ALLEVIATING TRAFFIC CONGESTION: ALTERNATIVES TO ROAD PRICING † Richard Arnott September 1994 Revised version Preliminary draft: Please do not cite or quote without the permission of the author For presentation at the Taxation, Resources and Economic Development (TRED) conference on "Alternative Strategies for Managing Externalities," Lincoln Institute of Land Policy, Cambridge, Massachusetts, September 30/October 1, 1994. † Department of Economics Boston College Chestnut Hill, MA 02167 Alleviating Traffic Congestion: Alternatives to Road Pricing I am a believer but I am also a realist. In an ideal world, full efficiency would be achieved by marginal (social) cost pricing, and equity through lump-sum redistribution. In such an ideal world, travelers would make efficient choices since they would face the true social costs of their actions on every margin of travel choice -- there would be perfect congestion pricing. And since prices would equal social costs, governments would be able to determine precisely the optimal capacity of all transport facilities. But we do not live in an ideal world. Whatever its virtues, congestion pricing will not be introduced soon in the United States, at least on a significant scale. When and where it is first introduced, it will take a crude form, which falls a long way short of having auto drivers face the true social costs of their actions on every margin of choice. Even with reasonably sophisticated electronic road pricing, many margins of driver choice will still be distorted. Thus, while full marginal cost pricing -- of which perfect congestion pricing is one element -- is the ideal to be strived for, realistically we must muddle through as best we can in the untidy world of the second best. In the context of urban travel, this entails a messy mix of policies, including some forms of road pricing. In this paper I do not argue against congestion pricing -- after all, I am a believer. But I do think there is the danger that striving for the first best may distract us from the second best, that in researching and advocating congestion pricing we take time away from consideration of the other policies that may be at least as practical and practicable. In section 1, I elaborate on my views on congestion pricing. The bottom line is that wherever, whenever, and in whatever form congestion pricing is introduced, a broad array of other policies are and will be needed for efficient urban traffic management. In section 2, I run through alternative urban traffic management policies. And in section 3, I provide an overview. 1. Congestion Pricing That road travel entails a congestion externality which can be internalized by means of tolling was recognized by Dupuit (1844) and has been widely-known and well- understood by economists since Knight (1924) and Pigou (1912). Today, almost all economists favor congestion pricing in principle, though some have doubts about its practicality. Almost forty years ago, Vickrey (1959) drew up a detailed scheme for electronic auto congestion pricing in Washington, D. C. While the scheme he proposed probably seemed futuristic at the time, today the technology for electronic auto congestion pricing is available and reliable, and with widespread use should be inexpensive. The technological and economic feasibility of electronic road pricing was persuasively demonstrated in a well-designed experiment in Hong Kong in the early 1980’s (Catling and Harbord (1985)). The wisdom of congestion pricing is accepted, at least among economists, and the technology is there to implement it. Why then has not a single jurisdiction introduced electronic road pricing? 1.1 Political opposition to road pricing The answer is that politicians have regarded it, with justification, as political poison. The Hong Kong experiment was sufficiently successful that serious consideration was given to the implementation of electronic road pricing in Hong Kong. But political opposition was mounted on the grounds that electronic road pricing constitutes an invasion of privacy that can be abused by the government, and resulted in the proposal being withdrawn (Borins (1986)). Under the proposed Hong Kong system, the government would have collected information on the travel history of all its residents. The potential for abuse is obvious -- the government could, for instance, discredit an opposition politician who travels extensively or parks frequently in the red light district. This argument against electronic road pricing is hardly a damning one, since the system can easily be designed not to record travel histories. 3 But the opposition to congestion pricing is more deep-seated than the Hong Kong example suggests. Congestion pricing entails the pricing of a service -- travel on city streets -- that was previously provided free. Relatedly, people tend to view free travel on city streets as a right, or at least as a right earned from paying local taxes. As well, citizens are justifiably cynical about political motives, and tend to view proposals for congestion pricing as another tax grab, another incursion by Leviathan. Also, the promised benefits of congestion pricing are intangible. Road pricing would reduce congestion but, given the substantial day-to-day variations in traffic, the reduction might not be obvious. Finally, opponents of congestion pricing argue, probably correctly, that the policy is regressive. Professionals can alter their travel schedules to avoid paying peak tolls; clerical and factory workers cannot. As well, congestion pricing entails paying with money rather than with time. A successful campaign for congestion pricing will have to overcome all these political obstacles. The proponents of congestion pricing are developing more political savvy. They are attempting to educate the public concerning the virtues of congestion pricing. Also, they are proposing to earmark the revenues raised from congestion tolls for the expansion and improvement of transportation infrastructure, or to rebate the revenue collected, perhaps via the income tax, in such a way that all major groups in the population benefit (Jones (1991), Small (1992)). There are indications that this campaign is having at least some success. A recent poll found that the majority of commuters in London would favor congestion pricing if the revenues collected were used to upgrade the transport system (Bayliss (1992)).1 I forecast that the political opposition to congestion pricing will be stronger in the U.S. than in any other industrialized country, and therefore that the U.S. will be among the last industrialized countries to adopt urban auto congestion pricing on a large scale. There are three reasons for my pessimism. First, Americans have a more individualistic and libertarian political culture; second, U.S. cities, with the exception of Manhattan, are significantly less congested that cities of similar size in Europe and Asia; and third, the extreme jurisdictional fragmentation in the U.S. is going to make congestion pricing significantly harder to implement. 1The adoption of congestion pricing in Oslo, Bergen, and Trondheim provides only weak evidence that the resistance to congestion pricing is weakening. Those cities did not introduce congestion pricing because they were persuaded of its effectiveness in reducing the volume of traffic or in spreading out the rush hour. Rather, the federal government refused to provide funding for the access roads, and the only way the cities had to raise revenue to finance the roads’ expansion and improvement was tolling (Ramjerdi (1994)). 4 The advocates of congestion pricing have recently shown more political awareness. They have not, however, shown the same sensitivity with respect to practical problems of implementation. Unless these are successfully addressed, a system could be so seriously flawed or be so unpopular that it would be withdrawn. The congestion pricing schemes in the Norwegian cities of Oslo, Bergen, and Trondheim are relatively crude, since they cover only major access roads and entail little variation in the toll by time of day. I would guess that these tolling systems have been (apparently) successful because the topography is such that major access roads are hard to avoid and because Norwegians have a collective spirit. But if such a tolling system were applied in the United States, at least a significant minority of commuters would take delight in taking detours on secondary roads around the tolling points and community residents would be on the warpath. This problem would be avoided if there were electronic road pricing on all roads. But comprehensive electronic road pricing will have problems of its own. First, there is the problem of who will pay for the installation of the metering hardware for a car. A sure way to guarantee unpopularity would be to require that car owners do so;2 all expenses associated with the metering hardware should come out of government revenue, and when the system is operational, out of toll revenue. Second, it will be vital that any system be user-friendly. Big city life is stressful, largely due to the many petty frustrations encountered. Bureaucratic hassles with billing, including overbilling, bureaucratic delays in getting the metering hardware installed and maintained, and unjust fines due to malfunctioning equipment, could condemn any system, however sound its basic design, to an early demise. Third, the billing system will have to be well-thought- out; prepayment cards which can be bought at convenience stores -- like lottery tickets -- and which would display how much of the card has been used up would seem a sensible option. Fourth, the metering hardware will have to be tamper-proof. One of the thorniest problems in the implementation of any system of congestion pricing is how to deal with non-residents. If the system were to entail tolling on only major access roads, then there could be toll booths where non-residents would purchase day passes. But a large proportion of non-residents might then choose to enter the city on secondary roads, reasoning that the jurisdiction would not bother to follow up on fine collection -- just as today many non-residents do not pay parking tickets. This problem could be mitigated by cooperative arrangements between jurisdictions or through regionalization of tolling. With electronic road pricing, the problem would be trickier. 2Kingston, Canada, finally adopted taxi meters only about five years ago. One of the contentious issues in moving from a zone to a meter system was who should bear the cost of the meters. 5 Presumably non-residents would have to purchase some gizmo -- a decal, a smart card, etc. -- that would identify the car as non-resident and which would be valid for a specified period. Such a system would be less than ideal since the toll paid by non- residents would be unrelated to their patterns of travel within the jurisdiction. In small jurisdictions, where a large proportion of road traffic is non-resident, this could seriously undermine the effectiveness of the tolling system. Another problem is that jurisdictions will have an incentive to delay implementation of congestion pricing because of uncertainty concerning which system to adopt -- the costs of irreversibility. Each jurisdiction will have an incentive to learn from the mistakes of other jurisdictions, and then to adopt the system that is the most cost- effective and hassle-free, particularly since the adoption of any system will entail sizable fixed costs. Smaller jurisdictions have an additional incentive to delay because of adoption externalities. The inner cities of Boston would, for instance, realize that it was in their mutual interests to adopt a common system, but there would be disagreement concerning what system to adopt, how to share costs and administration, etc. These problems would be less severe if tolling were managed at the metropolitan level. I do not want to paint too gloomy a picture. There are messy, practical problems of implementation with any policy, particularly when there is little experience to draw upon. But with a policy as politically sensitive as congestion pricing, it is imperative that these problems be thought through and resolved prior to implementation. Reformists tend to be dreamers. I remember a discussion with a group of transportation scientists in the early days of research on vehicle information systems. They envisaged widespread adoption of vehicle information systems by 1995. I was highly skeptical. I am equally skeptical of the rapid adoption of congestion pricing, and think that electronic road pricing will not be implemented in the United States for decades. I envisage the East Asian countries introducing electronic road pricing first, because of their receptiveness to technological innovation, their collectivist culture, their relatively high degree of political centralization, and the severity of traffic congestion in their cities. Northern Europe and France will follow. And major American cities will start adopting only after systems elsewhere have proved their effectiveness. Congestion pricing, in a crude form, followed by electronic road pricing, will come. It has to, since it is the only sensible, long-run solution to the urban traffic 6 problem. But, especially in North America, it will not come soon, and in the interim we must come up with alternative policies to deal with traffic problems. 1.2 Perfect versus imperfect congestion tolling We tend to speak rather carelessly about congestion pricing. On one hand, at a theoretical level, we use the term to refer to Pigouvian pricing -- imposing a tax or toll equal to the congestion externality, evaluated at the optimum -- so that price equals marginal social cost. On the other hand, at a policy level, we use the term to refer to any policy that raises the monetary cost of traveling in congested traffic. Let me refer to the former as perfect congestion tolling, and the latter as imperfect congestion tolling when price does not equal marginal social cost. The main point I want to make in this subsection is that all congestion pricing schemes that have been implemented or that are under study (Small and Gomez-Ibanez (1994)) are highly imperfect, and as a result additional policies to relieve traffic congestion are potentially welfare-improving. Thus, even if congestion pricing, in the policy sense, is in place, other policies are called for. Perfect road pricing would entail variation of the toll so that at each location and at each point in time, each driver would face the marginal social cost of all his travel and driving decisions. i) temporal variation Ignore stochasticity for the moment, so that the temporal variation of traffic is fully predictable. Then, at a specific location, a perfect toll would vary continuously over time to reflect the temporal variation of traffic. Current technology permits arbitrarily fine temporal variation of congestion tolls. Such fine variation would, however, be annoying since it would make the informal calculation of when to travel complex. Thus, one expects that electronic road pricing will entail step tolls, with several steps. An interesting issue is what proportion of the potential efficiency gains from tolling can be had with step tolls. In a model of bottleneck congestion (Laih (1994)), a toll that is optimal conditional on having n steps over the rush hour achieves a fraction n of the n+1 efficiency gains from a fully-flexible, time-dependent toll; for example, an optimized toll with three steps achieves three-quarters of the potential gains from tolling. Step tolls provide a good example of how complementary traffic management policies can be beneficial when tolling is less than perfect. Step tolls tend to create congestion peaks after step falls. The freeway congestion caused by these peaks can be 7 smoothed by limiting the flow of cars that enter the freeways through "on-ramp metering." ii) spatial variation Just as continuously varying the congestion toll over time would be annoying, so too would be continuously varying the toll over space. Thus, one expects that electronic road pricing will entail a zone toll system. This entails some inefficiency. But there is a far more important source of efficiency loss associated with the spatial variation of tolls, which has already been alluded to. Before sophisticated electronic tolling schemes à la Vickrey are introduced, it is likely that cities will start by tolling freeways and later freeways and major access roads to downtown. The problem with imposing congestion tolls on only some city streets is that drivers will tend to switch to those city streets without a toll -- the tolling system will create traffic diversion. Similarly, imposing congestion tolls only on urban freeways will induce drivers to switch to city streets. In both cases, the implementation of a congestion toll may actually worsen congestion. There are two effects. The toll will reduce the amount of traffic, which by itself reduces the amount of congestion, but it will also divert traffic to untolled links, and if the diversion is to "more congestible" links, congestion may increase. This is demonstrated in Figure 1 below, which depicts the MSC MSC 1 2 ¯ MPC 1 MPC 2 ¯ ¯ B A C N N 1 __ 2 N Figure 1: Tolling on only some streets can worsen congestion 8 extreme situation where demand is completely inelastic. There are N cars which travel between a given origin and destination along either of two routes, 1 and 2. In the absence of congestion tolling, the division of traffic between the two routes is shown by A, where the private costs are equalized. With perfect congestion tolling on both routes, the division of traffic is given by B, where the social costs are equalized. And with perfect congestion tolling on only the less congestible route -- route 2 -- the division of traffic is given by C. The important point is that the efficiency loss, relative to the first-best division of traffic, is larger when a perfect congestion toll is imposed on only the less congestible route that when there is no congestion tolling. Unfortunately, this situation is more than just a curiosum since it is likely that secondary urban streets are more congestible than major urban arterials which are in turn more congestible than urban freeways. Thus, it appears that a tolling scheme which covers only urban freeways, or only urban freeways and major arterials, may be highly inefficient. Complementary policies are needed to discourage cars from switching from tolled to untolled roads, or from making detours around tolling stations. To discourage cars from detouring around freeway tolling stations, tolling stations could be installed on entry and exit ramps. And commuting on secondary streets can be discouraged by introducing a complex system of one-way streets and by blocking some streets -- but this may give rise to more delay than it alleviates. ii) stochastic variation The efficiency loss due to stochastic variation in congestion depends on the magnitude of the stochasticity, the responsiveness of the congestion tolls to the stochasticity, and the responsiveness of drivers to both the stochasticity and the tolls. Consider a link with a time-varying toll that is not adjusted for stochastic variation in traffic density. When density is lower than average, the toll will be higher than optimal, and when density is higher than average, the toll will be lower than optimal. A completely responsive toll adjusts to the instantaneous realization of congestion. Vickrey has advocated responsive pricing in many contexts (Vickrey (1971)). An important point he has sometimes overlooked is that the effectiveness of responsive pricing is crucially dependent on the information that drivers have when making their travel decisions. Charging drivers for getting blocked behind an accident if they are unaware of either it or of the increase in the toll it induces is not only ineffective but also adds needless insult to injury. Thus, the effectiveness of a responsive tolling scheme in dealing with stochastic 9 variation depends on the quality and timeliness of information provided travelers about traffic conditions and the level of the toll.3 In recent years, there has been much discussion among transport engineers of vehicle information systems (Ref.?), and a number of experimental systems are being tested.(Ref.?) Congestion tolling will be more effective if it is combined with a vehicle information system which provides current information of toll levels and traffic conditions. Conversely, vehicle information systems will generate greater efficiency gains if they are combined with congestion pricing.4 iv) individual variation Individuals vary considerably in their quality as drivers. To charge each for the "average" congestion caused by a driver is to undercharge bad drivers and to overcharge good drivers, which entails an efficiency loss relative to the first best. This inefficiency calls for complementary policies which penalize bad drivers. To some extent traffic ticketing serves this purpose. But traffic ticketing as practiced is largely ineffective, and covers only gross violations such as not stopping at stop signs, speeding, and running red lights. There are many other forms of bad driving that exacerbate congestion but are not ticketable. The hardware that will be used in electronic congestion pricing could perhaps be adapted to impose an instantaneous fine for committing some of these offenses -- passing on the right, cutting in when the headway between cars is unsafely small, moving into an intersection so many seconds after the light has turned amber, not letting a car in when traffic is merging, accelerating rapidly, etc. v) accidents and breakdowns "U.S. studies in the Los Angeles conurbation show that more vehicle hours of delay results from extraordinarily and accidentally occurring traffic disturbances than from regularly occurring network overload during daily peak hours." (Busch, 1991). 3The proposed tolling scheme in Cambridge, England is responsive. Under the proposal, a driver will be charged on the basis of how long it takes her to travel between tolling stations. Unless drivers are informed of current traffic conditions, it would be preferable to charge them on the expected (e x ant e ) rather than realized ( e x post ) time. 4This theme is developed in Arnott, de Palma, and Lindsey (1991), Ben-Akiva, de Palma, and Kaysi (1991), and de Palma and Lindsey (1992). The basic idea is that, in the absence of congestion pricing, improved information will induce individuals to make travel decisions that are closer to privately optimal. But, with inefficient pricing, these need not be closer to socially optimal -- better information with distorted prices can make drivers worse off. Electronic congestion pricing and vehicle information systems exhibit strong technological complementarities. 10