated car parks in the fringe of the CBD and then took the bus to work. Not only did it not solve the congestion problem, but also it was inconvenient and cumbersome. The government has also encouraged the formation of car cooperatives. Car sharing using a car cooperative scheme allows an individual to have access to a car and use it without having to own it. The first car cooperative in Singapore started in 1997. The idea of a cooperative is to operate and maintain a small fleet of cooperative cars, which are parked at designated sites for the cooperative

Transport Policies in Singapore

229

members, who pay an annual membership in exchange for the right to use these cars, plus the cost of their trips. Similar schemes for the U.K. and the U.S. are mentioned in Chapters 11 and 12 of this volume respectively. Foo (2000) finds that a majority of car cooperative members commute to work by public transport and use the cooperative car for other purposes such as leisure and social trips. Advantages of car cooperatives over renting cars include ease of access and the option of hiring the car for only a few hours rather than per day. Finally, in an effort to create smooth-flowing traffic, the government also installed a Green Link Determining (GLIDE) system in 1996, which coordinates traffic signals to give more “green” time where traffic flow is heavy (LTA, 1996). The Expressway Monitoring and Advisory System was installed in 1998. The system uses cameras to detect accidents and traffic conditions. The control centre alerts drivers of problems ahead through the television, radio and electronic signboards placed along and at the approaches to expressways (LTA, 2002).

6. POLICY DEVELOPMENTS Willoughby (2000) finds that since the 1960s, the revenues collected by the government from road transport have systematically exceeded government expenditure on road transport by between three and six times. In 2002 for example, only 19.8% of road transport revenues were spent on road transport. Revenues from road transport were 7.6% of total government revenues and expenditure on road transport was 2% of total expenditures.21 Even when government investment in MRT is included, transport revenues have typically exceeded expenditure on transport. In 1995, 11% of Singapore’s land area was occupied by roads (LTA, 1996). At the time the government considered this to be an optimal share, since further increases would have been at the expense of housing, offices, factories, etc. Although many Asian cities have a similar share, the share for most European cities is twice as high, and the figure for many North American cities is three times or more higher (Willoughby, 2001). Hau (1992) states that road space should be added whenever the economic rent earned on a congested stretch of road exceeds the economic rent of having alternative uses for the adjoining land. A policy that limits roads to 11% of land area means that the decision to add new roads was not based on economic efficiency but rather on policy judgment and strategic intent. The extremely high VQS quota premiums seen in the mid-1990s is evidence that with increasing affluence and disposable incomes, Singaporeans wanted to own a car, even if they only made limited use of it. The ERP is able to charge efficiently for the use of individual roads by setting appropriate fees for varying

230

GEORGINA SANTOS ET AL.

levels of congestion over the day. It has also brought about a rationalisation of vehicle ownership and usage taxes: in 1998 the ARF was lowered and the annual growth rate of VQS quotas was increased by 9–12% (Toh & Phang, 1997). The ERP is now the centerpiece of the government’s transport policies, and is shifting the burden of taxation away from ownership taxes, with an aim to lowering the 3:1 ratio of upfront and annual fixed costs against usage costs (LTA, 1996) and the high annual mileage of the average car. The government introduced these changes gradually to prevent a sharp decline in motor vehicle prices and a destabilisation in banks’ consumer lending, much of which used to be used for car purchases (Willoughby, 2000). Following the LTA’s recommendation, the government has also allowed an increase in the vehicle population, with the goal of increasing car ownership from one car per ten people in 1995 to one per seven people by 2010 (LTA, 1996). The ERP charges have remained stable, but the ERP zones have been continuously expanded. Between 1996 and 2001, the road network increased by 415-lane-km (Singapore Ministry of Transport, 2001), a figure much higher than the 225-lane-km targeted by the end of 2001, as stated in LTA (1996). Although this may suggest that the government has relaxed the 11% land use policy, it should also be borne in mind that Singapore land area is growing due to land reclamation from the sea.22 In 2002, ARF charges were lowered, higher PARF discounts were given, and import duties were reduced from 31 to 20%. These recent developments indicate that the Singaporean government is sensitive to the people’s aspirations to own a car, and accordingly, has adjusted the transport policies to meet these aspirations. By shifting the cost of car ownership towards a usage-based charge under the expanding ERP scheme, and with further investment in the road network, the government has been able to cater for the expected increase in demand for road usage without compromising its goal to control urban congestion.

7. CONCLUSIONS The ERP scheme, the first of its kind in the world, has proven to be a flexible instrument that can price road usage, according to congestion conditions, time of the day, vehicle type and road. Since it was implemented in 1998, the charges have been fine-tuned to achieve the objective of maintaining traffic speeds within an acceptable range. By controlling the number of new vehicle registrations each year, the VQS has slowed the growth of the vehicle population. The introduction of the open-bidding format has made the VQS more transparent compared with the previous sealed

Transport Policies in Singapore

231

format. Sub-categorisation however has prevented the system from achieving equity or efficiency. Luxury car buyers continue to pay a lower COE premium relative to the car price, the incentives are distorted, and this favours the purchase of expensive cars. Non-transferability has also removed some flexibility in the system to adjust allocation in response to changing demand conditions. Further improvements can be made to the VQS. One solution proposed in this chapter is to raise the annual road tax. This would weaken the impact of the VQS by reducing its quota prices, but it would retain the present auction format. Although eliminating the open category would probably reduce the bias towards buying more expensive cars (as open category COEs effectively supplement Category B COEs at present), it would not resolve the problem of equity. Perhaps a simpler way to address the equity problem would be to change to an ad valorem bidding system, as suggested by Koh and Lee (1994). This would simplify the system and correct the distortions. Under ad valorem bidding, buyers of expensive cars would pay more for their COE, but the quota licence as a percentage of each successful bidder’s vehicle’s OMV would be the same for everyone. Although licence transferability has led to speculation in the past, some degree of transferability may be desirable in order to inject some flexibility in the final allocation, and improve efficiency. Singapore’s integrated transport policy has been largely successful in both controlling urban congestion and meeting the people’s aspirations to own a car. Most people day-to-day commute by public transport, and if they have a car, it is used mainly for leisure and social trips. By investing in public transport (MRT, taxis and buses) and encouraging its use, the provision of public transport has become profitable. This has produced a virtuous cycle as further improvements in infrastructure are being made to enhance the quality of services. Without the hands-on involvement of the government in driving the various transport policies, there would have been a vicious cycle with increased car usage, deteriorating quality of bus and rail services, worsening congestion and a sharp drop in the quality of public transport. The 10-yearly Household Expenditure Survey in both 1983 and 1993 finds that private vehicle ownership in Singapore has been a privilege of the rich, with the richest 10% of households accounting for 30% of total expenditure on cars (Willoughby, 2000). Although this is a reflection of the high cost of vehicle ownership, the government has ensured that the less wealthy households have access to convenient, frequent, reliable and safe public transport. In many cities in both the developed and developing world, traffic congestion threatens to seriously hinder the functioning of businesses. The success of road pricing in Singapore serves as model for other countries. Even less sophisticated systems, such as the one implemented in London in February 2003, reduce traffic

232

GEORGINA SANTOS ET AL.

congestion and increase economic efficiency. A lesson for London and other cities around the world that might introduce congestion charging is that the ERP in Singapore has shown that a per-entry charge is more effective at reducing congestion than a per-day charge. To conclude, Singapore’s experience with vehicle quotas and electronic road pricing has been successful and this gives hope to other towns and cities around the world. Most importantly, the Singaporean government has managed to convince the motoring public that controlling congestion is essential for an economy to function efficiently. Though many suggest that the one-party democracy in Singapore is uniquely able to carry out its policies decisively, Singaporeans in general understand the rationale and support the government’s transport policies, perhaps because these constitute an integrated package that increases the costs of driving but at the same time makes the option of public transport attractive (Chu & Goh, 1997). If the policies had only been aimed at raising the costs of owning and using cars they would have probably failed. This is where the Singaporean example can claim its major victory: the support from the public thanks to its integrated approach to transport policy.

NOTES 1. This was enhanced at the beginning of 1991 with the introduction of an automatic ticketing system, which consisted of a stored value card that could be used to pay for the MRT and for public buses. In April 2002 a smart-card ticketing system replaced the stored value one. 2. LRT is less costly and takes half the time to build (3–4 years) compared to the MRT. 3. The OMV is essentially similar to Cost plus Insurance and Freight (CIF). It includes purchase price, freight, insurance and all other charges incidental to the sale and delivery of the car to Singapore. 4. Land Transport Authority (LTA) website: http://www.lta.gov.sg. 5. It also recommended usage restraint measures and upgrading and expansion of the road infrastructure. 6. The introduction of the VQS was accompanied by a reduction of the ARF rate, from 175% in 1990 to 150% in 1991. The ARF has however remained high and as of 2003 it stands at 130%. 7. Initially the period of validity was six months but this was changed to three months from the second auction in August 1990. This was changed back to six months for Categories 1–4 and 6 in October 1991, when the transferability of the COEs was removed. 8. As originally intended, a Category 7 COE could also be used to purchase a motorcycle at one third of the quota premium, but given the wide disparity in the quota premiums for Category 7 and Category 6 (motorcycles and scooters), this flexibility was seldom, if at all, utilised. 9. Apart from paying tolls under the ERP system, the smart cards can also be used to pay for parking. Many privately run car parks are equipped with automatic sensors so that when

Transport Policies in Singapore

233

vehicles equipped with IUs and smart cards enter the car park the sensor detects the smart card, and when they leave a charge is automatically deducted according to the time the car was parked. Other uses include payment of National Library Board fees and purchases from retail shops that support cash card payments. These cards can store up to S$500. 10. These estimates were produced by the Centre for Transportation Studies at Nanyang Technological Institute, Singapore (1991) and are still used by the government when setting the level of fees (Chin, 2002). 11. Here as in all other instances where the figures given are not referenced, the source is the LTA website, http://www.lta.gov.sg. 12. This has always been the case for the VQS. Theoretically, a large quota for the open category would, by arbitrage, equate the quota premiums of all the categories and defeat the point of sub-categorisation. 13. This time period was chosen for the analysis here because in May 1999, Categories 1 and 2 were combined and Categories 3 and 4 were also combined. 14. At the same time, Categories 5, 6 and 7 were simply renamed Categories C, D and E. 15. In the VQS, the successful bidders pay the lowest successful bid. With many bidders in the market so that bids are continuously distributed, this is analytically equivalent to paying the highest rejected bid. All bidders would be expected to bid their valuation and everyone gets a surplus except the lowest winning bidder who gets zero surplus. 16. In Singapore the road tax structure is split into categories by engine size but within each category there is a sliding scale. Source: LTA website: http://www.lta.gov.sg. 17. In the case of the RPS on the expressways, the licences also offered an unlimited number of passages through the RPS manually controlled points. 18. All the data on costs in this section were kindly provided by the Land Transport Authority. 19. This fixed S$10 fee was computed by the Land Transport Authority by taking into account the cost of renting a temporary IU and the average ERP charges incurred by a foreign motorist each day. 20. http://www.lta.gov.sg/. 21. Government expenditure on road transport in Singapore in 2002 was S$379 million, out of total expenditures of S$19,344 million; government revenues from road transport were S$1,559 million from vehicle taxation plus S$359.3 million from vehicle import duties, out of total revenues of S$25,401 million (Customs and Excise, 2003; Department of Statistics, 2003). 22. The area increased from 633 km2 in 1990 to 683 km2 in 2003 (equivalent to a growth of 8.2%).

ACKNOWLEDGMENTS The authors are indebted to Kian Keong Chin, Eddie Lim Sing Loong, Geraldine Chan and the Land Transport Authority in Singapore, for provision of useful information, and to David Newbery, Martin Richards and Piotr Olszewski for helpful comments on an earlier draft. Any remaining errors are the authors’ own. Support from the British Academy for Georgina Santos is gratefully acknowledged.

234

GEORGINA SANTOS ET AL.

REFERENCES Bhagwati, J., & Srinivasan, T. (1969). Optimal intervention to achieve noneconomic objectives. Review of Economic Studies, 36, 27–38. Centre for Transportation Studies, Nanyang Technological Institute, Singapore (1991). Speed-flow model for traffic within the restricted zone. Nanyang Technological Institute. Chin, K. K. (2002). Road pricing: Singapore’s experience. Essay prepared for the third seminar of the IMPRINT-EUROPE Thematic Network: Implementing reform on transport pricing: Constraints and solutions: Learning from best practice. Brussels, October 23–24. http://www.imprint-eu.org/public/Papers/IMPRINT3 chin.pdf. Chu, S., & Goh, M. (1997). The price of car ownership in Singapore: An empirical assessment of the COE scheme. International Journal of Transport Economics, 24, 457–472. Cited in Willoughby (2001). Customs and Excise, Singapore (2003). Revenue collected by Singapore customs department. http://www.gov.sg/customs/trade/trde12 3.html. Department of Statistics, Singapore (2003). Economic survey of Singapore: First quarter, 2003. http://www.singstat.gov.sg/. Falvey, R. E. (1979). The composition of trade within import-restricted product categories. Journal of Political Economy, 87, 1105–1114. Foo, T. S. (1997a). An effective demand management instrument in urban transport: The area licensing scheme in Singapore. Cities, 14, 155–164. Foo, T. S. (1997b). Experiences from Singapore’s park-and-ride scheme (1975–1996). Habitat International, 21, 427–443. Foo, T. S. (2000). Vehicle ownership constraints and car sharing in Singapore. Habitat International, 24, 75–90. Goh, M. (2002). Congestion management and electronic road pricing in Singapore. Journal of Transport Geography, 10, 29–38. Hau, T. D. (1992). Economic fundamentals of road pricing: A diagrammatic analysis. World Bank Policy Research Working Paper Series 1070, The World Bank, Washington DC. Holland, E. P., & Watson, P. L. (1978). Traffic restraint in Singapore. Traffic Engineering and Control, 19, 14–22. Koh, W., & Lee, D. (1993). Auctions for transferable objects: Theory and evidence from the vehicle quota system in Singapore. Asia Pacific Journal of Management, 10, 177–194. Koh, W., & Lee, D. (1994). The vehicle quota system in Singapore: An assessment. Transportation Research A, 28, 31–47. Koh, W. (2003). Control of vehicle ownership and market competition: Theory and Singapore’s experience with the vehicle quota system. Transportation Research A, 37, 749–770. Krishna, K., & Tan, L. H. (1997). A note on India’s MFA quota allocation system: The effect of sub-categorisation. Annales d’Economie et Statistique, 47, 39–50. Krishna, K., & Tan, L. H. (1999). Transferable licences vs. non-transferable licences: What is the difference? International Economic Review, 40, 785–800. Land Transport Authority (1996, January). A world class land transport system. White Paper presented to the Singaporean Parliament. Singapore. Land Transport Authority (2002). Annual report 2001–2002: Riding the next wave. Singapore. http://www.lta.gov.sg/AbtLTA/pdf/annual LTA.pdf. Li, M. Z. F. (1999). Estimating congestion toll by using traffic count data – Singapore’s area licensing scheme. Transportation Research E, 35, 1–10.

Transport Policies in Singapore

235

McCarthy, P., & Tay, R. (1993). Economic efficiency vs. traffic restraint: A note on Singapore’s area licence scheme. Journal of Urban Economics, 34, 96–100. Menon, A. (2000). ERP in Singapore – A perspective one year on. Traffic Engineering and Control, 41, 40–45. Milgrom, P., & Weber, R. (1982). A theory of auctions and competitive bidding. Econometrica, 50, 1089–1122. Mohring, H. (1983). The Singapore MRT: What price a large central area work force? Suara Ekonomi, 20, 9–15. Cited in Phang (2003). Olszewski, P., & Turner, D. (1993). New methods for controlling vehicle ownership and usage in Singapore. Transportation, 20, 355–371. Phang, S. Y. (2003). Strategic development of airport and rail infrastructure: The case of Singapore. Transport Policy, 10, 27–33. Phang, S. Y., & Chin, A. (1990). An evaluation of car ownership and car usage policies in Singapore. Report of the Select Committee on Land Transportation Policy, 105–117. Cited in Toh and Phang (1997). Phang, S. Y., & Toh, R. S. (1997). From manual to electronic road congestion pricing: The Singapore experience and experiment. Transportation Research E, 33, 97–106. Phang, S. Y., Wong, W. K., & Chia, N. C. (1996). Singapore’s experience with car quotas: Issues and policy processes. Transport Policy, 3, 145–153. Singapore Ministry of Transport (2001). Parliament questions & answers: World class transport system. http://www.mot.gov.sg/newsroom nav/parliament matters/pq worldclasstransport.html. Tan, L. H. (2001). Rationing rules and outcomes: The experience of Singapore’s vehicle quota system. IMF Working Paper 136, Washington DC. www.imf.org/external/pubs/ft/wp/2001/ wp01136.pdf. Toh, R. S. (1992). Experimental measures to curb road congestion in Singapore: Pricing and quotas. Logistics and Transportation Review, 28, 289–317. Toh, R. S., & Phang, S. Y. (1997). Curbing urban traffic congestion in Singapore: A comprehensive review. Transportation Journal, 37, 24–33. Wilbur Smith & Associates (1974). Singapore mass transit study, phase 1. Final report. Wilbur Smith and Associates, Singapore. Cited in Foo (2000). Willoughby, C. (2000). Singapore’s experience in managing motorization, and its relevance to other countries. World Bank Transport Division TWU-43. www.worldbank.org/transport/ publicat/twu 43.pdf. Willoughby, C. (2001). Singapore’s motorization policies 1960–2000. Transport Policy, 8, 125–139.

This Page Intentionally Left Blank

10.

NORWEGIAN URBAN TOLLS

Farideh Ramjerdi, Harald Minken and Knut Østmoe 1. INTRODUCTION The introduction of urban tolls in Norway should be viewed in the light of the limited public funds for financing the necessary road infrastructure to cope with the growing traffic. Toll financing of road infrastructure in Norway, in the traditional sense, dates back to the 1930s. For many years, toll projects mainly took place on the peripheries. Before the introduction of the first urban toll in Bergen in 1986 the contribution of toll revenues to the total funds for road infrastructure investment did not exceed 4–5%. In the year 2000 the contribution of toll financing (net toll income plus loans)1 to the total funds for road infrastructure investment increased to about 33%. At the same time there was an increase in the number of toll-financed projects that were located in urban areas. Different aspects of the Norwegian urban tolls in Bergen, Oslo and Trondheim have been well covered in the literature since their introduction (G´omez-Ib´an˜ ez & Small, 1994; Hau, 1992; Jones, 1998; Larsen, 1988; Larsen & Østmoe, 2001; Meland & Polak, 1993; Odeck & Br˚athen, 1997; Ramjerdi, 1992, 1994a, b, 2002; Vold et al., 2001). Nearly all of the toll-financed projects in Norway are supplemented by grants from the central government. The toll-financed projects are based on local initiatives. However, they require approval by the Norwegian parliament. The Norwegian law sets a limit on the period of toll collection for financing purposes and the approval for toll collection is usually granted for 10–15 years. The Norwegian Public Roads Administration has the responsibility for the planning, construction and maintenance of the toll projects as well as the toll collection Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 237–249 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09010-9

237

238

FARIDEH RAMJERDI ET AL.

system. A private company with limited responsibility, established by the local authorities and interest groups, operates the toll system. The shareholders in these private companies are most often public bodies such as the affected municipalities. The responsibility for financing as well as collecting and administrating the toll revenues lies with the private toll companies. The Bergen cordon toll scheme, a manually operated system,2 was successfully implemented in January 1986. This was followed by the introduction of two full-scale electronic toll payment projects in Norway. The first was a toll project ˚ in Alesund (on the west coast of Norway) in 1987 and the second, a road toll station at Ranheim (east of Trondheim) in 1988. These successful demonstrations of technologies alongside with the experience from Bergen prepared the way for the introduction of electronic pricing on a full-scale basis in Oslo in 1990, in Trondheim in 1991, and more recently, in Kristiansand in 2000, and in Stavanger in 2001. Since the mid-1990s amendments to the Road Acts have made it possible to allocate some of the revenues from a toll scheme to investment in public transport. A new amendment, which was approved in June 2002, sanctions the use of a toll scheme for demand management and has opened the possibilities for introducing congestion charging in urban areas. According to this amendment, there is no time limit on a congestion pricing scheme in an urban area, but any scheme has to be reviewed every ten years. The proceeds from a congestion pricing scheme have to be used for local transport purposes.3 In that sense, the Norwegian legislation is not too different from the British legislation by which, as explained in Chapter 11 of this volume, local authorities are granted powers to introduce congestion charges (and workplace parking levies) in order to manage congestion in towns and cities, under the condition that the revenues are hypothecated to improve local transport, for at least ten years. In this chapter the urban toll rings in Norway are discussed and some lessons drawn.

2. THE BERGEN TOLL RING The toll ring in Bergen, the second largest city in Norway, was introduced in January 1986. The city council approved the scheme in January 1985 and the Norwegian Parliament approval came in June 1985. The scheme was originally approved for fifteen years, from 1986 to 2001. In 2002 the Norwegian Parliament approved the extension of the deadline to 2011. The scheme in Bergen is manually operated. Electronic collection will start in 2004, when toll fees will be increased by approximately 50%. A time differentiated toll scheme, with the aim of

Norwegian Urban Tolls

239

reducing congestion during peak periods, is an alternative that has been discussed for Bergen. The original aim of the scheme was to provide supplementary funding for a package of road projects. There was no provision of funds for public transport, even though a small amount was allocated for the construction of special bicycle lanes. Some of the projects were tunnels to divert traffic from the city streets, which had some positive environmental impacts. As of 2003, about 45% of toll revenues is now allocated to road construction and 55%, to projects that improve environmental quality and safety. Bergen has a population of about 233,000. About 10% of the population lives inside the toll ring. The limited number of entries to the central city was an important factor for the successful introduction of a manual system. Figure 1 shows the location of the Bergen toll ring around the central business district. The toll ring comprises 7 toll stations located at about two to eight kilometres from the city centre. Inbound traffic is tolled Monday through Friday 6 AM to 10 PM, except on holidays. Altogether twenty traffic lanes lead to these stations.

Fig. 1. Toll Ring in Bergen.

240

FARIDEH RAMJERDI ET AL.

Ten of these lanes are reserved for traffic with seasonal passes. The capacity of a reserved lane is about 1500 vehicles per hour while the capacity of a manual lane is 400–600 vehicles per hour. The reserved lanes for non-stop traffic (with seasonal passes) have a conventional video enforcement system. The penalty for violation is NOK 300. The violation rate is about 1.5%. The impacts of the Bergen toll scheme on travel behaviour have been small. The authorities claim that the impact has been less than the original estimate of 3% reduction in car traffic crossing the ring. Toll fees are set to promote the use of seasonal passes. The toll fee in 2002 was NOK 10 for light vehicles. A seasonal pass for a light vehicle was NOK 200 for one month, NOK 600 for 3 months, NOK 1150 for 6 months and NOK 2,200 for one year. A seasonal pass allows the driver to cross the toll ring an unlimited number of times during a specified period. Prepayment of NOK 180 for 20 crossing tickets is available, to be used at the manual gates. Toll fees for heavy vehicles are twice these amounts. The annual gross toll revenue in 2002 was about NOK 156 million. The annual operating cost was about NOK 30 million, approximately 19% of the revenue. The toll revenue has proved to be far higher than originally expected.

3. THE OSLO TOLL RING The toll ring in Oslo was introduced in February 1990. The electronic payment system became operational in December 1990. An extension of the scheme has been approved for up to 2007. In 2000 the Parliament approved an increase in the toll fee in order to finance an investment package on public transport projects, referred to as “Oslo Package 2.” Originally, the toll revenue, supplemented by about equal funds from the central government, was to finance the “Oslo Package” (now referred to as “Oslo Package 1,”) comprising some 50 new road projects. The aim of the “Oslo Package 1” was to increase road capacity. About 30 of these road projects are tunnels that divert traffic from city streets. It is estimated that by 2007 the total contribution of the scheme to “Oslo Package 1” will amount to NOK 9.1 billion at 2002 prices, approximately 15–20% above the initial estimate. There is much debate and some interest in changing the direction of the scheme to a congestion charging scheme. Amongst the different alternatives that have been evaluated for Oslo, there is a time differentiated toll scheme with the purpose of reducing car traffic during peak periods. Revenues would be allocated to public transport and extension and improvement of roads in the region. The Oslo scheme is very likely to continue with some modifications.

Norwegian Urban Tolls

241

The population in the Oslo region is about one million,4 About 52% of work locations are inside the toll ring, whereas only about 30% of people actually reside inside the toll ring. The greenbelt areas in the north and the east of Oslo combined with the Oslo fjord result in three corridors leading to the central parts of Oslo, where the toll ring is located. The three corridors go from the city centre to the West, to the West/North-West and to the South. The home and work locations, outside the toll ring, are almost equally distributed amongst the three corridors. Figure 2 shows the location of the toll ring in the Oslo/Akershus region. The location of the toll stations is mainly the outcome of practical considerations and political negotiations. As explained in Chapter 4 of this volume, cordons based on judgement may not always be optimally located. The toll ring consists of 19 toll stations located at about three to eight km from the city centre.5 Inbound traffic is tolled all day long, every day of the year. Each toll station has a manual lane with an attendant. Most stations also have lanes with coin machines for toll payment. The capacity of a reserved lane for

Fig. 2. Toll Ring in Oslo.

242

FARIDEH RAMJERDI ET AL.

non-stop traffic with seasonal passes is 1600 vehicles per hour, whereas the capacity of a manual lane is about 400 vehicles per hour, similar to Bergen. The toll fee in 2002 was NOK 15 for light vehicles and twice as much for heavy vehicles.6 The seasonal pass for light vehicles in 2002 was NOK 400 for one month, NOK 2,250 for 6 months and NOK 4,100 for one year. The system allows for a number of prepaid passes that can be used over an unlimited period of time. These cost NOK 4,000 for 350 crossings, NOK 2,200 for 175 crossings and NOK 340 for 25 crossings. In 2002 the annual gross toll revenue was NOK 1046 million with an annual operating cost of NOK 103 million, approximately 9.8% of the revenue. Each reserved lane for non-stop traffic is equipped with a registration unit that reads the identity of the passing passive tags. The identity is fed into a computer to check for its status. The passing driver receives a signal on the status of the tag on the vehicle. The toll station computer activates a video camera when an illegal passage is registered. The video photographs of the licence plates of the infractors are recorded for fining. The average violation rate in a non-stop lane is about 0.2%. The accuracy rate of the system is over 97% and is expected to improve significantly with some modifications. Very strict rules apply to the data compiled during electronic toll collection in order to ensure the privacy of drivers. The reported impact of the Oslo scheme in its first year of operation ranges from insignificant (Wærsted, 1992) to about 10% reduction in the number of car crossings (Solheim, 1992). Ramjerdi (1994b) puts this estimate at 3–4%. The introduction of the toll scheme in Oslo did not have significant impact on the demand for public transport or on car occupancy (Gylt, 1991; Nordheim & Sælensminde, 1991).

4. THE TRONDHEIM TOLL RING The toll ring in Trondheim started to operate in October 1991 with the full use of the electronic toll payment system. The scheme was approved for the period of 1991–2006. Unlike the schemes in Oslo or Bergen the extension of the Torndheim scheme beyond the approved dates seems unlikely. The net revenue of the toll scheme finances a package of road infrastructure projects, mainly to increase road capacity, with some earmarking to public transport and facilities for pedestrians and cyclists. The toll revenue will contribute to about 40% of the total cost of the package, estimated at NOK 2 billion in 1992 (Tretvik, 1992) or NOK 2.52 billion at 2002 prices. Trondheim is the third largest city in Norway with a population of about 150,000. About 40% of the population live inside the toll ring. There are 22 toll stations.7

Norwegian Urban Tolls

243

Fig. 3. Toll Ring in Trondheim.

One of these toll stations was abandoned in July 2003. The station on highway E6-East operates 24 hours a day seven days a week. In this sense, it might be considered not to be a part of the Trondheim toll ring. For the others, inbound traffic is tolled Monday through Friday from 6 AM to 6 PM with different fees according to time of the day. Figure 3 shows the location of the toll ring in Trondheim. All toll stations have lanes with machines that accept coins and magnetic strip cards. Unlike Oslo, toll stations are unattended except for two. These two stations cater for non-local traffic, which is significant during the tourist season. The toll fee at the manual stations was NOK 15 for light vehicles in 2001. There is no electronic payment option with unlimited use in Trondheim. With a prepayment of NOK 6,000, the toll fee is NOK 9 between 6 and 10 AM and, NOK 6 between 10 AM and 6 PM. With a prepayment of NOK 3,000, the corresponding toll fees are NOK 10.5 and NOK 7.5. With a prepayment of NOK 1000, these fees are NOK 12 and NOK 9. A post-payment option is available in Trondheim. Toll fees with the post-payment option are NOK 13.5 between 6 and 10 AM and NOK 10.5 between 10 AM and 6 PM. In order to encourage the use of the electronic system,

244

FARIDEH RAMJERDI ET AL.

users of the electronic system have discounts compared to those paying manually. Discounts per trip are lower during peak times than during off-peak times. Drivers using the electronic system are only charged for their first 60 entries each month and for only one entry in any one-hour. The electronic tags have become widely accepted in the Trondheim area, even amongst the infrequent users of the system. The annual gross toll revenue in 2002 from the Trondheim scheme, excluding revenue from the toll station on E6-East, was NOK 168.2 million, and the gross revenue from the toll station on E6-East was NOK 120.6 million. The corresponding annual operating costs were NOK 17.1 million (approximately 10% of the revenue) and NOK 18.3 million (approximately 15% of the revenue).8 Meland and Polak (1993) report that the impacts of the toll ring on mode choice or the total traffic that crosses the toll ring have been quite small. However Meland (1995) reports that inbound car traffic decreased by 10% during charging periods, and this decrease was almost offset by an 8–9% increase in inbound car traffic during non-charging periods at evenings and weekends.

Fig. 4. Toll Ring in Stavanger.

Norwegian Urban Tolls

245

5. THE STAVANGER TOLL RING The toll ring in the county of Stavanger/Sandnes started operation in 2001 with the use of the electronic toll payment system. Stavanger/Sandnes has a population of 162,000. The scheme has been approved for a period of 10 years. The proceeds from the scheme will contribute to about 41% of the cost of an investment package on road, public transport and traffic safety. The cost of this package has been estimated at NOK 2,639 million at 2002 prices. The scheme consists of 21 toll stations located on the county borders on all main roads. Inbound traffic is tolled Monday through Friday from 6 AM to 6 PM. The toll fee is highest between 7 and 10 AM and between 2 and 5 PM. Figure 4 shows the location of the cordon in Stavanger/Sandnes. Drivers using the electronic toll payment system are only charged for their first 75 entries each month and for only one entry in any one-hour period. Table 1 shows the different fees for light vehicles. The fee for a heavy vehicle is twice as much. The annual gross toll revenue in 2002 was NOK 80 million and the annual operating cost was NOK 21.1 million, approximately 26% of the revenue.

6. THE KRISTIANSAND TOLL RING Kristiansand has a population of about 74,000. The toll ring in Kristiansand started at the end of 1997 with only one toll station. In early 2000 three new toll stations were opened. The proceeds from the scheme will contribute to about 56% of the financing of a road infrastructure project that is estimated at NOK 920 million at 2002 prices. The scheme is manual. The toll fee is NOK 10 at the manual stations. A yearly pass costs NOK 1500, a six month pass costs NOK 800, and a monthly Table 1. Type of Toll Payments in Stavanger. Type of Payment

Toll Fee in NOK During the Time Period of 6.00–7.00

7.00–9.00

9.00–14.00

14.00–17.00

17.00–18.00

Manual Prepayment, NOK 175 Prepayment, NOK 1050 Prepayment, NOK 1750

5.00 3.50 3.00 2.50

10.00 7.00 6.00 5.00

5.00 3.50 3.00 2.50

10.00 7.00 6.00 5.00

5.00 3.50 3.00 2.50

Postpayment arrangement 0–5 crossings 6–10 crossings >10 crossings

5.00 4.50 4.00

10.00 9.00 8.00

5.00 4.50 4.00

10.00 9.00 8.00

5.00 4.50 4.00

246

FARIDEH RAMJERDI ET AL.

pass costs NOK 150. The annual gross toll revenue in 2002 was NOK 95 million and the annual operating cost was NOK 19.7 million, approximately 20.5% of the revenue. In 2003 another small community, Namsos, introduced a toll ring, probably the smallest toll ring in the world. Another community, Tønsberg, was planning to introduce a toll ring in January 2004. A poll carried out in 2003 showed that 80% of the population in Tønsbery were against the introduction of the ring. The opposition to the toll gained the majority in the local election held in September 2003, thus putting the plan for the introduction of a toll ring under question.

7. CONCLUSIONS Toll financing of road infrastructure has been common practice in Norway since the 1930s. However, toll financing of large road investment packages in urban areas only began with the introduction of the Bergen toll ring in 1986. This scheme was implemented in response to the limited funds available for road infrastructure investment and growing traffic in Bergen. Oslo and Trondheim followed and other towns and cities have also implemented or will implement some form of tolling in order to finance transport projects. Additional grants from the central government to supplement toll revenues have also acted as an additional incentive for local authorities to implement toll rings. Long and careful negotiations and compromises at political level and between different interest groups were necessary for the different urban toll rings in Norway to be implemented. Public attitutude has fluctuated. Despite the initial negative public attitude (Odeck & Br˚athen, 1997, 2002), toll rings were introduced in the three largest cities in Norway. After the introduction of the toll rings, public opinion changed to some extent, probably because drivers realised that their money was being spent on transport projects from which they would benefit. Also, there has always been good and clear information available to the motoring public. Each of the urban toll rings was linked to a well-defined package of road infrastructure improvements in the relevant urban area, which became visible shortly after the introduction of each scheme. In the case of Oslo for example, public support for the scheme increased from 38% in 1991 to 48% in 1997, when it reached its maximum (Odeck & Br˚athen, 2002). It then dropped to 36% in 2001, when fees increased to finance the “Oslo Package 2.” The Norwegian experience with urban tolls suggests that political consensus is more important than public support. Public support may fluctuate. On the other hand, since all proposed and tabled toll projects are initiated locally and sanctioned in Parliament, wide political acceptance is crucial.

Norwegian Urban Tolls

247

The different urban toll rings in Norway have achieved the objective of financing transport projects, for which they were implemented in the first place. The question is whether they could be extended or modified to manage traffic demand at peak times and reduce congestion as well. The necessary amendments to the Road Act and Road Traffic Act have been completed and this already enables local authorities to introduce congestion charging if they wish to. The authorities in Bergen and Oslo are currently considering the possibility of managing congestion with some kind of time-varying toll. Toll fees might need to be increased in order to discourage drivers from crossing the cordon at congested times. This could of course have equity implications, as discussed in Chapter 8 of this volume for the case of Washington, DC. The amendments to the Road Traffic Act ensure that toll revenues in a congestion charging scheme can only be used for local transport purposes. If revenues are allocated in such a way that they are not only returned to the transport sector but they are also used to compensate losers (for example through subsidised public transport fares) the problem of potential regressivity could be at least in part, solved.

NOTES 1. Toll companies in Norway are responsible for collecting tolls as well as for borrowing the necessary money in order to finance a given package. At any time toll revenues have to cover the cost of toll collection, and the amortisation of previous loans. The net toll income plus any new loan goes to finance the package that the toll is supposed to finance (road infrastructure, public transport, etc.). 2. Half of the lanes are now reserved for drivers with seasonal passes and video-enforced. In that sense, the system is not totally manual any longer. 3. The proceeds from a congestion pricing scheme can be used for investment on road and public transport infrastructure as well as road and public transport operation or maintenance. 4. Oslo has a population of about 512,000. 5. There were originally 17 toll stations. 6. The toll fee for light vehicles when the scheme was implemented was originally 10 NOK. 7. There were originally 18 toll stations. 8. The E6-East toll station has high revenues and high operating costs because it operates 24 hours a day seven days a week.

ACKNOWLEDGMENTS The authors wish to thank James Odeck, from the Public Roads Adminstration, as well as different local authorities for providing recent information on toll rings

248

FARIDEH RAMJERDI ET AL.

in Norway, and Astrid Horrisland, from the Institute of Transport Economics, for her support in library services. The authors are also grateful to Georgina Santos and three anonymous referees for their helpful comments on an earlier draft of this paper.

REFERENCES G´omez-Ib´an˜ ez, J. A., & Small, K. A. (1994). Road pricing for congestion management: A survey of international practice. Washington, DC: National Research Council, National Academy Press. Gylt, S. (1991). Manuelle traffikktellingen. Oslo: Scandiplan. Hau, T. (1992). Congestion pricing mechanisms for road: An evaluation of current practice. World Bank Policy Research Working Paper Series 1071, The World Bank, Washington, DC. Jones, P. M. (1998). Urban road pricing-public acceptability and barriers to implementation. In: K. Button & E. Verhoef (Eds), Road Pricing, Traffic Congestion and Environment. Cheltenham: Edward Elgar. Larsen, O. I. (1988). The toll ring in Bergen, Norway – The first year of operation. Traffic Engineering and Control, 29, 216–222. Larsen, O. I., & Østmoe, K. (2001). The experience of urban toll cordons in Norway: Lessons for the future. Journal of Transport Economics and Policy, 35, 457–471. Meland, S. (1995). Generalised and advanced urban debiting innovations. The GAUDI project 3. The Trondheim toll ring. Traffic Engineering and Control, 36, 150–155. Meland, S., & Polak, J. (1993). Impact of the Trondheim toll ring on travel behaviour. In: Proceedings of PTRC (Planning and Transport Research and Computation) Seminar F, Transport Policy and its Implemenation, 103–117. Nordheim, B., & Sælensminde, K. (1991). Effekter av bomringen p˚a kollektivtransporten, PROSAM del rapport, Oslo: PROSAM. Odeck, J., & Br˚athen, S. (1997). On public attitudes towards implementation of toll roads – The case of Oslo. Transport Policy, 2, 89–98. Odeck, J., & Br˚athen, S. (2002). Toll financing in Norway: The success, the failures and perspectives for the future. Transport Policy, 9, 253–260. Ramjerdi, F. (1992). Road pricing in urban areas: A means of financing investment in transport infrastructure or improvements in resource allocation, the case of Oslo. In: Selected Proceedings of the 6th World Conference on Transport Research, Lyon, France, July. Ramjerdi, F. (1994a). The Norwegian experience with electronic toll rings. In: Proceedings of the International Conference on Advanced Technologies in Transportation and Traffic Management, Centre for Transportation Studies, Nanyang Technical University, Singapore, May. Ramjerdi, F. (1994b). An evaluation of the impact of the Oslo toll scheme on travel behaviour. In: B. Johansson & L. G. Mattsson (Eds), Road Pricing: Theory, Empirical Assessment and Policy. Dordrecht/Boston/London: Kluwer. Solheim, T. (1992). Bompengeringen i Oslo – Effekter p˚a trafikk og folks reisevaner Sluttrapport fra før-etter undersøkelsen. PROSAM Rapport N◦ 8, Oslo. Tretvik, T. (1992). The Trondheim toll ring: Applied technology and public opinion. In: R. L. Frey & P. M. Langloh (Eds), The Use of Economic Instruments in Urban Travel Management (pp. 69–78). Wirtschaftswissenschaftliches Zentrum der Universit¨at, Basel.

Norwegian Urban Tolls

249

Vold, A., Minken, H., & Fridstrøm, L. (2001). Road pricing strategies for the Greater Oslo Area. TØI Report 507/2001, Oslo. Wærsted, K. (1992). Automatic toll ring no stop electronic payment system in Norway-Systems layout and full scale experience. Proceedings of the 6th International Conference on Road Traffic Monitoring and Control, Conference Publication 355, London: IEEE.

This Page Intentionally Left Blank

11.

URBAN ROAD PRICING IN THE U.K.

Georgina Santos 1. INTRODUCTION In the early 1960s the Ministry of Transport commissioned a panel chaired by Prof. R. J. Smeed to look into the possibility of road pricing in the U.K. The report (Ministry of Transport, 1964), which put forward the idea of pricing roads as a function of congestion costs, was the first study to consider the ideas advanced by Walters (1961) as a practical possibility. The conclusion was that practical pricing methods could be devised and that pricing would yield a net gain to the community from higher speeds, even before accounting for benefits from less noise and fumes. Although throughout the years the Department for Transport, under its different names, has always been interested in the area of road pricing, it was not until the year 2000 that the legislation was passed. Before then, studies on road pricing design, potential impacts and technologies in the U.K. had not really gone beyond desk calculations and occasional on-site testing. Cambridgeshire County Council for example conducted a field trial in the city of Cambridge in the period October 1990 to October 1993 (Ison, 1996). The Council was keen to pursue the development of road charging but did not have the necessary legal powers to introduce a congestion charging scheme at the time. The idea the Council had was to promote its own Act of Parliament. This would have given it powers to require the fitting of meters to cars and charge for road use. In reality, it was hoped that when the process was started the Government would step in and produce an Act itself. The

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 251–282 Copyright © 2004 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09011-0

251

252

GEORGINA SANTOS

experiment in Cambridge however found public and political opposition and the project never went ahead. It was only in December 2000 that the Parliament passed the Transport Act. There were two documents that prepared its way. In the first one, A New Deal for Transport: Better for Everyone (Department of the Environment, Transport and the Regions, 1998a), the government expressed its intention to reduce the problems of congestion and pollution and announced that they would introduce legislation to allow local authorities to charge road users and to levy a new charge on workplace parking. Later that year, a second document came out, Breaking the Logjam (Department of the Environment, Transport and the Regions, 1998b). In this consultation document economic instruments to manage road congestion were contemplated, including road user charges and workplace parking levies, both earmarked to local transport. Finally, the Transport Act (Acts of Parliament, 2000) gave local authorities in England and Wales powers to introduce road user charges and/or workplace parking levies to tackle congestion when it appeared it would help achieve the policies in the charging authority’s local transport plan. The Act requires spending of (net) revenues on measures for improving local transport and guarantees that the revenues will be hypothecated for at least ten years to the local authority introducing a scheme within ten years of the legislation coming into effect.1 The Act also allows for joint schemes, including ones involving London authorities, as long as the order has been submitted to and confirmed by the Greater London Authority. A similar legislation was passed in Scotland in 2001. The Transport (Scotland) Act 2001 (Acts of the Scottish Parliament, 2001) grants local authorities powers to introduce road user charging, though not workplace parking levies, with the purpose of facilitating the achievement of policies in their transport strategy. Although nothing is said about authorities keeping revenues for ten years, the Act states that charging schemes should be introduced only where the charging authority has a local transport plan. The intention is that local authorities will have an estimate of how much the scheme will raise and what local transport improvements will be funded with the net proceeds. In 1999 the Mayor of London had been given legal powers to implement road user charging and/or workplace parking levies by the Greater London Authority Act 1999 (Acts of Parliament, 1999). That legislation states that Transport for London (TfL), any London borough council or Common Council may establish and operate, solely or jointly, road user charging schemes and/or workplace parking in Greater London.2 It does not allow both TfL and a borough to introduce road user charges or workplace parking levies in the same part of London. It does

Urban Road Pricing in the U.K.

253

however allow road user charges and workplace parking levies to be implemented in the same area. Local traffic authorities in England, Wales and Scotland can introduce road user charging if they wish to do so.3 As of October 2003 there are two road user charging schemes in operation: the London Congestion Charging Scheme and the Durham Scheme. The City of Edinburgh Council has been contemplating the possibility of a cordon toll since the Scottish Act was passed. This chapter was written only a few months after the London Congestion Charging Scheme was implemented. The scheme in Durham had been in place for only a year, and plans for Edinburgh were still at very early stages. As a consequence of that, present tense is used throughout and the results and conclusions can only be seen as preliminary. Section 2 concentrates on the London experience. Section 3 describes the scheme in Durham and its impacts. Section 4 reports the plans for Edinburgh. Section 5 gives some conclusions.

2. THE LONDON CONGESTION CHARGING SCHEME The London Congestion Charging Scheme started on February 17, 2003. As stated above, powers to introduce congestion charging were granted to Transport for London in the Greater London Authority Act 1999 (Acts of Parliament, 1999). The legal framework for congestion charging in London is contained in the Consolidated Scheme Order (TfL, 2003a), and its supporting documents. These set out how and where the Scheme operates, who qualifies for a discount or exemption, and how the revenues are to be spent. 2.1. The Run-Up to the Scheme The introduction of congestion charging was a central part of the Mayor Ken Livingstone’s manifesto for election in May 2000. The Review of Charging Options for London (ROCOL) report had been published in March that same year (ROCOL Working Group, 2000). The report was prepared by an independent group of transport professionals and overseen by the Government Office for London. It advised on how the new road user charging and workplace parking levy powers in the Greater London Authority Act 1999 could be put into practice by the Mayor, and what their impact might be. After being elected, Ken Livingstone decided to take forward the ROCOL proposals for a Central London Congestion Charging Scheme and he set out his

254

GEORGINA SANTOS

initial ideas in a document entitled “Hearing London’s Views” (TfL, 2003b). This document was sent to key stakeholders including local councils, businesses and road user representatives in July 2000. The comments on this document helped shape the Mayor’s draft Transport Strategy, which was published in January 2001. Following public consultation, where interested parties were given the opportunity to comment on the draft Transport Strategy including the proposed Central London Congestion Charging Scheme, the Mayor published his final Transport Strategy in July 2001. The proposed Congestion Charging Scheme was then sent out for public consultation in its own right from July to September 2001. 6,000 notices were placed 250 metres apart on streets in and around the London Inner Ring Road. Consultation meetings were held with key stakeholders. 66,000 public information leaflets on the proposed Scheme were distributed to all the 33 London boroughs. Also, details of the Scheme and how to participate in the consultation exercise were published in eleven London newspapers and broadcast on eleven London radio stations. Following the publication of the proposed modifications to the Scheme in November 2001, there was a further consultation period until January 2002. This involved contacting everyone who had responded until then. An advertisement in the Evening Standard detailing the amendments and inviting further comments was also published. Again, 6,000 notices of the proposed modifications to the Scheme were placed 250 metres apart on streets in and around the charging zone. Meetings were also held with key stakeholders. Eventually, the Mayor confirmed the Order on February 26, 2002. This Order was subsequently modified several times until February 14, 2003 (TfL, 2003c). Once the Scheme Order was confirmed, informing the public and the business community became a priority. The idea was to get people to register for discounts and exemptions before the end of January 2003. Discounts and exemptions are explained below, but it can be advanced here that roughly 22,000 car owning residents inside the charging zone and around 100,000 Blue Badge holders,4 as well as a number of other vehicles, qualified for exemptions or discounts and needed to get engaged with the system before the Scheme started. If all these people had waited until middle February to register it would have simply not worked. There was also a massive public information campaign. Three million leaflets were produced and delivered to every single household in London twice, once before Christmas and once after. The leaflets contained information on where the charging area was, who would be affected and what those affected would need to do. Information on the Scheme was broadcast on all main radio stations and TV channels, and published in most newspapers. A website with information and the possibility of making inquiries was also opened in July 2002.

Urban Road Pricing in the U.K.

255

2.2. How the Scheme Operates 2.2.1. Charging Area and Times The Scheme is an area licensing one. All vehicles entering, leaving, driving or parking on a public road inside the zone between 7 AM and 6.30 PM Monday to Friday, excluding public holidays are charged £5 (around $8). The charging period was originally going to finish at 7 PM but it was subsequently changed to 6.30 PM, so as not to damage the West End and the entertainment industry. Similarly, the proposed £15 ($24) charge for Heavy Goods Vehicles was reduced to £5, following the consultation process. Figure 1 shows the limit of the area, the Inner Ring Road, which runs along Euston Road, Pentonville Road, City Road, Old Street, Commercial Street, Tower Bridge Road, New Kent Road, Kennington Lane, Vauxhall Bridge Road, Park Lane, Edgware Road and Marylebone Road. No charge is made for driving on the Inner Ring Road itself. As it can be seen on Fig. 2, the charging area is relatively small. It only covers 21 km2 (8.4 mi2 ), representing 1.3% of the total 1,579 km2 (617 mi2 ) of Greater London. There are 174 entry and exit boundary points around the zone. Traffic signs make it very clear where exactly the charging zone is. These are accompanied by a red symbol on each lane of traffic at the entry points to the charging zone. Figure 3 shows some of the signs around London. The charge is not linked to inflation. If it were to be linked to inflation a modification of the Scheme Order would be needed. 2.2.2. Exemptions and Discounts The Scheme allows for a variety of 90–100% discounts, as well as exemptions. A summary is shown in Table 1. In addition to the exemptions and discounts described on the table, National Health Service staff and firefighters may be able to claim a refund of the congestion charge from their employers for certain operational journeys undertaken as part of their work. Patients who are clinically assessed as being unable to travel to an appointment by public transport may also be eligible to claim a refund of the congestion charge from their treating hospital.5 Fleet operators with a minimum fleet of 25 vehicles not eligible for discounts or exemptions can pay the congestion charge in the same way as other drivers or they can make use of the fleet schemes. There is an administration charge of £10 ($16) per vehicle per year. Rather than having each vehicle pay the charge each day the whole fleet can be registered. The Fleet Operator pays monthly in advance for anticipated vehicle movements (based on previous average use). Monthly

256 GEORGINA SANTOS

Fig. 1. London Congestion Charging Area. Source: TfL (provided on request). Reproduced with permission from TfL.

Urban Road Pricing in the U.K.

257

Fig. 2. London Congestion Charging Area in relation to Greater London. Source: TfL (provided on request). Reproduced with permission from TfL.

payments are adjusted each month according to actual vehicle congestion charges incurred during the previous month. At the end of each month a statement of vehicle congestion charges is issued (TfL, 2003e). 2.2.3. Methods of Payment The charge has to be paid in advance or on the day until 10 PM with late payment available between 10 PM and midnight but with the charge rising to £10 ($16). The charge can be paid daily, weekly, monthly or yearly. There are several methods of payment. According to the Six Months On report (TfL, 2003f), the percentage split between all sales channels in the first six months was 35% via retail (outlets in selected shops, petrol stations and car parks), 20% via the call centre (of which 7% used Interactive Voice Recognition), 25% via the Internet, 19% via Short Message Service (SMS) on mobile phones, and 1% by post, which were mainly payments made by residents when registering for the discount. 2.2.4. Enforcement and Penalties There are no toll-booths or barriers around the charging zone and no physical tickets or passes. Enforcement is undertaken with video cameras. There is a network of

258

GEORGINA SANTOS

Fig. 3. Congestion Charging Signs in Central London. Source: TfL. www.cclondon.com/ signsandsymbol.shtml. Reproduced with permission from TfL.

203 camera sites, with these located at every entrance and exit to the congestion charging zone as well as inside the charging zone (TfL, 2003g). Each camera site consists of at least one colour camera plus a monochrome camera for each lane of traffic being monitored. The cameras provide high-quality video signals to Automatic Number Plate Recognition (ANPR) software, which reads and records each number plate with a 90% accuracy rate. At midnight, images of all the vehicles that have been in the congestion charging zone are checked against the vehicle

Urban Road Pricing in the U.K.

259

Table 1. Exemptions and Discounts. Discount/Status

Category

Fully exempt

Motorcycles, mopeds and bicycles Emergency vehicles Public service vehicles with 9 or more seats licensed as buses Vehicles used by disabled persons that are exempt from VEDa Licensed London taxis and mini-cabs

100% discount with free registration

Certain military vehicles Local government service vehicles (e.g. refuse trucks, street maintenance) Vehicles with 9 or more seats not licensed as buses (e.g. community minibuses)

100% discount with a one-off £10 registration

Vehicles driven for or by individuals or institutions that are Blue Badge holders

100% discount with £10 registration per year

Alternative fuel vehicles – requires emission savings 40% above Euro IV standards Roadside assistance and recovery vehicles (e.g. motoring organisations such as the Automobile Association)

90% discount with £10 registration per year

Vehicles registered to residents of the central zone

Source: TfL (2003d). www.cclondon.com/exemptions.shtml. a VED: Vehicle excise duty.

registration numbers of vehicles which have paid their congestion charge for that day. The computer keeps the registration numbers of vehicles that should have paid but have not done so. A manual check of each recorded image is then made and a Penalty Charge Notice of £80 ($126) is then issued to the registered keeper of the vehicle. As with parking penalties, this amount is reduced to £40 ($63) for prompt payment within 14 days. Failure to pay the penalty charge within 28 days results in the penalty being increased to £120 ($190). Since the start of the Scheme, an average of 106,200 penalty charge notices have been issued per month. Representations against penalty charge notices have fallen from 62% in the early weeks of the Scheme to 16% six months later. Representations accepted by TfL have also fallen from 66% in June to 55% in October (TfL, 2003f). As of September 2003, less than 2% of all penalty charge notices issued up to then had been appealed against.6 Vehicles with three or more outstanding congestion charging penalty charges, with no representation or appeal pending, may be clamped or removed, not only in the charging zone but also anywhere in Greater London. As of October 2003 the clamp fee is £45 and the removal fee is £125. In addition to that there is a

260

GEORGINA SANTOS

storage cost of £15 a day. If the offender does not pay for the clamp or removal fee and the storage cost, the vehicle may be disposed at auction or by scrapping. The registered keeper will remain liable for all outstanding charges, including a £60 disposal fee (TfL, 2003g).

2.3. Impacts on Traffic According to both the Three Months On (TfL, 2003h) and the Six Months On (TfL, 2003f) reports, the average travel speed in the charging zone is 17 km per hour (10.6 mi per hour), 21% higher than the average speed pre-charging, which was 14 km per hour (8.7 mi per hour), computed as the average of the average speeds during the AM peak, the Inter-peak, and the PM peak throughout 2002, and shown in Table 3.3 of the TfL First Annual Report (TfL, 2003i, p.52). Congestion is defined by TfL as “the difference between the average network travel rate and the uncongested (free-flow) network travel rate in minutes per vehicle-kilometre” (TfL, 2003i, p.46). Using the uncongested network travel rate of 1.9 min per km (approximately 32 km per hour) from TfL (TfL, 2003i, p.52), and pre- and post-charging average travel rates of 4.2 and 3.5 min per km respectively, it can be seen that congestion has decreased from 2.3 to 1.6 min per km. It should be noted that the optimal amount of congestion is not zero congestion. Zero congestion would suggest an under-utilisation of road-space. TfL recognises that there is an “optimal” level of congestion, which is achieved at the “optimal” level of traffic. At the same time it considers the optimal level of congestion difficult to define and that is the reason why it defines congestion using free-flow time as the base (TfL, 2003i, p.45). TfL’s definition of congestion reduction, although practical, is slightly misleading in terms of illustrating the movement towards an optimal level of congestion. Bearing that caveat in mind, it can be concluded that congestion has been reduced by 30.4%. This calculated value is not too different from the values quoted in the Six Months On report (TfL, 2003f, point 3.15) of 32% in March/April, 28% in May/June, and 38% in July/August, and seems to be in line, although at the upper end, with TfL’s expectations of congestion reduction of between 20 and 30%. The Six Months On report also notes that the decreased congestion is mainly the result of less time spent stationary or moving in queues, rather than the result of higher driving speeds (TfL, 2003f, point 3.20). The total number of vehicles with four or more wheels entering the zone during the charging hours has been reduced by 16% (TfL, 2003f, point 3.37), just above TfL’s expectation of a 10–15% reduction. When Spring 2002 (pre-charging) and Spring 2003 (post-charging) are compared, some changes in vehicle counts can be observed. For example, the count of all cars entering the central zone decreased by

Pedal Cycles

Motor Cycles

Cars

Taxis

Bus & Coach

LGVs

HGVs & Other

4+ Wheels

Total

Total Non-Cars

Pedal and Motorcycles

13,836

25,840

193,912

55,971

13,393

53,780

15,329

332,386

372,062

178,149

39,676

11,346

22,940

192,840

57,036

13,079

59,487

16,256

338,697

372,982

180,143

34,285

25,181

48,780

386,752

113,007

26,472

113,267

31,585

671,083

745,044

358,292

73,961

18,131

30,779

133,016

66,836

15,518

48,745

13,476

277,591

326,501

193,485

48,910

12,535

25,426

125,151

64,917

15,735

50,660

14,402

270,865

308,826

183,675

37,961

30,666

56,205

258,168

131,753

31,253

99,405

27,878

548,456

635,328

377,160

86,871

Changes (%) Incoming Outgoing

31 10

19 11

−31 −35

19 14

16 20

−9 −15

−12 −11

−16 −20

−12 −17

9 2

23 11

Total

22

15

−33

17

18

−12

−12

−18

−15

5

17

Spring 2002 incoming Spring 2002 outgoing Spring 2002 total Spring 2003 incoming Spring 2003 outgoing Spring 2003 total

Urban Road Pricing in the U.K.

Table 2. Vehicle Counts Pre and Post Charging.

Note: Incoming and outgoing refer to in and out of the charging zone respectively. LGV: Light Goods Vehicles. HGV: Heavy Goods Vehicles. Source: TfL, data provided on request.

261

262

GEORGINA SANTOS

31%, slightly above the upper end of TfL’s expectations of 17–28% (TfL, 2003f). However, the reduction in the number of cars has been partially offset by increases in incoming taxis and buses in the order of 19 and 16% respectively. This is shown in Table 2, which summarises vehicle counts. It can also be seen from the table that important increases in the number of pedal and motorcycles have occurred. These typically follow commuting patterns (inbound in the morning and outbound in the evening), which suggest an increase in commuting by these modes (TfL, 2003f, point 3.39). Surprisingly, the number of powered 2-wheelers (motorcycles and mopeds) and bicycles involved in accidents following the introduction of the Scheme has fallen by 15 and 17% respectively, when compared to the same period in 2002 (TfL, 2003f, point 3.97). These results probably reflect the long-term trend of declining accidents in London rather than any feature linked to the Scheme. The increase in the use of buses, taxis, motorcycles and bicycles does not jeopardise the reduction in overall traffic. Even though the increase in motorcycle and bicycle movements has been higher than expected (TfL, 2003f, point 3.38), these are less disruptive and therefore impose less congestion. Trips by bus were expected to increase and TfL provided more bus spaces anticipating this. With a larger than expected reduction in incoming vehicles with four or more wheels there seem to be no grounds to extend the charge to include other vehicle categories, like it was done in Singapore.7 Finally, travel times also show significant improvements. Table 3 summarises travel times and their changes. Since traffic travelling on the Inner Ring Road does not pay the congestion charge, TfL expected that through traffic, with origin and destination outside the charging zone, would divert and use the Inner Ring Road instead. However, improved traffic management arrangements were put into place on the Inner Ring Road before the Scheme started and this prevented an increase in congestion. For example, between one and two seconds were taken off green light time on radial roads, which were anticipated would have less traffic, and added on to green light Table 3. Comparison of Return Journey Times Pre and Post Charging.

Travel times to central zone To/from outer London To/from inner London To/from central London Basket of 5,000 journeys Source: TfL (2003h, Fig. 5, p. 5).

Pre-Charging (Min)

Post-Charging (Min)

Change (%)

59 37 38 47

52 33 35 41

−12 −11 −8 −13

Urban Road Pricing in the U.K.

263

Table 4. Average Workday Travel Times into London (Morning Peak). Route

Orientation

Feb-Sep 2002 (Min)

Feb-Sep 2003 (Min)

A1 Mill Hill to Islington A41 Mill Hill, Five Ways to Regent’s Park A40/M Denham to Marylebone A4 Langley, Slough to Talgarth Road A30 Stanwell to Osterley A316 Sunbury Cross to Ravenscourt Park A3 Cobham to Clapham A23 Hooley to Brixton A20 Swanley to Eltham A2/A102 Dartford to Blackwall Tunnel A12 Harold Wood to Blackwall Tunnel A10 Waltham Cross to Stoke Newington

Change (%)

N-NW NW

46 36

49 38

7 6

W

62

61

−3

W

57

57

1

W-SW W-SW

15 40

16 41

6 2

SW S SE E-SE

50 59 20 39

62 64 23 44

25 8 14 14

E-NE

66

72

9

N

44

51

15

Source: Trafficmaster PLC (data supplied on request).

time on the Inner Ring Road. That made a sufficient difference to keep the Ring Road operating satisfactorily with marginally lower levels of congestion, when compared to pre-charging conditions. Although the Six Months On report states that congestion on main roads across Inner London outside the charging zone has not increased (TfL, 2003f, point 3.30) it also states that further surveys and analysis are needed to fully understand any congestion changes across Inner London (TfL, 2003f, point 3.33). Trafficmaster, a private company that provides real-time traffic information on major routes, started a study on February 17, 2003 to assess the commuting impacts of the Scheme outside the zone. Although over the period February/September 2003 travel time on most routes has increased, there are no constant patterns and it is difficult to draw any final conclusions. Table 4 shows the average of travel times (in minutes) and travel time changes for the first seven months after the Scheme was introduced. Table 5 gives some examples of how travel time on some routes has increased and decreased in comparison to the same month the year before, and shows that there is no constant pattern. The apparent contradiction between Tables 3 and 4 can be explained by the fact that Table 3 measures travel times all the way up to and including the

264

GEORGINA SANTOS

Table 5. Examples of Non-Constant Patterns in Percentage Changes in Travel Times. Route

A1 Mill Hill to Islington A41 Mill Hill, Five Ways to Regent’s Park A40/M Denham to Marylebone A4 Langley, Slough to Talgarth Road A30 Stanwell to Osterley A316 Sunbury Cross to Ravenscourt Park A3 Cobham to Clapham A23 Hooley to Brixton A20 Swanley to Eltham A2/A102 Dartford to Blackwall Tunnel A12 Harold Wood to Blackwall Tunnel A10 Waltham Cross to Stoke Newington

April ‘02 (Min)

April ‘03 (Min)

Change (%)

May ‘02 (Min)

May ‘03 (Min)

Change (%)

50.1 35.8

51.7 38.4

3 7

55.4 42

53.9 46.4

−3 10

65.4

68.9

5

62.6

58.7

−6

58.6

56.1

–4

64.4

59.7

−7

16.5 42.9

17.5 39.7

6 –7

14.8 44.3

17.7 48.4

20 9

58.4 66 23 46.7

59.3 66.7 23.2 46.7

2 1 1 0

51.2 64.8 21.7 40.7

66 71.6 26 45.8

29 10 20 13

72.5

70.7

–2

71.7

69.4

−3

55.8

58.1

4

53.4

57

7

Note: Both months are within school term. Source: Trafficmaster PLC (data supplied on request).

charging zone, where travel times have decreased and therefore push the average down. Table 4 on the other hand, does not include any road inside the charging zone.

2.4. Impacts on Public Transport TfL estimates that between 50 and 60% of the reduction in car trips is the result of car users having transferred to public transport (TfL, 2003f, point 3.64). This represents a 2% increase in overall public transport passenger levels coming into the zone. The reason for this percentage increase to be small is that before the Scheme was implemented the majority of commuters used public transport anyway. In the morning peak, 7 to 10 AM, between 14,000 and 15,000 additional bus passengers were expected on buses serving the charging zone, 7,000 of which were expected to travel between 8 and 9 AM. The results for the first six months

Urban Road Pricing in the U.K.

265

are in line with expectations. In order to cope with this expected increase in demand TfL had provided 11,000 extra spaces on London’s buses during the peak hour by the time the Scheme started. This was achieved with a combination of more frequent services, new and altered routes, and bigger buses. Average bus speeds in the morning peak have not changed too much and it is difficult to establish a pattern of variation. Whilst speeds on some route sections have increased, speeds on others have decreased (TfL, 2003f, point 3.78). On the plus side, additional time waited by passengers over and above the route schedule has been reduced by 25% across Greater London and by over 33% in the routes serving the charging zone and the Inner Ring Road (TfL, 2003f, point 3.75). TfL had predicted that approximately 5,000 commuters would switch from the car to the Underground (TfL, 2003j). This did not happen. Underground usage across London and specially in Fare Zone 1 decreased. The reason for this decrease is obviously not related to the congestion charge in any way. If anything the congestion charge might have caused a marginal increase in demand. The reasons for the decrease in passenger levels on the London Underground are probably linked to the slowdown of the economy and the decrease in tourism in London, which in turn may be linked to the war in Iraq (TfL, 2003f). In addition to that, the Central Line was temporarily closed for almost three months following a derailment at Chancery Lane station in January. No significant changes in demand for trips by rail have resulted from the Scheme, and this is in line with TfL’s expectations (TfL, 2003f).

2.5. Generalised Cost Elasticity of Demand Using the changes in speed and traffic levels registered after the Scheme was implemented, the elasticity of demand for trips by car with respect to generalised travel costs in London can be computed. The generalised costs include both time and distance costs. To calculate these the U.K. Automobile Association estimates of fixed and variable motoring costs (monetary costs) were used, together with the time savings reported in the Three Months On report (TfL, 2003f) and value of travel time savings (VTTS) estimates. Two sets of VTTS estimates were produced. The first one follows the recommendations of the Transport Economics Note (Department of the Environment, Transport and the Regions, 2001), and the second one follows the recommendations set out in Mackie et al. (2003). The working value of time from the Transport Economics Note was updated to 2003 prices and increased by 34%, to reflect the difference in the average earnings

266

GEORGINA SANTOS

Table 6. Motoring Costs at 2003 Prices. Standing costs (£ annual) Road tax Insurance Cost of capital Depreciation Standing costs total (pence per km)

160 448 412 2080 35.2

Running costs (pence per km) Fuel Service: tyres, parts and labour Parking Running costs total (pence per km)

7.18 3.95 1.13 12.25

Total costs (pence per km)

47.48

Note: The numbers in the table correspond to a car that costs between £13,000 and £20,000 if bought new. Average distance driven by a Londoner is assumed to be 8,800 km (5,466 mi) (TfL, 2001, p. 2). Source: U.K. Automobile Association (2003). www.theaa.com/allaboutcars/advice/advice rcosts home.html.

in London as indicated in the New Earnings Survey 2003 (National Statistics, 2003). The non-working value of time was also updated to 2003 prices. Two categories of trips were considered: working, as trips made in the course of work, excluding commuting, and non-working. Although commuting trips are non-working they tend to have higher VTTS than shopping or leisure trips because of the potential penalty for arriving late. Thus, for the second set of calculations, three categories of trips were considered: working, commuting, and other. 10% of all trips made by car in London are for business purposes, 26% for commuting, and 64% for other purposes such as leisure, shopping, etc.8 The value of working time was derived from the gross weekly earning in London indicated in the New Earnings Survey 2003 (National Statistics, 2003) and increased by 24.1% to include the non-wage labour costs for the employer, as recommended in the Transport Economics Note. The VTTS for commuting and other purposes for different income levels were taken from Mackie et al. (2003), updated to 2003 values, and calibrated to the average salary in London. Tables 6 and 7 summarise the numbers used in the calculations. Table 8 presents the results. The generalised cost elasticity values presented in Table 8 are not out of line with those reported in the literature. Dodgson et al. (2002) for example summarise various studies for Singapore suggesting point elasticities in the order of −0.12 to −0.35 with respect to congestion charges. The ROCOL report (ROCOL Working

Urban Road Pricing in the U.K.

267

Table 7. Details of an Average Trip by Car in London (2003 Prices). Distance travelled per daya Time travelled per day (pre-charging)b Time savings per day (post-charging)b Vehicle costs Running costs

23.4 km 94 min 12 min 47.5 pence/km 12.25 pence/km

Value of time per car (i) (ii) Average occupancyc

11.7 pence/min 12.7 pence/min 1.35

Note: Estimate (i) follows the Transport Economics Note (Department of the Environment, Transport and the Regions, 2001) and estimate (ii) follows Mackie et al. (2003). Details are explained in the text. a Average car trip in London is 11.7 km (TfL, 2002, Table 7.1, p. 25) and two trips per day are assumed. b Average travel time and time savings taken from the basket of 5,000 journeys in Table 3. c TfL (2002), Table 5.1, p. 19.

Group, 2000, p.154) gives arc elasticities with respect to a £5 charge for numbers of trips that range between −0.03 and −0.63, depending on the method used, the journey purpose, and the other charge assumed (£2.50 or £10). Goodwin (1992) suggests fuel price elasticities for vehicle-kilometres (veh-km) of −0.16 for the short run and −0.32 for the long run. These numbers are still considered standard values for the responsiveness of car travel demand with respect to changes in fuel prices (Graham & Glaister, 2002). As a general rule, the demand elasticity with respect to generalised cost changes is roughly equal to the fuel price elasticity divided by the fuel share of generalised cost (Dodgson et al., 2002, p. 28). In heavily congested London, where time costs have a high share of total costs, the share of fuel costs can be estimated at roughly 7–15%, where the upper range corresponds to cases where only running costs are taken into account. Therefore, Goodwin’s value of −0.16 for short-run fuel price elasticity corresponds to a generalised cost elasticity of between −1.1 and −2.3, not too different from the values presented in Table 8. Goodwin et al. (2004) review 69 empirical studies published since 1990 and conclude that the fuel price elasticity for volume of traffic ranges from −0.10 for the short run (one year), to −0.30 for the long run (around five years). This slightly lower short run elasticity yields lower elasticities than those of Table 8 when divided by the fuel share of generalised cost. The elasticities from Table 8 were computed with data collected after the Scheme was implemented. Traffic reduction has been higher than expected, which means that elasticities might have been underestimated prior to the implementation of the Scheme. Goodwin (2003) suggests that elasticities were revised down by a

268

GEORGINA SANTOS

Table 8. Elasticities of Demand for Car Trips with Respect to Costs. All GC Occupancy rate GC per day, £ (I) (ii)

1.35 22.1 23.0

GC Exc. Fixed Costs 1 19.2 19.9

1.35 13.8 14.8

1 11.0 11.7

Toll, £

5.00

5.00

5.00

5.00

Time Benefits, £ (i) (ii)

1.40 1.52

1.04 1.13

1.40 1.52

1.04 1.13

Reliability Benefits,a £ (i) (ii)

0.35 0.38

0.26 0.28

0.35 0.38

0.26 0.28

−3.25 −3.09

−3.70 −3.59

−3.25 −3.09

−3.70 −3.59

Change in GC, £ (i) (ii) Change in GC, % (i) (ii)

−14.7 −13.5

−19.3 −18.0

−23.5 −20.7

−33.7 −30.7

Change in Demand,b %

33

33

33

33

Elasticity (i) (ii)

−2.2 −2.5

−1.7 −1.8

−1.4 −1.6

−1.0 −1.1

Note: GC: Generalised Costs. Estimates (i) and (ii) correspond to the two sets of values presented in Table 7. a Dodgson et al. (2002) argue that reliability benefits are worth approximately 25% of time benefits. b Table 2.

sort of “ratchet” effect from one study to the next, probably because their authors wanted to be conservative, and would always choose the lowest estimate. Before concluding this section, it should be noted that the estimates presented in Table 8 are based on vehicle counts, not on veh-km. The latter however would be a more appropriate statistic, as fewer cars travelling on average more km could compensate the reduction in vehicle counts. Data on changes in veh-km will only become available in Summer 2004 and it is therefore not possible at this stage to compute an elasticity using veh-km. Preliminary estimates however point to a reduction in veh-km driven by vehicles with four or more wheels of between 5 and 15% (TfL, 2003f, point 3.49). Since there is no data on veh-km reduction for cars, choosing a number would be pure speculation.

Urban Road Pricing in the U.K.

269

2.6. Marginal Congestion Cost If traffic is assumed to be homogenous inside the charging zone, area marginal congestion costs (MCC) can be computed. Speed is assumed to have risen from 14 km per hour to 17 km per hour (TfL, 2003f). The calculations are done using passenger car units (PCUs), which give a measure of the relative disruption that different vehicle types impose on the network. A car has a PCU rating of 1, a Light Goods Vehicle (LGV) has a PCU rating of 1.5, a Heavy Goods Vehicle (HGV) has a PCU rating of 2.5 or 3, a bus has a PCU rating of 2.5, a motorcycle has a PCU rating of 0.5, and a bicycle has a PCU rating of 0.2.9 Average VTTS within the central zone were estimated at 29.6 and 35.1 pence per PCU-min at 2003 prices before and after the Scheme was implemented. The reason for the higher post-charging value is that although approximately the same number of people enter and leave the charging zone, there has been a shift from cars to bicycles and motorcycles, which have lower PCU ratings. Although there has been an increase in bus counts, there has been a higher decrease in LGVs and HGVs. The decrease in PCUs due to the decrease in LGVs and HGVs’ counts is almost three times the increase in PCUs due to increased bus ridership. The average VTTS expressed in pence per PCU-min post-charging is thus higher than the VTTS pre-charging. The average VTTS were estimated using the shares of traffic as implied by Table 2, their associated PCU values, occupancy rates as given in the London Travel Report 2002 (TfL, 2002, Table 5.1, p.19) and in the Transport Economics Note (Department of the Environment, Transport and the Regions, 2001, Tables 2/3), working, commuting and other values of time as explained in Section 2.5, and trip purpose by vehicle type as provided by TfL on request.10 Following the practice set for the London Congestion Charging Research Programme (MVA, 1995) and for the ROCOL study (ROCOL Working Group, 2000) a single value of time was used for all modes in the case of working time, and a single value of time was used for all person types in the case of non-working time. The increase in speed was 21% and the decrease in traffic volume, for all vehicle types measured in PCUs, was 15%. Using the standard Eq. (1a) MCC =

b e sq s(q)

(1a)

where b is value of time in pence per PCU-hour, s is speed in km per hour, dependent on traffic volume in the area, q, in PCUs per hour, and esq is the elasticity of speed with respect to traffic volume, the area MCC in the charging zone was estimated.

270

GEORGINA SANTOS

Equations (1b) and (1c) give the MCC estimates pre- and post-charging respectively: 29.6 × 60 0.21 × = 186.5 pence per PCU-km (1b) 14 0.15 35.1 × 60 0.21 MCC = × = 182 pence per PCU-km (1c) 17 0.15 The area MCC has decreased by 2.4% since the implementation of the Scheme. The values can be converted into pence per veh-km by using the PCU/veh rate for London of 1.13. Thus, MCC is 165 pence per veh-km and 161 pence per veh-km pre- and post-charging respectively. These values are almost twice as high as the value reported by Sansom et al. (2001) of 86 pence per veh-km updated at 2003 prices in a major urban centre such as London. The values estimated here rest on the latest data and empirical evidence and are therefore more reliable. It should be noted that if the £5 congestion charge were to reflect on average the congestion externality, an average vehicle would need to travel a distance of around 3 km per day, which is a reasonable expectation given that the charging zone has a diametre of roughly 5 km. The £5 charge, never computed as a function of MCC, seems to be about right (!). MCC =

2.7. Monitoring Programme A five-year monitoring programme has been set up, one year before the start of charging and four years after. It consists of over 100 surveys and studies, designed to measure and understand the impacts of the Scheme. The Scheme is affecting congestion levels, traffic patterns, public transport, and travel behaviour. Although six months into the Scheme may still be early to draw any conclusions, it is clear that the Scheme will also have economic, social and environmental impacts. All these different aspects are already being assessed by the monitoring programme and annual reports describing and explaining them will be produced. The first one, published in June 2003 (TfL, 2003f), summarises the conditions prior to charging and sets the indicators and issues being monitored since the Scheme was introduced. In September 2003 Westminster City Council conducted a survey to find out how businesses felt about the congestion charge. From all the respondents, 68% have their businesses inside the charging zone, 44% are retailers and 27% are bars and/or restaurants (Westminster City Council, 2003). Almost 69% of the respondents feel the Scheme has had a negative impact on their business, 8% feel it has had a positive impact, and 23% feel it has had no impact. The Six Months

Urban Road Pricing in the U.K.

271

On report (TfL, 2003f) on the other hand, attributes the downturn in retail activity to the reduction in trips in Central London on all modes of transport, including the Underground, which according to the report, accounts for 70% of the decline. The report claims that since the majority of people travel to Central London by public transport, the relative impact of reduced trips by car is small (TfL, 2003f, point 5.5). According to the business survey conducted by Westminster City Council, almost 28% of the respondents are considering relocating outside of the zone as a result of the charge. Whether the Scheme is responsible for the reduction in sales, or other factors such as general economic slowdown, reduction in the number of tourists, and the war in Iraq, remains to be determined. Six months is too short a period to conclude on any definite economic or relocation effects.

2.8. Use of Revenues The idea of earmarking the revenues from congestion charging to the transport sector has been advocated for many years and indeed has often been cited as a powerful tool to gain public acceptability (Goodwin, 1989, 1990; Harrington et al., 2001; Ison, 2000; Odeck & Br˚athen, 1997, 2002; Small, 1992). The ROCOL Consultants carried out a series of surveys between March and August 1999 and found that 67% of the general public thought road user charges in Central and Inner London would be a “good thing” if revenues were spent on a mix of transport improvements. The percentage increased to 73% when respondents’ spending package preferences were introduced (ROCOL Working Group, 2000). The Greater London Authority Act 1999 (Acts of Parliament, 1999) provides that revenues from charging schemes introduced during the first ten years of the legislation coming into force will be earmarked for at least ten years from their implementation to projects included in the Mayor’s Transport Strategy. In fact, any charging scheme in London must include a plan of how revenues will be used during the first ten years. The Mayor’s Transport Strategy includes objectives such as for example reducing traffic congestion, investing on the Underground, improving bus services, integrating National Rail with other transport systems, and improving journey time reliability for car users (Greater London Authority, 2001). Although the revenues from the London Scheme will be lower than predicted, mainly because of the higher than expected reductions in traffic levels and exemptions and discounts, they will still be a substantial £68 million, net of costs in the financial year 2003/04 (TfL, 2003f). The net revenues forecast for 2004/05 onwards are between £80 and £100 million per year, at 2003 prices.

272

GEORGINA SANTOS

Table 9. Preliminary Estimates of Annual Costs and Benefits of the London Scheme (£ Million at 2003 Prices). Annual costs TfL administrative and other costs Scheme operation Additional bus costs Chargepayer compliance costs Total

5 90 20 15 130

Annual benefits Time savings to car and taxi occupants, business use Time savings to car and taxi occupants, private use Time savings to commercial vehicle occupants Times savings to bus passengers Reliability benefits to car, taxi and commercial vehicle occupants Reliability benefits to bus passengers Vehicle fuel and operating savings Accident savings Disbenefit to car occupants transferring to public transport, etc. Total

75 40 20 20 10 10 10 15 −20 180

Source: TfL (2003f, Table 3).

These net revenues will be spent on transport improvements in London. Plans include bus network improvements, contributions to the costs of developing possible tram or high quality segregated bus schemes, safety and security improvement schemes, road and bridge maintenance programs, and improvement of late night public transport. In the medium to long term, revenues could help finance the development and funding of expanded Underground and rail capacity with new services across Central London, together with improved orbital rail services, new Thames Gateway river crossings, schemes to provide improved access to London’s many town centres, and selected improvements to London’s road system (TfL, 2003k).

2.9. Cost-Benefit Analysis Table 9 shows costs and benefits, as detailed in the Six Months On report (TfL, 2003f). The table suggests a net benefit of around £50 million per year. According to the numbers presented in Table 2 and the occupancy rates given in the London Travel Report 2002 (TfL, 2002, Table 5.1, p. 19) and in the Transport Economics Note (Department of the Environment, Transport and the Regions,

Urban Road Pricing in the U.K.

273

2001, Tables 2/3), 52% of all people travelling to or from the charging zone used buses before the Scheme was introduced. If taxi and pedal and motorcycle users are added as well, the total share of people that did not use a chargeable mode of transport before the Scheme raises to 63.9%. These are winners, in the sense that they are enjoying lower congestion without paying a penny, and they do not have any disutility from changing mode because they did not change mode. From a very conservative point of view the remaining 36.1% would be losers. However, those with very high values of time also have a net benefit after paying the congestion charge. In addition to that, there are a number of exemptions and discounts, as explained in Section 2.2. The share of people travelling by car was reduced from 27 to 18%, which means that 9% of the original car users have transferred to some other mode or made alternative arrangements. This is the group listed in the second part of Table 9 as having a disbenefit of £20 million.

3. THE DURHAM CONGESTION CHARGING SCHEME On October 1, 2002 Durham County Council implemented a £2 ($3) charge for all vehicles using Saddler Street and the Market Place between 10 AM and 4 PM, Monday to Saturday (Durham County Council Press and Publicity Unit, 2002). This was the first road toll introduced in the U.K. following the Transport Act 2000. It is a modest scheme as it only operates on Saddler Street, a narrow road, which runs through the heart of the historic city of Durham. The historic city centre is also known as the Peninsula, as it is surrounded by the River Wear. Saddler Street is the only public access road onto the Peninsula, where the Cathedral and the Castle are, as well as some businesses, a school, parts of Durham University and a small number of private houses. The daily number of vehicles using the road before the scheme was implemented was around 2,000. Being a narrow street, only wide enough for one vehicle at a time, this level of traffic presented a significant hazard to the 14,000 pedestrians that also used the road every day (Durham County Council, 2003). The main aim of the scheme was to reduce traffic levels in order to improve pedestrian safety, improve access for the disabled, and enhance the World Heritage Site whilst preserving the viability of the Peninsula as a working part of the city centre. All these objectives have been achieved. The number of vehicles using the road, from the Market Place to the Cathedral fell by between 50 and 80%, depending on the traffic count used as the base (Durham County Council, 2003, Chart 3). Preliminary results also indicate that the number of pedestrians increased by 10%

274

GEORGINA SANTOS

on average, from 14,000–14,500 a day to 15,000–16,000 a day, after the scheme was introduced (Durham County Council, 2003, Chart 13). The charge is paid on exit from the area on a ticketing machine, monitored by Closed Circuit Television (CCTV). Alternatively, payment is accepted before 6 PM at the National Car Parks (NCP) Parking Shop. Although the aim of the scheme was never to raise revenues, the income from the £2 ($3) charge was supposed to cover the costs of the scheme operation as well as the Cathedral Bus. This was a previously underutilised bus service into the area, which was extended and enhanced (Durham County Council, 2003) to link the town’s main car parks, bus and railway stations with the city centre. The reduction in traffic has been greater than expected and the revenues are lower than expected and are therefore not enough to cover the above. Durham County Council however is able to cover the shortfall with revenues from parking charges. A limited number of exemptions are in place and these include permanent residents and those establishments that hold a parking space inside the Peninsula. The Cathedral buses and some service providers such as emergency service vehicles, Royal Mail and recovery vehicles are also exempt. Disabled people can be issued exemption permits by the establishments they choose to visit or can reserve a permit in advance by contacting the NCP Parking Shop. Delivery or service vehicles servicing establishments that do not hold a parking space have to pay the charge if servicing during the restricted period. Although drivers who fail to pay the charge are still permitted to use the road, a £30 ($50) penalty notice is issued to the vehicle registered keeper if payment is not made before the end of the working day. The number of violators however is negligible. Vehicles are recorded on the CCTV system and owners traced through the Driver and Vehicle Licensing Agency (DVLA). Although the scheme is not technologically advanced, it is a measure that has proved effective for the purpose intended: to reduce traffic on one road in the historic Peninsula. Durham County Council took the plan seriously and carried out two major public consultation exercises, backed up by exhibitions and leaflet drops. This meant that by the time the scheme was introduced Durham County Council had obtained 49% public support, which increased to 70%, after the scheme was introduced (Durham County Council, 2003).

4. PLANS FOR EDINBURGH Edinburgh is a historic city and also the capital of Scotland. With a population of almost 500,000, it is a commercial and financial centre, and the site of the new Scottish Parliament.

Urban Road Pricing in the U.K.

275

Forecasts show that if no congestion charging is implemented and investment in public transport is not increased traffic will increase by 20% between 2001 and 2021, and time spent in congested conditions will double (Transport Initiatives Edinburgh, 2002). The 2002 Strategic Regional Consultation “Have your Say” generated 20,000 responses from around the region (TIE, 2002) showing that the majority of the residents of both Edinburgh and South-East Scotland believe congestion has to be reduced and public transport, improved. As explained earlier, the Transport (Scotland) Act 2001 (Acts of the Scottish Parliament, 2001) grants local authorities powers to introduce road user charging with the purpose of facilitating the achievement of policies in their transport strategy. The City of Edinburgh Council has been contemplating the possibility of a cordon toll since the Act was passed. It commissioned Transport Initiatives Edinburgh (TIE), a company 100% owned by the City of Edinburgh Council, the task of looking into different options. TIE found that a single cordon around the city centre would not adequately address the increasing congestion forecast in the outer parts of the city and recommended a double cordon (TIE, 2002). Although, as of October 2003, the plans are not finalised, the reminder of this section describes the scheme under consideration. Subject to the approval of the City of Edinburgh Council and the Scottish Executive, and subsequent to an affirmative referendum in Edinburgh, the scheme could come into force in 2006.

4.1. How the Scheme Will Operate 4.1.1. Charging Area and Times The inner cordon would be operational Monday to Friday except public holidays between 7 AM and 6.30 PM, and the outer cordon would be operational on the same days from 7 AM to 10 AM and from 4 PM to 6.30 PM. Both cordons would finish at 6.30 PM to coincide with parking regulations. A daily charge of £2 ($3) would be levied on vehicles crossing either one or both cordons inbound. In contrast with the London Scheme, after introduction in 2006, the charge would be linked to inflation. The proposed inner charging boundary surrounds the Old and New Towns of Edinburgh that are covered by UNESCO World Heritage Site designation. The proposed outer boundary is inside the outer city bypass, at the edge of the built-up area. A consultation with local residents and businesses will help define the exact boundaries.

276

GEORGINA SANTOS

Table 10. Methods of Payment in Edinburgh. Period of Licence

Method of Payment

One day

Direct debit, telephone, on-line, retail outlet, self-service machine or text message Direct debit, telephone, online or retail outlet Direct debit, telephone, online or retail outlet Post, direct debit, telephone, online or retail outlet

Period of 5 consecutive charging days Period of 20 consecutive charging days Period of 258 consecutive charging days

Source: Draft Edinburgh Congestion Charging Order. www.tiedinburgh.co.uk/iti/downloads/Charging Order-Consultation.DOC.

4.1.2. Exemptions Like in London, emergency vehicles, vehicles used by disabled people, buses, licensed taxis, and motorcycles will all be exempt. City Car Club will also be exempt.11 Finally, residents of Edinburgh, living outside the outer cordon (including Currie, Balerno, Juniper Green, Ratho, South Queensferry, Kirkliston), and will be exempt from paying the charge at the outer cordon (TIE, 2003a). 4.1.3. Methods of Payment The methods of payment will include self-service machines, retail shops, call centre, online, SMS on mobile phones, post, and direct debit. Payments will be made on a daily, weekly, monthly or annual basis. Table 10 details the methods according to the period of licence, as presented in the Draft Edinburgh Congestion Charging Order. 4.1.4. Enforcement and Penalties The scheme will use an Automatic Number Plate Recognition (ANPR) system, as used in London. If a driver enters a cordon and does not pay the standard charge by midnight the same day, a penalty charge will be issued to the registered keeper of the vehicle. This will be the same amount as a parking ticket, with a 50% reduction if the owner pays within a fortnight of the notice being issued, and a 50% increase if the payment is not made within 28 days. As of October 2003, a parking ticket is £60 (TIE, 2003a). Violators with three or more outstanding penalty notices may have their vehicles clamped or removed. These vehicles will only be released if all outstanding penalty charges in respect of that vehicle are paid to the Council, including the clamping, removal and storage fees. If the vehicle has been sold or destroyed, the owner shall also be liable for a penalty charge for its sale or destruction (Draft Edinburgh Charging Order, 2003).

Urban Road Pricing in the U.K.

277

4.2. Integrated Transport Initiative An associated program of transport improvements will be undertaken before the scheme is in place. This program comprises bus and rail improvements, including two new stations for the Edinburgh Cross Rail, the West Edinburgh Busway, new Park & Ride facilities, bus priority lanes, integrated bus ticketing system, and real time information at bus stops. The revenues from the scheme (net of operating costs) will be reinvested in transport related projects within Edinburgh and South-East Scotland. In pragmatic terms for local authorities, the ability to invest in better transport using the revenues generated from charging schemes can be an important objective in its own right, not just a useful by-product of a demand management scheme. In this context, the approach taken by the City of Edinburgh Council is to develop an “integrated transport package,” underpinned by congestion charging in order to achieve congestion reduction, but consisting of a major programme of investment as well. This integrated transport investment programme includes three tram lines, train, bus and highway improvements, and better facilities for cyclists and pedestrians. Although plans are still at an early stage, the figures that are being assumed in October 2003 for the transport investment package (TIE, 2003b), assuming the charging scheme starts in 2006 and operates for twenty years, are as follows:  Government funding will continue to be available for transport projects, with an estimated total of just over £100 million (at 2002 prices) for Edinburgh and the rest of the South-East Scotland Transport Partnership (SESTRAN) area.12  Additionally, specific funding of £375 million (at 2002 prices) has been made available by the Scottish Executive towards a tram system for Edinburgh, subject to a satisfactory business case being made for it.  On the basis of the proposal to implement a congestion charging scheme based on two cordons, the scheme will deliver approximately £900 million (at 2002 prices) on new transport projects between 2006 and 2026.

4.3. Expected Impacts on Traffic The expected impacts from the proposed transport initiative described above, including congestion charging, are presented in Table 11. Congestion is measured as time lost due to congestion, in vehicle-hours per day. All percentages compare the situation with and without the proposed transport initiative.

278

GEORGINA SANTOS

Table 11. Expected Traffic Impacts from the Integrated Transport Initiative for Edinburgh. Congestion (%)

2006 2016 2026

−7 −36 −32

Traffic Across Cordons (%) Outer Inner −31 −34 −25

Orbital Traffic (%)

Travel to City Centre (%)

7 5 6

2 5 6

−9 −15 −10

Travel by Public Transport (%)

2 35 41

Source: TIE (2003c). www.tiedinburgh.co.uk/iti/.

Following an extensive consultation process over 2003 and 2004, the plan is to introduce the toll in 2006. However, the City of Edinburgh Council is committed to holding a referendum on the final proposals before any finalised scheme can be agreed (City of Edinburgh Council, 2002).

5. CONCLUSIONS This chapter describes the London Congestion Charging Scheme, the Durham Scheme, and the plans to introduce congestion charging in Edinburgh. Charging for congestion, obtaining public support to do it, improving public transport to provide an alternative to the car and returning the revenues of the scheme to the transport sector has finally become a reality in the U.K. The systems in place may not be perfect and charges may not equal marginal congestion costs. However, both the Durham and the London schemes seem to be working smoothly. The scheme in Durham is a small scale one with a toll on just one road. Yet, the aim for which it was conceived, to reduce traffic levels on Saddler Street, has been achieved. The London Scheme is undoubtedly unsophisticated. Heavy Goods Vehicles that tend to slow traffic more pay the same charge as cars. Also, vehicles pay the same charge regardless of whether they are driving (and thus causing congestion) all day long or are parked on a public road inside the charging zone. In spite of this, it has achieved its aim: to reduce congestion in Central London. Although even TfL admits that there might be additional external reasons for this reduction, such as the slowdown in economic activity and decline in the number of tourists visiting the capital, it has to be recognised that travel delays in Central London would have not been reduced to the extent which they have been, had it not been for the Scheme.

Urban Road Pricing in the U.K.

279

The long-term impacts of the scheme are still unknown and the monitoring programme will take care of their assessment. The reasons for the success of the London Scheme are being and will continue to be scrutinised for many years to come. Congestion charging in London fits into an integrated transport strategy and Londoners, happy or unhappy with the Scheme, can clearly see at least part of the immediate results (reduced traffic levels and more reliable bus services). The consultation process and the public information campaign ensured that different views were taken into account and that the public knew what, where, how, and when. Edinburgh is imitating London’s example and may have a good chance of success. It took almost 40 years to put the prescribed solution suggested by the Smeed Report (Ministry of Transport, 1964) into practice. The schemes are not fine-tuned but they are aimed at reducing congestion and they have succeeded in doing so.

NOTES 1. Since the Transport Act 2000 came into effect on 1 February 2001, for the local authority to keep the revenues the congestion charging scheme would have to start before February 2011. 2. The power to make joint local-London charging schemes later conferred by the Transport Act 2000 does not limit any of the powers to introduce road user charging in Greater London given by the Greater London Authority Act 1999. Furthermore, the Transport Act 2000 explicitly states that the London charging schemes do not need the consent of the Secretary of State. 3. Subject to confirmation by the Secretary of State for Transport in England, the National Assembly for Wales in Wales, and the Scottish Ministers in Scotland. In the case of London the approval by the Secretary of State is only needed if the scheme is a joint local-London charging scheme. 4. Blue Badges, which existed before the scheme was implemented, are special parking permits issued to disabled people to allow them to park near shops, stations, and other facilities. The badge belongs to the disabled person who qualifies for it (who may or may not be a car driver) and can be used in any vehicle they are travelling in. The badge belongs to an individual, not a vehicle. As of October 2003 there are around 113,800 Blue Badge holders registered for a discount. There are over 200,000 Blue Badge holders in London altogether, and nearly 2 million across England. The discount applies to individual Blue Badge holders anywhere in the EC. 5. As of October 2003 TfL is working with the National Health Service and the London Fire and Emergency Planning Authority to develop and implement these reimbursement schemes. 6. An appeal may be made if the representation is rejected by TfL. 7. Transport policies in Singapore are explained in detail in Chapter 9 of this volume. 8. These figures were provided by TfL on request and are part of the London Area Transport Survey 2001, Household Survey, Interim weighted data. They will be included in the London Travel Report 2003 when it is published.

280

GEORGINA SANTOS

9. Although PCU ratings vary from country to country, these are the ratings mostly used in the U.K. 10. The numbers provided by TfL will be included in the London Travel Report 2003 when it is published. 11. The Edinburgh City Car Club provides an alternative to traditional car ownership. Members of the club pay for the time they use the cars. The club has a fleet of 17 cars and over 200 active members. Although there is a City Car Club in London as well, as of October 2003, it is still very small. There are similar schemes in place in Singapore, known as car cooperatives, and in the U.S., known as car sharing programmes. These are described in Chapters 9 and 12 of this volume respectively.. 12. SESTRAN local authorities include: Borders, East Lothian, Midlothian, West Lothian, Fife, Clackmannanshire, Stirling, Falkirk, and the Forth Estuary Transport Authority.

ACKNOWLEDGMENTS Support from the British Academy is gratefully acknowledged. The author is indebted to Jeremy Evans, Simon Burton, Charles Buckingham, Sharon Cartwright, Ruth Excell and Karen Grayson, from Transport for London, to Georgina Osbourn, from Trafficmaster PLC, to John Saunders from Transport Initiatives Edinburgh, and to John McGargill and Danny Harland from Durham County Council, for provision of data. The author is also indebted to Blake Shaffer, with whom she worked on the section on the London Congestion Charging Scheme, and to David Reams, Peter Jones, Peter Mackie and Mike Goodwin for comments on an earlier draft. Any remaining errors are the author’s sole responsibility.

REFERENCES Acts of Parliament (1999). Greater London Authority Act 1999 c. 29. London: HMSO. www.hmso. gov.uk/acts/acts1999/19990029.htm. Acts of Parliament (2000). Transport Act 2000 c. 38. London: HMSO. www.hmso.gov.uk/acts/ acts2000/20000038.htm. Acts of the Scottish Parliament (2001). Transport (Scotland) Act 2001 asp 2. London: HMSO. www.hmso.gov.uk/legislation/scotland/acts2001/20010002.htm. Automobile Association (2003). Your guide to motoring costs. Automobile Association, 2003. www.theaa.com/allaboutcars/advice/advice rcosts home.html. City of Edinburgh Council (2002). Integrated transport initiative: Report from Transport Initiatives Edinburgh (TIE). Report No. CEC/112/02–03/CS. Department of the Environment, Transport and the Regions (1998a). A new deal for transport: Better for everyone. The Government’s White Paper on the Future of Transport. London: The Stationery Office. www.dft.gov.uk/stellent/groups/dft transstrat/documents/page/dft transstrat 021588.hcsp.

Urban Road Pricing in the U.K.

281

Department of the Environment, Transport and the Regions (1998b). Breaking the Logjam, December. www.detr.gov.uk/itwp/logjam/foreword.htm. Department of the Environment, Transport and the Regions (2001). Transport Economics Note. March. www.dft.gov.uk/stellent/groups/dft transstrat/documents/page/dft transstrat 504865.hcsp. Dodgson, J., Young, J., & van der Veer, J. (2002). Paying for road use. Technical Report. A report to the Commission for Integrated Transport, National Economic Research Associates (NERA), London, February. www.cfit.gov.uk/research/pfru/pdf/pfru-tech.pdf. Draft Edinburgh Congestion Charging Order (2003). www.tiedinburgh.co.uk/iti/downloads/ ChargingOrder-Consultation. DOC. Durham County Council Press and Publicity Unit (2002). Country’s first town centre road toll goes live next week. www.durham.gov.uk/durhamcc/pressrel.nsf/Web+Releases/ 822B70C34F24AF5280256C400055290B?OpenDocument. Durham County Council (2003). Saddler Street road user charge scheme monitoring report. Durham: Durham County Council. Goodwin, P. (1989). The rule of three: A possible solution to the political problem of competing objectives for road pricing. Traffic Engineering and Control, 30, 495–497. Goodwin, P. (1990). How to make road pricing popular. Economic Affairs, 10, 6–7. Goodwin, P. B. (1992). A review of new demand elasticities with reference to short and long run effects to price changes. Journal of Transport Economics and Policy, 26, 155–169. Goodwin, P. (2003, October 29–31). The economic cost of congestion when road capacity is constrained: Lessons from congestion charging in London. CD-ROM 16th International Symposium on Theory and Practice in Transport Economics of the European Conference of Ministers of Transport, Budapest. Goodwin, P., Dargay, J., & Hanly, M. (2004). Elasticities of road traffic and fuel consumption with respect to price and income: A review. Transport Reviews, 24, 275–292. Graham, D., & Glaister, S. (2002). The demand for automobile fuel: A survey of elasticities. Journal of Transport Economics and Policy, 36, 1–25. Greater London Authority (2001, July). The Mayor’s transport strategy. London: Greater London Authority. Harrington, W., Krupnick, A., & Alberini, A. (2001). Overcoming public aversion to congestion pricing. Transportation Research A, 35, 87–105. Ison, S. (1996). Pricing road space: Back to the future? The Cambridge experience. Transport Reviews, 16, 109–126. Ison, S. (2000). Local authority and academic attitudes to urban road pricing: A U.K. perspective. Transport Policy, 7, 269–277. Mackie, P., Wardman, M., Fowkes, A., Whelan, G., Nellthorp, J., & Bates, J. (2003). Values of travel time savings in the U.K. – Summary report: Report to the Department of Transport. Institute for Transport Studies, University of Leeds. www.its.leeds.ac.uk/working/downloads/ [VOTSummary.pdf]. Ministry of Transport (1964). Road pricing: The economic and technical possibilities. London: HMSO. MVA (1995). The London congestion charging research programme. Final report, Vol. 1: Text, Government Office for London. London: HMSO. National Statistics (2003, October). New earnings survey. www.statistics.gov.uk/pdfdir/nes1003.pdf. Odeck, J., & Br˚athen, S. (1997). On public attitudes toward implementation of toll roads – The case of Oslo toll ring. Transport Policy, 4, 73–83. Odeck, J., & Br˚athen, S. (2002). Toll financing in Norway: The success, the failures and perspectives for the future. Transport Policy, 9, 253–260.

282

GEORGINA SANTOS

ROCOL Working Group (2000). Road charging options for London: A technical assessment. London: Stationary Office. www.go-london.gov.uk/transport/publications/rocol.asp. Sansom, T., Nash, C., Mackie, P., Shires, J., & Watkiss, P. (2001). Surface transport costs and charges: Great Britain 1998. Final report to the Department of Environment, Transport and the Regions, Institute for Transport Studies, University of Leeds. www.its. leeds.ac.uk/projects/STCC/downloads/SurfaceTransportCostsReport.pdf. Small, K. A. (1992). Using the revenues from congestion pricing. Transportation, 19, 359–381. Transport for London (2001). Transport statistics for London 2001. London: Transport for London. www.tfl.gov.uk/tfl/pdfdocs/stats2001.pdf. Transport for London (2002). London travel report 2002. London: Transport for London. www.tfl.gov.uk/tfl/reports library stats.shtml. Transport for London (2003a). Congestion charging: Legal framework. www.londontransport.co. uk/tfl/cc consolidated scheme order.shtml. Transport for London (2003b). Congestion charging: Public consultation. www.londontransport.co.uk/ tfl/cc public consultation.shtml. Transport for London (2003c). Consolidated scheme order. www.londontransport.co.uk/tfl/ pdfdocs/congestion charging/cc consolidatedschemeorder.pdf. Transport for London (2003d). Congestion charging: Exemptions and discounts. www.cclondon. com/exemptions.shtml. Transport for London (2003e). Congestion charging: Business and fleet information. www.cclondon. com/businessandfleet.shtml. Transport for London (2003f, October). Congestion charging: Six months on. Transport for London, London. www.tfl.gov.uk/tfl/downloads/pdf/congestion-charging/cc–6monthson.pdf. Transport for London (2003g). Congestion charging: Payments and penalties. www.cclondon. com/paymentsandpenalties.shtml. Transport for London (2003h, June). Central London congestion charging scheme: Three months on. Transport for London, London. www.tfl.gov.uk/tfl/pdfdocs/congestion charging/cc-threemonth-report.pdf. Transport for London (2003i, June). Impacts monitoring programme: First annual report. Transport for London, London. www.tfl.gov.uk/tfl/cc monitoring.shtml. Transport for London (2003j). Congestion charging: Public transport improvements before. www. tfl.gov.uk/tfl/cc improve pt before.shtml. Transport for London (2003k). Congestion charging: Public transport improvements after. www. tfl.gov.uk/tfl/cc improve pt after.shtml. Transport Initiatives Edinburgh (2002). Integrated transport initiative for Edinburgh and South-East Scotland. Edinburgh: Transport initiatives Edinburgh. www.tiedinburgh.co.uk/iti/. Transport Initiatives Edinburgh (2003a). Investing in travel improvements: Frequently asked questions. www.tiedinburgh.co.uk/iti/. Transport Initiatives Edinburgh (2003b). Investing in travel improvements. www.tiedinburgh.co.uk/iti/. Transport Initiatives Edinburgh (2003c). Investing in travel improvements: Traffic and economic impact. www.tiedinburgh.co.uk/iti/. Walters, A. (1961). The theory and measurement of private and social cost of highway congestion. Econometrica, 29, 676–699. Westminster City Council (2003). Congestion charging – Business survey: Results of congestion charging survey. www.westminster.gov.uk/roadsandstreets/congestioncharging/business form.cfm.

12.

RECENT U.S. EXPERIENCE: PILOT PROJECTS

Patrick DeCorla-Souza 1. INTRODUCTION In the United States, market-based approaches to reduce congestion are now widely referred to as “value pricing,” a term often used synonymously with the more traditional “congestion pricing.” Both of these terms are broader than the frequently-used “road pricing,” in that they are intended to include a variety of pricing strategies applied to high-occupancy vehicle (HOV, or carpool) lanes, vehicle use, and parking as well as pricing of highway or road use itself. The term “value pricing” was proposed by the U.S. Department of Transportation during the development of federal legislation to promote use of this broader range of pricing strategies, in order to emphasise the positive benefits (or value) of using pricing to reduce congestion. The U.S. established the Value Pricing Pilot Program in 1998. This Federal grant program, authorised under the Transportation Equity Act for the 21st Century, provides states, local governments, or other public entities with 80% Federal matching funds to establish, maintain, and monitor a wide variety of pricing projects. Since Program authorisations first became available in the fiscal year 1999, about $29 million have been obligated under the Program to support more than 36 projects in 15 states. This amount is in addition to about $30 million expended under the predecessor Congestion Pricing Pilot Program authorised in 1991 under the Intermodal Surface Transportation Equity Act. While many of the projects are at early stages of development, several have already been implemented and have

Road Pricing: Theory and Evidence Research in Transportation Economics, Volume 9, 283–308 Published by Elsevier Ltd. ISSN: 0739-8859/doi:10.1016/S0739-8859(04)09012-2

283

284

PATRICK DECORLA-SOUZA

proven to be successful. This chapter focuses on recent U.S. experience with these projects. Value pricing encompasses a variety of strategies to manage congestion on highways and surface streets, including both tolling of highway facilities and other strategies not involving tolls. There are four broad types of pricing strategies that have been implemented or are under consideration in the U.S.: (A) Newly-imposed tolls on existing toll-free facilities (usually electronicallycollected), including:  Tolls for vehicles not meeting normal occupancy requirements for use of High Occupancy Vehicle (HOV) lanes;  “Cordon tolls” around a designated area, and other forms of area pricing; and  Tolls on one or more general purpose lanes of a multi-lane facility, with toll credits provided to users of adjacent lanes, a concept known as “FAIR” (Fast and Intertwined Regular) lanes. (B) Tolls on lanes added to existing highways (usually electronically-collected), including:  Tolls on newly-constructed general purpose lanes;  Tolls on new HOV lanes for vehicles not meeting occupancy requirements; and  Tolls on “Queue Bypass” lanes that are added to arterial streets at intersections, or to freeway entrance ramps. (C) Variable tolls (usually electronically-collected) on existing or newly-built toll roads, bridges, and tunnels. The difference between this strategy and the preceding two strategies (A and B) is that strategies A and B impose new (generally variable) tolls on existing toll-free facilities, while with strategy C, flat tolls on existing or new toll facilities are changed to variable tolls. Of course, when new toll facilities are built, generally traffic will need to build up for several years before variable tolls are needed to manage congestion. (D) Pricing strategies that do not involve tolls, including:  Usage-based vehicle charges, including mileage-based charges for insurance, taxes, or leasing fees; and car sharing; and  Market pricing of employer provided parking spaces (called “Cash-out” when accompanied by payments to former recipients of free parking);  Payments to households to reduce their use of cars. In addition, the Value Pricing Pilot Program supports region-wide studies within metropolitan areas attempting to identify candidates for implementation of pilot pricing projects. Table 1 through 3 provide listings of projects funded under the Program by type. It should be noted that the lists in Tables 2 and 3 include current

Recent U.S. Experience: Pilot Projects

285

Table 1. Value Pricing Projects by Type-Operational Projects. State

Locality/Year Implemented

A. Pricing on existing roads California San Diego/1996 (low tech) 1998 (electronic tolls) Texas Houston/1998

Texas

Houston/2000

B. Pricing on new lanes California Orange County/1995 C. Pricing on toll roads California Orange County/2002

Florida

Lee County/1998

New York

New York metropolitan area/2001

New Jersey

Statewide/2000

D. Pricing of parking and vehicle usea California San Francisco/2001

Washington

Seattle/2002

Washington

Seattle/2000

Project

HOT lanes on I-15: Toll varies dynamically from 50 cents to $4 depending on traffic demand HOT lanes on Katy Freeway (I-10): $2 toll charged to two-person carpools in the peak hour of the peak period; 3-person and larger carpools are free HOT lanes on U.S. 290: Toll policy same as for I-10, but applies only to morning peak period Express lanes on SR91: Toll varies from $1 to $5.50 depending on traffic demand Peak pricing on the San Joaquin Hills toll road: Toll surcharge of 25 cents during peak period at several entrances to the facility; toll surcharge of 50 cents at one mainline toll plaza Variable pricing of two bridges: 50% toll discount (amounting to 25 cents) offered in shoulders of the peak periods Variable tolls on Hudson River crossings: Off-peak tolls discounted by 20% relative to peak period tolls, i.e. $4 vs. $5 Variable tolls on New Jersey Turnpike: Peak period toll exceeds off-peak toll by 12.4%; for the entire 238 km (148 mile) length, off-peak toll is $4.85 vs. peak toll of $5.45 Car sharing: Charges are $4 per hour (10 AM – 10 PM) and $2 per hour (other times); plus 44 cents per mile Parking cash-out: Monthly average parking cost in downtown Seattle is about $175. This is the amount those cashing out might expect to get Cash out of cars: Weekly average cost for owning a car was estimated at $63.90. This is the amount those “cashing out” their cars might expect to save

Note: Acronyms explained in the text and listed in Appendix. a Car sharing and parking cash out have also been implemented in other locations in the U.S. The projects shown are only those that have received federal Value Pricing Pilot Program funding.

286

PATRICK DECORLA-SOUZA

Table 2. Value Pricing Projects by Type-Projects Under Development. State

Locality

A. Pricing on existing roads A-1. Conversion of HOV to HOT lanes California Alameda County Colorado Denver Florida Miami-Dade County Minnesota Minneapolis-St. Paul

Project

HOT lanes on I-880 HOT lanes on I-25 HOT lanes on I-95 HOT lanes on I-394

A-2. Cordon tolls Florida New York

Lee County New York City

Cordon pricing in Ft. Myers Beach Tolls on East River and Harlem bridgesa

A-3. FAIR lanes California Georgia Oregon Texas

Alameda County Atlanta Portland Houston

FAIR lanes on I-580/I-680 FAIR lanes simulation on GA 400 FAIR lanes on entrance ramps to Hwy. 217 Managed lanes on the Katy Freeway

B. Pricing on new lanes California California Colorado Florida North Carolina Oregon Texas Texas Texas

Alameda County San Diego Denver Lee County Raliegh/Piedmont Portland Dallas Houston San Antonio

HOT lanes on I-680 Extension of I-15 HOT lanes HOT lanes on C-470 Priced queue jumps HOT lanes on I-40 HOT lanes on Hwy. 217 Managed Lanes on LBJ Freeway Managed lanes on the Katy Freeway HOT lanes on I-35

C. Pricing on toll roads Florida Florida Florida Illinois Ohio Pennsylvania

Broward County Lee County Miami-Dade County Chicago area Statewide Philadelphia

Variable tolls with open road tolling Variable pricing of heavy vehicles Pricing options on Florida Turnpike Variable tolls on North-West Tollway Discount truck tolls on Ohio Turnpike Variable tolls on Pennsylvania Turnpike

D. Pricing of parking and vehicle use Georgia Atlanta Minnesota Statewide Oregon Statewide Washington Seattle

Mileage-based insurance Variabilisation of fixed auto costs Financing infrastructure with value pricing GPS-based pricing

Note: Acronyms explained in the text and listed in Appendix. a Project is not currently a Federal pilot project.

Recent U.S. Experience: Pilot Projects

287

Table 3. Value Pricing Projects – Regionwide Studies. State

Locality

Project

Arizona Connecticut Maryland Minnesota Texas Virginia Washington

Phoenix New Haven Statewide Twin Cities Dallas/Ft. Worth Northern Virginia Seattle area

Value lanes studya Regional value pricing studya Feasibility of value pricing at 10 locations Regional study and outreach Region-wide value pricing study Regional HOT lanes study Regional HOT lanes study

Note: Acronyms explained in the text and listed in Appendix. a Project is not currently a Federal pilot project.

Program participants, as well as those that have made significant progress in their efforts and have expressed interest in participating in the Program. The projects listed in Table 1 that are now operational are discussed further in the remaining sections of this chapter.

2. CONVERTING HOV LANES TO HOT LANES “HOT” is the acronym for “High Occupancy/Toll.” On HOT lanes, low occupancy vehicles are charged a toll, while High-Occupancy Vehicles (HOVs) are allowed to use the lanes for free or at a discounted toll rate. HOT lanes create an additional category of eligibility for travellers wanting to use HOV lanes, since drivers can be eligible to use the facility either by meeting its minimum passenger requirement, or by choosing to pay a toll to gain access to the HOV lane. HOT lanes currently operate in Houston, Texas and San Diego, California. There is increasing interest in HOT lanes in the U.S., due to their many potential advantages. HOT lanes can:  Reduce congestion during the peak period by taking some traffic off the regular lanes;  Offer drivers the option to bypass congestion when in a hurry, so that drivers can avoid delays when it is important to do so (for example if they have to catch a flight at the airport);  Provide revenue to pay for congestion-reducing road improvement projects, expansion of roads, public transport improvements, or park-and-ride programs;  Create financial incentives to make public transport and carpooling more attractive, while continuing to ensure congestion-free travel by these vehicles;

288

PATRICK DECORLA-SOUZA

 Reduce air pollution resulting from cars idling in traffic jams by reducing congestion; and  Reduce fuel consumption resulting from stop-and-go traffic. These benefits are not exclusive to HOT lanes, since other forms of congestion pricing can also yield many of these same benefits. The difference between HOT lanes and other pricing systems, however, is that with HOT lanes drivers can choose between meeting the vehicle occupancy requirement or paying the toll in order to use the HOV lane.

2.1. Operational Projects 2.1.1. San Diego’s Priced Express Lanes San Diego’s “FasTrak” pricing program was implemented in April 1999. Under this program, customers in single-occupant vehicles pay a toll each time they use the Interstate-15 (I-15) HOV lanes, shown with a thick line on Fig. 1. The unique feature of this pilot project is that tolls vary dynamically with the level of congestion on the HOV lanes. Fees can vary in 25-cent increments as often as every six minutes to help maintain free-flow traffic conditions on the HOV lanes. Motorists are informed of the toll rate changes through variable message signs located before the entrance to the Express Lanes, so that they can elect to enter the Express Lanes or remain on the free lanes. The normal toll varies between $0.50 and $4, but during very congested periods it can be as high as $8. All transactions are electronic. Overhead antennas read a transponder affixed to the inside of a vehicle’s windshield and deduct the toll electronically from the driver’s pre-paid account. As of August 19, 2003, 24,369 transponders had been issued. During August 2003, average daily traffic on the Express Lanes reached a peak of 22,063 total vehicles, including both vehicles meeting the occupancy requirement for free use and those paying the toll. This represents an increase of 140% from the 9,200 daily vehicles prior to the initiation of the program. On average, 79% of the daily traffic is from high occupancy vehicles (HOVs), and 21% is from toll-paying customers. Total revenue in 2002 was around $2.2 million. Approximately 50% of the revenues are used to fund the Inland Breeze express bus service that operates in the corridor. The remainder is used to fund enforcement by the California Highway Patrol, and operation of the Customer Service Centre. Extensive outreach was conducted to measure public response to the concept. The outreach included 25 stakeholder interviews, three focus groups, 100 intercept surveys at park-and-ride lots and transit centres, and a telephone survey of 800 I-15

Recent U.S. Experience: Pilot Projects

289

Fig. 1. San Diego’s I-15 “FasTrak” Lanes.

corridor users. The surveys (Wilbur Smith Associates et al., 2002) found that corridor users did not consider equity to be a major issue or obstacle to implementing pricing on the managed lanes. The majority of those interviewed in the telephone survey (71%) feel that pricing the lanes is “fair” for travellers on the main lanes. Furthermore, 66% approve of the currently operating HOT lanes, and 71% believe that tolls are an effective way to manage demand. Both users and non-users of

290

PATRICK DECORLA-SOUZA

Fig. 2. Houston’s “QuickRide” Program on I-10 and U.S. 290.

the dynamically priced I-15 HOT lanes support the use of pricing. Support is high across all income groups, with the lowest income group expressing as much support as the highest income group (78%). 2.1.2. QuickRide Program in Houston, Texas The “QuickRide” pricing program was initially implemented on an existing reversible HOV lane on Interstate-10 (I-10, also known as the Katy Freeway) in Houston in January 1998. A similar project was subsequently implemented on Houston’s U.S. 290 highway in November 2000. Figure 2 shows the segments where the HOV lanes operate with thick lines. The HOV lanes are reversible and restricted to vehicles with three or more people during the core hours of the peak periods. The pricing program allows a limited number of two-person carpools to pay a toll to access the HOV lanes during these hours. Single-occupant vehicles are not allowed to use the HOV lanes. Participating two-person carpool vehicles pay a $2 per trip toll, while vehicles with higher occupancies continue to travel for free. As in San Diego, the QuickRide project is completely automated and no cash transactions are handled on the facility. Results from surveys conducted on I-10 indicate that the primary source of QuickRide participants is persons who formerly travelled in single-occupant vehicles on the regular lanes (Berg et al., 1999). Toll revenues from the several

Recent U.S. Experience: Pilot Projects

291

hundred two-occupant vehicles electing to pay the $2 toll each day pay for all program operation costs.

2.2. Projects Under Development Under the Value Pricing Pilot Program, pricing of existing HOV lanes is being studied for implementation in Denver, Colorado on Interstate-25 (I-25)/US 36. In addition, the potential conversion of existing HOV lanes to HOT lanes is under study for route I-95 in Miami-Dade County, Florida.

3. CORDON TOLLS Introducing new tolls on existing toll-free facilities without continuing to allow some opportunity for free travel on the same facility (as HOT lanes allow) has generally been considered to be unacceptable to the public in the U.S. However, two such projects are under consideration in Florida and in New York.

3.1. Projects Under Development 3.1.1. Cordon Pricing in Lee County, Florida The Town of Fort Myers Beach in Lee County, Florida, is an island community that experiences a heavy influx of visitors during the tourist seasons, thus aggravating the problem of traffic congestion. Due to the relatively small land area occupied by the Town and potential environmental costs, construction of new roads or widening of the existing ones would be difficult. The Town is studying the feasibility of introducing a new variable cordon toll at each of the two approaches to the Town. 3.1.2. Tolling Existing Free Bridges Over the East River and Harlem River in New York The Mayor of New York City has proposed to place tolls on 12 city-owned bridges over the East and Harlem Rivers, which connect Manhattan with the Burroughs of Queens, Brooklyn, and the Bronx. Use of these bridges is currently free. The proposed tolls, which would not vary by time of day or level of congestion, are expected to bring in $600 million in revenues annually. A report by the Bridge Tolls Advocacy Project (Komanoff, 2003) indicates that 98% of New York residents of driving age do not drive to work using the East River bridges and on average would pay less than $50 a year with the proposed tolls in effect. Regular users of

292

PATRICK DECORLA-SOUZA

the East River bridges, who have higher average salaries than those who do not use the bridges to drive to work, would pay tolls that would average $1,500 per car per year.

4. FAIR LANES “FAIR” lanes stands for “Fast and Intertwined Regular” lanes. This type of pricing seems likely to be more acceptable to the motoring public (DeCorla-Souza, 2000) because it would provide drivers the option of paying to use faster lanes or being compensated for continuing to use unpriced lanes on the same facility. The system would involve separating multiple freeway lanes, typically using plastic pylons and striping, into two sections: “fast” lanes and “regular” lanes. The fast lanes would be electronically tolled express lanes, where tolls could change dynamically to manage demand. In the remaining unpriced lanes, drivers whose vehicles were equipped with transponders would be compensated with credits that would be based on the tolls in effect at the time they travelled. These credits could be used as toll payments on days when drivers accumulating them chose to use the fast lanes, or as payment for transit fares, paratransit fees (vanpool membership fees, for example), or parking at commuter park-and-ride lots in the corridor. The credits would be funded from toll revenues generated by charges imposed for use of the fast lanes. Buses, paratransit vehicles and carpools could use the fast lanes without paying any toll. FAIR lanes have benefits similar to those identified above for HOT lanes. They increase freeway throughput and speed transit service, generate revenues to finance road and public transport improvements, and allow motorists to bypass congestion as they choose. However, in addition to these benefits, FAIR lanes make it easier to provide more than one express lane, thereby allowing faster vehicles to overtake slower-moving vehicles. This prevents queuing of vehicles behind the slower vehicle, and prevents gaps from developing in front of the slower vehicle, which results in lower vehicle throughput. Providing credits to the accounts of drivers using the regular lanes is intended to increase the public acceptability of taking an existing free lane from a facility for use as an express lane. Making more capacity available for paying motorists by transferring multiple lanes on the same facility from free to fast lane status might help to maintain tolls at affordable levels for those with lower ability to pay, thus allowing more motorists to make use of this premium service. The strategy can be established on any existing congested freeway facility, preferably a facility with four or more lanes in each direction, or a facility with three lanes in each direction that is proposed to be widened. When adding new

Recent U.S. Experience: Pilot Projects

293

freeway lanes, an existing adjacent free lane could also be combined with the added lane to create a wider fast section. On congested toll roads or bridges, higher tolls could be charged on fast lanes, while other motorists could be given discounts rather than offered completely free use of the remaining lanes.

4.1. Projects Under Development A feasibility study involving FAIR lanes is underway in Alameda County, California in the San Francisco Bay area. A FAIR lanes simulation study has also been proposed in Atlanta, Georgia. FAIR lanes are also being studied at freeway entrance ramps on Highway 217 in Portland, Oregon and as an alternative to the proposed extension of HOT lanes on the Katy Freeway in Houston, Texas.

5. PRICED NEW LANES State and local budget cuts and unsuccessful attempts to fund transportation improvements through taxation have increased the interest of states in financing lane additions to existing highways using toll revenues. Newly-constructed express lanes with tolls have been implemented to date in only one location in California, but similar strategies are under development in many states. Tolls on added lanes could be allowed to vary by time of day and be collected without slowing highway speeds using electronic toll collection technology. Tolls could also be set “dynamically,” i.e. they could be increased or decreased every few minutes in response to fluctuating demand so as to ensure that the lanes are fully utilised, yet remain uncongested.

5.1. Operational Project 5.1.1. Express Lanes on State Route 91 in Orange County, California The State Route 91 (SR 91) express lanes in Orange County, California opened in December 1995 as a four-lane toll facility in the median of a 16 km (10 mile) section of one of the most heavily congested highways in the U.S. The segment where the express lanes operate is shown on Fig. 3. The toll lanes are separated from the general purpose lanes by a painted buffer and plastic pylons. Toll revenues have been adequate to pay for construction and operating costs. In fact, the private company that had the franchise to build and operate the facility recently sold the franchise to the Orange County Transportation Authority for a profit.

294

PATRICK DECORLA-SOUZA

Fig. 3. Orange County’s SR 91 Express Lanes.

As of August 2003, tolls on the express lanes vary between $1 and $5.50. Tolls differ by direction, and are set by day of the week and time of the day to reflect the level of congestion delay in the adjacent free lanes that can be avoided by using the toll lane, and to maintain free-flowing traffic conditions on the toll lanes. Drivers can observe message signs before entering the SR 91 Express Lanes to obtain the current toll schedule, which is subject to change without notice in order to optimise traffic flows (Orange County Transportation Authority, 2003). As Poole and Orski (2003) note, the SR 91 Express Lanes represent only 33% of the SR 91 freeway capacity (i.e. two out of six lanes in each direction), but are carrying 40% of the traffic in the busiest peak hours, at speeds of 104 km per hour (65 mi per hour) vs. 16–32 km per hour (10–20 mi per hour) in the adjacent free lanes. As is well known among traffic engineers, congestion results in reduced throughput on the regular lanes, accounting for the higher relative throughput on the free flowing Express Lanes in peak hours. SR 91 Express Lanes customers pay tolls by having them electronically deducted from pre-paid accounts. All vehicles travelling on the express lanes must be equipped with a “FasTrak” transponder mounted on the inside of the

Recent U.S. Experience: Pilot Projects

295

windshield. Vehicles with three or more occupants are not charged except when travelling eastbound from 4 PM to 6 PM on weekdays, the peak period in the heavy traffic direction, and during that time receive a 50% discount from the posted toll. There were almost 139,000 transponders in circulation at the end of 2002. The facility served 9.5 million vehicles in 2002, averaging over 26,000 vehicles per day, and yielding revenues of over $29 million.

5.2. Projects Under Development Pricing of new lanes is being studied for implementation on I-680 in Alameda County, California in the San Francisco Bay Area; on I-15 in San Diego, California; on C-470 in Denver, Colorado; in Lee County, Florida on Queue-bypass lanes at two intersections; on Interstate-40 (I-40) in the Raleigh-Durham and Piedmont areas in North Carolina; on Highway 217 in Portland, Oregon; and on the Katy Freeway (I-10) in Houston, and the LBJ Freeway (I-635) in Dallas, Texas. In addition, a study in Sonoma County, California in the San Francisco Bay Area has recommended new HOT lanes on Highway 101, and a study has recently begun to plan for new HOT lanes on I-35 in San Antonio, Texas.

6. VARIABLE PRICING ON FIXED-TOLL FACILITIES Facilities that are already tolled – but on which tolls do not vary by time of day or traffic conditions – can introduce variable rates in order to reduce traffic during peak times. Thus, existing tolls on congested facilities may be varied by day of the week or time of the day with the intention of encouraging some travellers to use the roadway during less congested periods, to shift to another mode of transport, or to change route. If congestion at peak times is reduced, the remaining peak period travellers will experience decreased delays. Ultimately, such shifts will result in less need for roadway expansion on the toll facilities. All of the projects in this category that have been implemented to date use electronic technology to vary tolls by time period, and to be eligible vehicles must be equipped with transponders.

6.1. Operational Projects Four projects have been implemented in four states in the U.S.: Florida, New York, New Jersey, and California.

296

PATRICK DECORLA-SOUZA

6.1.1. Bridge Pricing in Lee County, Florida In August 1998, Lee County implemented a value pricing strategy on two toll bridges between the cities of Ft. Myers and Cape Coral, shown on Fig. 4. The project created a peak/off-peak pricing structure offering bridge users a discount from the prevailing toll during times immediately before and after the peak periods. Under the time varying toll schedule, a 50% toll discount (from a base toll of $0.50 normally charged to vehicles with electronic transponders) is provided for trips made during the half-hour period before the morning peak of 7–9 AM, as well as during the two-hour period following it. In the evening, the discount periods are the two-hour period before the evening peak of 4–6:30 PM, and the half-hour period following it. The program has been successful in inducing significant shifts in traffic out of the peak times. Surveys indicate that over 71% of eligible motorists (i.e. those with vehicle transponders) shifted their time of travel at least once a week to obtain a toll discount amounting to just 25 cents (Burris et al., 2002). 6.1.2. Variable Tolls on Hudson River Crossings in New York The Port Authority of New York and New Jersey adopted a variable toll strategy for users of the electronic toll collection system (E-ZPass) in March 2001. The Port Authority provides a 20% discount from normal tolls for off-peak use of its bridges and tunnels crossing the Hudson River between New York and New Jersey, shown on Fig. 5. Peak periods are weekdays 6–9 AM and 4–7 PM, and Saturdays and Sundays 12 noon to 8 PM. An estimated 121.4 million vehicles and approximately 65 million interstate transit system riders use the interstate crossings annually. Studies show that morning peak period traffic in May 2001 was reduced by 7% compared to the same month in 2000. Evening peak period traffic dropped by 4%; and overall traffic remained stable (Port Authority of New York and New Jersey, 2001). Muriello and Jiji (2004) provide evidence that a significant share of morning traffic has shifted to the 5–6 AM hour when off-peak rates are in effect. Weekday 5–6 AM traffic increased from 10.6% of total 5–10 AM traffic in 2000 (before the value pricing program began) to 12.9% of total 5–10 AM traffic in 2002. 6.1.3. Variable Tolls on the New Jersey Turnpike The New Jersey Turnpike Authority operates a 238 km (148 mile) facility with 28 interchanges, shown on Fig. 6. It is one of the most heavily travelled roadways in the country with average daily trips exceeding 500,000 vehicles. The Turnpike’s variable pricing program began in the fall of 2000. The program provides for car tolls that are currently 12.4% higher during peak traffic hours (7–9 AM and 4.30–6.30 PM, Monday through Friday) than during off-peak periods for users of the electronic toll collection system. When the value pricing program

Recent U.S. Experience: Pilot Projects 297

Fig. 4. Lee County’s Variably Priced Bridges.

298 PATRICK DECORLA-SOUZA

Fig. 5. New York’s Hudson River Crossings.

Recent U.S. Experience: Pilot Projects

299

Fig. 6. New Jersey Turnpike.

300

PATRICK DECORLA-SOUZA

initially started, the price differential was 7.6%. The differential between peak and off-peak tolls is scheduled to increase in a phased manner over several years. Preliminary data show that value pricing is working to shift traffic out of the peak period. Most of the recent growth in traffic on the Turnpike has been in the

Fig. 7. San Joaquin Hills Toll Road in Orange County, California.

Recent U.S. Experience: Pilot Projects

301

off-peak hours, with total traffic up by around 7%, but morning peak traffic up by only 6% and afternoon peak traffic up by only 4%. The proportion of daily Turnpike traffic accounted for by the morning peak dropped from 14 to 13.8%, and the afternoon peak’s share of traffic decreased from 14.7 to 14.3%. 6.1.4. Variable Tolls on the San Joaquin Hills Toll Road The San Joaquin Hills Toll Road (State Route 73) in California is 24 km (15 miles) long and extends from Interstate-5 (I-5) near San Juan Capistrano to Interstate-405 (I-405) in Newport Beach, as shown on Fig. 7. It provides an alternative to heavily congested portions of I-5 and I-405, two North-South freeways in the southern portion of the Los Angeles metropolitan area. The Toll Road carries over 2.3 million vehicles annually on a six-lane facility, and is currently near capacity during peak periods. A small peak period premium of 25 cents (in addition to the normal toll, which varies by distance) was implemented at most entrances to the facility in February 2002. The premium was calibrated to reduce congestion and spread peak demand to shoulder and off-peak periods, while maintaining revenues at levels required to maintain the covenants on the Agency’s revenue bonds. Evaluation results showed that there was a net reduction of 2.7% in total traffic along with a net increase of 5.8% in toll revenue due to the premium tolls. On October 5, 2003, an additional peak period premium of 50 cents was implemented at the Catalina View Mainline toll plaza (in addition to the normal toll of $2.50).

6.2. Projects Under Development Several new projects are under consideration for implementation, including three in Florida (Florida Turnpike, Sawgrass Expressway, and pricing of heavy vehicles on bridges in Lee County) and one in Pennsylvania (Pennsylvania Turnpike). In addition, studies to implement variable tolls have begun for Chicago’s North-West Tollway; and discounted tolls for trucks are being planned on the Ohio Turnpike, to reduce truck traffic diversions to parallel free highways.

7. USAGE-BASED VEHICLE CHARGES Three types of creative uses of this pricing strategy are being explored in the U.S.: Pay-As-You-Drive Automotive Insurance; Mileage-Based Automotive Leasing and Vehicle Taxation; and Car Sharing.

302

PATRICK DECORLA-SOUZA

7.1. Pay-As-You-Drive (PAYD) Automotive Insurance By converting automotive insurance from a fixed to a per mile cost, insurance companies may more accurately bill their customers based on crash risk and provide them a financial incentive to drive less. This may in turn reduce accidents, public infrastructure costs, and congestion and environmental externalities. A study by the Economic Policy Institute (Baker & Barrett, 1999) estimates that conversion of automotive insurance costs to PAYD could reduce accidents, congestion, and emissions from cars by 10–20%. A simulation study of this strategy is underway in Atlanta, Georgia.

7.2. Mileage-Based Automotive Leasing and Vehicle Taxation Over 80% of the costs of owning and operating a vehicle are fixed (Litman, 1997). Once a person has chosen to acquire a vehicle, the incremental costs of operating it are comparatively low. Converting some fixed vehicle costs to a pay-as-you-drive (PAYD) fee schedule financially rewards consumers for reducing their driving and related congestion and vehicle emissions. Pilot simulation tests of various types of mileage based pricing strategies are underway in the Twin Cities, Minnesota, the State of Oregon, and the Puget Sound (Seattle) region of Washington State.

7.3. Car Sharing This strategy involves automated hourly neighbourhood car rentals that substitute for car ownership. Similar schemes for Singapore and the U.K. are mentioned in Chapters 9 and 11 of this volume respectively. By sharing a neighbourhood car, individuals eliminate their fixed monthly car expenses such as car loan and insurance costs, and instead incur a variable car payment based on usage. In the U.S., there are active and growing car sharing programs in Seattle, Boston, San Francisco, Portland (Oregon), Chicago, New York, and Washington, DC. Under the Value Pricing Pilot Program, an evaluation of the impacts of car sharing on driving and congestion is underway in San Francisco. After two years of operation of the program, nearly 30% of those who signed up for the program (i.e. “members”) have reduced their car ownership by at least one car, and two-thirds have opted not to purchase another car because of their participation in the program. In a matched pair comparison with non-members, Cervero and Tsai (2003) estimate that members drove 6.46 miles less per day than non-members.

Recent U.S. Experience: Pilot Projects

303

8. “CASH-OUT” STRATEGIES Cash out strategies involve paying car users some kind of compensation for not using their cars. The two options that have been considered to some extent are Parking Cash Out and Car Cash Out, which are described below.

8.1. Parking Cash Out With Parking Cash Out, employers offer their employees the option of receiving an increase in taxable cash income in lieu of free or subsidised parking provided by the employer. Parking Cash Out works best in areas where transit is accessible or where employees are willing to carpool, telecommute, cycle or walk. Parking Cash Out has been implemented at several employment sites in the states of Washington, Minnesota, and California. Studies at seven employment sites in Minnesota have shown that, on average, Parking Cash Out at those sites resulted in an 11% reduction in solo-driving (Van Hattum et al., 2000). A similar study, conducted for eight employers in California concluded that solo driving to work at those eight sites fell by 17%, carpooling increased by 64%, transit ridership increased by 50%, walking and cycling increased by 33%, and commuter parking demand fell by 11% (Shoup, 1997). Preliminary results from a Parking Cash Out demonstration project under way in downtown Seattle, Washington, show that 17 out of 167 employees offered a parking cash out option (i.e. 10% of those eligible) accepted cash in lieu of parking (Glascock et al., 2003).

8.2. Car Cash Out Car cash-out involves paying households to use one less car for a certain period of time. The idea is to provide incentives for households to consider alternative modes of transport such as transit, carpool, cycling, or walking. Three demonstration studies were carried out in Seattle, Washington. Participating households were asked to use one less car and keep daily records of their trips and transport modes used. Households were paid a weekly stipend (equal to the average national cost of owning a second vehicle) during the time of the study to simulate the savings they would realise if they actually were to sell one of their cars. Daily records, odometer readings, and anecdotal stories were analysed to document costs and to assess whether households made significant behaviour changes such as carpooling, using public transport, cycling or walking. Results indicate that participating households reduced solo driving by 27% during the

304

PATRICK DECORLA-SOUZA

periods they were required to refrain from using one car. Of the 86 participating households in the three demonstration phases, 14 (16%) sold their cars after the study ended, seven (8%) pledged not to replace their cars after their vehicles were retired, and nine (10%) plan to sell their cars (Hoffman & Wiger, 2003).

9. CONCLUSIONS Table 4 summarises key information about the various types of value pricing projects implemented or under consideration in the U.S. during the past decade. With regard to projects involving tolling, the Value Pricing Pilot Program has demonstrated that:  Pricing can be politically and publicly acceptable – so far, four priced lane 







projects and four variably priced toll facility projects are operating without any significant public or political controversy. Pricing keeps congestion from occurring on priced lanes, as demonstrated by the HOT lanes in the Houston, San Diego and Los Angeles metropolitan areas; and it reduces congestion on toll facilities, as demonstrated by shifts in traffic on variably priced toll facilities in New York, New Jersey and Florida. Pricing changes travel behaviour, as demonstrated by travel choices made by those motorists on toll facilities who choose to shift their time of travel to offpeak periods to take advantage of lower tolls (e.g. New York and Florida); and motorists who choose priced lanes (e.g. in Los Angeles, San Diego and Houston) to take advantage of faster and more reliable travel times. Pricing can improve utilisation of existing highway capacity, as demonstrated in San Diego, where traffic volumes have increased on the HOT lanes by as much as 140% (without loss of speed) to make use of spare capacity on these lanes; this took traffic off the regular lanes and thereby reduced the congestion levels that they would have otherwise experienced. Pricing can provide funding for transportation improvements – new transit service was funded from toll revenues in San Diego, and the construction and operation of the new SR 91 Express Lanes in Orange County is supported entirely from toll revenues.

While many of these impacts are what theory predicts (and indeed has predicted for decades), the contribution of the pilot projects is that they provide valuable real world, on-the-ground evidence that has been very useful to U.S. transportation professionals in their efforts to convince elected officials and the public about the potential impacts and benefits of pricing strategies. Elected officials have seen that some forms of pricing can indeed be acceptable to the public, and are more willing

Priced Lanes on Otherwise Free Facilities, Including Conversions of HOV Lanes (Type A) and New Priced Lanes (Type B)

Variable Tolls on Toll Facilities (Type C)

Mileage-Based User Charges for Insurance, Taxes and Leasing Fees and Car Sharing (Type D)

Parking Pricing with Cash-Out of Existing Free Parking (Type D)

How does it reduce congestion?

Keeps traffic free flowing on the priced lanes, maintains high vehicle throughput, accommodates some traffic previously using regular lanes

Shifts peak period travellers to other modes, routes and times

Reduces use of driving for all trips, both peak and off-peak

Induces solo-drivers to shift to other modes for their work trips

What economic incentive is offered to change travel behaviour?

Prices change in the priced lanes to influence traveller choice and keep demand within pre-determined limits

Off-peak toll discounts, or higher peak tolls

Travellers save money by reducing driving

Cash or transit fare subsidies are offered in lieu of free parking

What are the observed travel impacts?

In San Diego, the number of vehicles carried on HOV lanes has increased by 140%, increasing freeway throughput and saving time for toll-paying motorists

4–7% reduction in peak period traffic observed in New York; 71% of participants shifted time of travel to get discount at least once a week in Florida

Simulation tests of mileage-based charges are underway in Georgia and Minnesota; San Francisco, California’s car sharing members drove 6.46 miles less per day than non-members

Average reduction in solo-driving of 11% observed at work sites in Minnesota, and 17% at work sites in California

Recent U.S. Experience: Pilot Projects

Table 4. Comparison of Key Aspects of Pricing Strategies.

305

306

PATRICK DECORLA-SOUZA

to explore this option. Several metropolitan areas in the U.S. have completed or have initiated efforts to assess the feasibility of regional pricing programs. Despite several early successes, however, value pricing involving tolls still encounters opposition. This has slowed efforts in some states such as Maryland and Minnesota. One of the most important challenges to overcome has typically been the problem of equity. This was essentially the concern expressed by the former Governor of Maryland in his highly publicised rejection of a proposed HOT lane project in Maryland. Similar concerns were expressed with respect to a proposed conversion of a HOV lane to a HOT lane in Minnesota. What was learned from these two experiences and from the successful projects still in place, is that there are two key strategies that need to be employed to get public acceptance: (1) conduct an effective public information campaign early in the process; and (2) implement an integrated package that benefits all income groups. These two strategies have been employed in many of the successful value pricing projects in the U.S., and in the recent successful London Congestion Charging Scheme. However, both of them were missing in the early experiences in Maryland and Minnesota. If a proposed value pricing project is to be publicly acceptable, its benefits must be clearly identified to motorists. Motorists may benefit either directly in the form of reduced travel delay and enhanced travel options, or indirectly through appropriate use of toll revenues. To help address equity, pricing may need to be combined with some form of direct benefits to those who pay tolls or those who give up the right to use facilities that were formerly provided without charge. These can take a number of forms, including highway improvement or expansion, construction of new highways, provision of alternative modes of transport such as transit, or investment in other areas within the transport sector such as safety and environment. Other revenue allocations may include some kind of explicit compensation to low-income groups, such as toll credits similar to credits provided to low income public utility customers, tax credits to low-income commuters for tolls paid by them on value priced lanes, or toll credits provided to those who choose not to use value priced lanes, such as in the FAIR lanes concept.

ACKNOWLEDGMENTS Much of the information included in this chapter has been obtained from pilot project reports prepared by Value Pricing Pilot Program grantees, publications of the Federal Highway Administration (FHWA), and FHWA’s web site at: http://www.fhwa.dot.gov/policy/otps/valuepricing.htm. However, the author alone is responsible for any errors or omissions, and the views expressed do not necessarily reflect the policies of the FHWA.

Recent U.S. Experience: Pilot Projects

307

REFERENCES Baker, D., & Barrett, J. (1999). The feasibility of pay by the mile automobile insurance. Economic Policy Institute. Technical Paper No. 231. Berg, J. T., Kawada, K., Burris, M., Swenson, C., Smith, L., & Sullivan, E. (1999). Value pricing pilot program. TR News, 204, 3–10. Burris, M., Swenson, C., & Crawford, G. (2002). Lee county’s variable pricing project. ITE Journal, 72, 36–41. Cervero, R., & Tsai, Y. (2003). San Francisco City CarShare: Second year travel demand and car ownership impacts. http://knowledge.fhwa.dot.gov/cops/hcx.nsf/home?openform&Group= Value%20Pricing&tab=REFERENCEBYALPHA. DeCorla-Souza, P. (2000). FAIR lanes: A new approach to managing traffic congestion. ITS Quarterly, 8, 5–13. Glascock, J., Cooper, C., & Keller, M. (2003). The downtown Seattle access project parking cash out experience: Results and recommendations. King County Metro, Seattle, Washington. http:// knowledge.fhwa.dot.gov/cops/hcx.nsf/All+Documents/A19C77018189D09F85256DBA0063 D8F4/$FILE/Seattle%20Parking%20Cash%20Out%20Draft%20Report.pdf. Hoffman, J., & Wiger, R. (2003). Way to go Seattle: “One-less-car” demonstration study. Draft report. Revision September 2, 2003. Seattle Department of Transportation. http://knowledge. fhwa.dot.gov/cops/hcx.nsf/home?openform&Group=Value%20Pricing&tab=REFERENCE BYALPHA. Komanoff, C. (2003). East River Bridge tolls: Who will really pay? The Bridge Tolls Advocacy Project. http://www.bridgetolls.org/whowillpay/. Litman, T. (1997). Distance-based vehicle insurance as a TDM strategy. Transportation Quarterly, 51, 119–138. Muriello, M., & Jiji, D. (2004). The value pricing toll program at the Port Authority of New York & New Jersey: Revenue for transportation investment and incentives for traffic management. Paper No. 04-3116, CD-ROM 83rd Annual Meeting of the Transportation Research Board, Washington DC, January 11–15, 2004. http://knowledge.fhwa.dot.gov/cops/hcx.nsf/home? openform&Group=Value%20Pricing&tab=REFERENCEBYALPHA. Orange County Transportation Authority (2003). 91 expresslanes: Toll schedules. http://www.91expresslanes.com/tollschedules.asp?p=m3. Poole, R. W., & Orski, C. K. (2003). HOT networks: A new plan for congestion relief and better transit. Reason Public Policy Institute. Policy Study No. 305. http://www.rppi.org/ps305.pdf. Port Authority of New York and New Jersey (2001). The early returns: Port Authority releases preliminary congestion pricing data for bridges, tunnels and PATH. Press Release No. 92-2001. June 26. http://www.panynj.gov/pr/92-01.html. Shoup, D. C. (1997). Evaluating the effects of cashing out employer-paid parking: Eight case studies. Transport Policy, 4, 201–216. Van Hattum, D., Zimmer, C., & Carlson, P. (2000). Implementation and analysis of cashing-out employer paid parking by employers in the Minneapolis-St. Paul Metropolitan Area. U.S. EPA and the Minnesota Pollution Control Agency. http://www.mplstmo.org/parking alts.html. Wilbur Smith Associates in association with FPL and Associates, Judith Norman Transportation Consultant, Fairfax Research, Frank Wilson Associates, ESTC and ALESC (2002). I-15 managed lanes value pricing project planning study: Volume 2 public outreach. San Diego Association of Governments. January 21, http://argo.sandag.org/fastrak/pdfs/concept plan vol2.pdf.

308

PATRICK DECORLA-SOUZA

APPENDIX: LIST OF ACRONYMS HOV: SOV: HOT: FAIR: I: SR: US:

High Occupancy Vehicle Single Occupancy Vehicle High Occupancy/Toll Fast and Intertwined Regular (lanes) Interstate (always followed by a number) State Route (always followed by a number) United States (followed by a number when referring to a highway)