A sample text widget

Etiam pulvinar consectetur dolor sed malesuada. Ut convallis euismod dolor nec pretium. Nunc ut tristique massa.

Nam sodales mi vitae dolor ullamcorper et vulputate enim accumsan. Morbi orci magna, tincidunt vitae molestie nec, molestie at mi. Nulla nulla lorem, suscipit in posuere in, interdum non magna.

Traffic congestion externalities

Road use congestion costs arise because travellers do not consider the impact of their travel decisions on the travel times of other road users. This creates what is the most significant externality associated with road travel. For the UK congestion is estimated by Samsom et al. (2001) to provide between 75-84 per cent of total estimated marginal external road costs.  For the US, Parry et al. (2007, p. 384) estimate that congestion costs constitute about 50 per cent of all distance-related externality costs.  There are no comparable estimates of the relative importance of congestion costs for Australia but the estimates of absolute size that are available suggest a substantial importance. 

The most plausible traffic congestion pricing solutions for Australia involve specific cordon pricing for some of its major cities along with pricing of major arterials and ring roads. Most plausibly these cities would be Sydney and Melbourne.  Other cities face high per km congestion costs but have limited aggregate congestion costs that would mean fixed costs of operating cordon pricing schemes would make them at best, only marginally viable.  There are serious issues of providing extra public transport infrastructure to encourage modal shifts from the use of private vehicles.


Australian per capita incomes should double over the next 20 years which will impact on transport demands and on the liveability of Australian cities.  Our increasingly affluent capital city populations will grow by 3 million with car traffic forecast to increase by 33 per cent, light freight by 41 per cent and heavy traffic by 39 per cent.  Growth in total traffic in already congested major cities will be considerable.  In the absence of pricing Sydney’s traffic will grow 47 per cent and Melbourne’s by 40 per cent (Gargett and Cosgrove, 2004).  Moreover, some key congestion impacts will occur in state capitals such as Brisbane and Perth.

These congestion impacts will be increasingly difficult to reverse using conventional supply measures as land supply constraints bite.  Increasingly pressure will fall on a more comprehensive reliance on public transport along with pricing congestion.  As cross price elasticities of demand for car travel with respect to public transport prices are low this puts added pressure on the need to price congestion.

Theory. The background static speed/flow theory of congestion pricing is exposited in Hau (1992). This theory ignores traffic bottleneck issues and the dynamics of pricing. Then, if motorists have the same value of time, charging motorists the dollar value of the marginal congestion costs they impose provides an efficiency gain since losses to travellers as a consequence of higher commuting costs are exceeded by the revenues yielded by the congestion charges.  An important feature of this outcome is that there are no uncompensated gains to motorists – road users are only better-off with appropriate redistributions of tax revenues.  This seems to create substantial ‘political economy’ issues in implementing congestion pricing since there are issues of trying to convince motorists that are compensated.   Uncompensated gains can arise if motorists attach different values to their travel since then those with a high value of travel time (generally those on high incomes) derive net benefits that exceed the costs to those making low-valued journeys.  There can also be gains to individual motorists if they prioritise the same trip differently though time.  Finally, if there is extremely high congestion so that, for example, large queues form – this is termed hypercongestion – then uncompensated gains can arise even if motorists have a uniform value of time.

The main shortcomings of this model are that it cannot adequately accommodate bottleneck constraints that cause hypercongestion. Nor does it account for the dynamics of driving decisions – motorists can not only choose if they travel but also when they travel.  They can avoid the worst congestion during peak periods by changing their departure time.  The ‘bottleneck’ congestion model addresses these issues (Vickrey, 1969; Arnott et al, 1993).  Here motorists have a preferred arrival time and incur increasing costs with departures from this time.  These costs are traded-off against time savings costs of leaving before or after peak travel periods.  With a single bottleneck a queue forms at peak travel times and then declines.  The optimal congestion toll then varies with time as an inverted U-shaped function which seeks to ‘flatten out’ the peak by inducing more drivers to leave earlier or later.  Unlike the speed/flow model this toll eliminates congestion entirely by eliminating queuing at the bottleneck.  Empirical estimation of this model has proven difficult but numerical simulations suggest that half the welfare gains yielded from congestion pricing now come from trip rescheduling rather than avoidance of peak travel altogether.  This reflects the fact that the costs of congestion arise because people are forced to shift away from their desired departure times as well as that they incur extra travel times.  This means that the welfare gains are much greater than in the speed/flow model and roughly of the order of the revenue collected.

Finally, it is worth emphasising that congestion can arise from non-recurring events such as accidents and bad weather. The types of models discussed above do not directly address such concerns but, with congestion charging, such events should have less severe impacts (BITRE, 2008a).

Empirics. There are various methods for estimating congestion costs – macro or aggregative approaches (an example is BTCE, 2007), network simulation-based (e.g. the Saturn model, Van Liet & Hall 1997) and most recently game-theoretic (Viauroux, 2006). Most empirical models use the speed/flow approach.

The most comprehensive and up-to-date study of congestion costs in Australia is BTCE (2007) and it relies on the speed/flow model.  This study provides estimates of the congestion costs in the eight Australian capital cities and base case (business as usual, ‘bae’) projections of these costs to 2020.  Conceptually the study estimates these correctly as deadweight loss (DWL) estimates.  The methodology is aggregative and relies on broad indicators of a city’s overall traffic rather than being based on network simulations in each city. It is intended to be appropriate for use in national level studies.  Comprehensive network modelling estimates are not yet available for Australia.

For 2005 the estimated DWL is $9.4b for the 8 Australian capital cities comprising private time costs ($3.5b), extra business time costs ($3.6b), extra vehicle operating costs ($1.2b) and extra pollution costs ($1.1 billion).  This aggregate figure is subject to considerable estimation uncertainty – sensitivity analysis suggests the true figure lies between $5-$15b. The worst congestion is concentrated in Sydney and Melbourne (costs are $3.5 and $3.0b respectively) though it is growing strongly in the smaller cities such as Brisbane. These estimates ignore costs of introducing new congestion control regimes, ignore other consequences of congestion other than those mentioned and assume a 1 per cent annual expansion in lane-kilometre road provision through to 2020. Under bau traffic will grow 37 per cent from 2005-2020 and congestion costs will grow strongly to $20.4b. 

The congestion costs identified are around 8 cents per km in Sydney, 7.5 cents in Melbourne, 6.5 cents in Brisbane, 5.4 cents in Perth and 5.2 cents in Adelaide.  The average for all 8 cities considered was 6.8 cents per km.  These figures are mid-range compared to national estimates for the US and the UK.  (Specifically the year 2000 weighted estimated average across all capital cities was 6 cents compared to an estimate of 6.6-7.2 cents US/km for the UK in that year and 1.5-3.3 cents US/km for the US (Parry and Small (2005, p 1282)).

All of these studies use aggregative speed/flow models which raises the question of whether they understate the gains from introducing congestion by pricing by not recognising gains associated with smoothing out bottleneck issues.

The cities that would be eligible candidates for reasonably comprehensive pricing, assuming estimates from the aggregative BITRE (2007) study are plausible, are Sydney and Melbourne.  Brisbane, Perth and Adelaide have quite high per km costs but much lower aggregate congestion costs and will be less obviously candidates for comprehensive pricing given the fixed costs of establishing a pricing scheme that are discussed below.  There is no case at all for attempting to capture congestion costs using congestion tolls outside these major cities.

Current Australian experience. Most road pricing in Australia – certainly CityLink and Eastlink in Melbourne – are geared toward achieving cost-recovery and do not target congestion explicitly.   Such tolls do not achieve the objectives of efficient congestion pricing since tolls do not increase at peak travel times when traffic is most intense.  Thus such tolls do not provide incentives to switch travel times away from peak periods to ‘smooth out the peak’.  There are inefficiencies here that reflect the levying of hefty charges on motorists when there is no congestion and hence when there are no congestion costs.  Private firms operate these tollways under contract with government so that an appropriate reform would be to renegotiate these contracts so that short-term marginal cost rather than uniform charges were levied (Clarke and Hawkins, 2006).  This move would eliminate DWLs stemming from current inefficient charges.   If private management of tolled roads is to be maintained then future contact designs for tolling should require efficient pricing perhaps along with perhaps a compensatory transfer to private operators from the government.

There is one significant attempt to introduce ‘time-of-day’ pricing in terms of tolling for use of the Sydney Harbour Bridge.  This tolling came into operation on 27th January 2009. The explicit objective was to ease traffic congestion by encouraging motorists to travel outside peak periods if possible.  On weekdays during the peak period 6-30am-9-30am and 4pm-7-00pm the maximum toll of $4 is charged.  From 9-30am-4-30pm the toll is reduced to $3 while from 7pm-6-30am the toll is lower still at $2-50. On weekends lower tolls are levied.  It is to early to make assessments on the impact on peak traffic flows since although the initial effects look favourable – peak traffic flows crossing the bridge have eased – the longer-term effects cannot yet be determined.  The important implication is that traffic authorities have made a move towards specific ‘time-of-day’ pricing, a move which essentially endorses the logic of the congestion pricing model.

The Sydney Harbour Bridge experience should provide useful evidence on the effects of differential time-of-day pricing on traffic congestion.

Second-best issues. The ‘first best’ case for comprehensive road pricing is immediate provided that the costs of comprehensive pricing (analysed below) do not exceed the DWLs avoided.  There are two complications that limit the applicability of this conclusion however (i) that comprehensive pricing may in fact be impractical so that only pricing of a subset of roads (including perhaps a cordon around a city’s CBD) may be feasible and (ii) that general equilibrium impacts of pricing may trigger additional distortions in such areas as labour markets.  General equilibrium complications are raised in Section 1.7 but, given that practical road pricing issues are almost always ‘second-best’ issues,  it is essential to discuss some ‘partial pricing’ second-best issues now.

Second-best ‘partial pricing’ issues are discussed in Choe and Clarke (2000), Clarke (2008).  The main insight is that if congestion pricing is restricted to a city cordon or to major arterial or ring roads in a city that the gains from congestion pricing can only be realised with auxiliary policies to address ‘second-best’ constraints and that, even with such side policies, gains may be substantially reduced.   

If pricing is imposed around a cordon then congestion can develop on its boundary.  This must be addressed through parking restrictions and other policies imposed around the boundary.  Pricing only major roads can lead to ‘rat-running’ along alternative unpriced routes that can create urban disamenities.   Pricing and investment policies are altered by these constraints as well.  Generally tolls should be set lower so that less DWLs are captured than would be without the constraints and installed road capacity on the tolled roads should also be lower.  This inevitably means lower welfare gains from congestion pricing.

If major arterial roads and ring roads are to be priced into cities – there is a strong case for doing so in Melbourne at least given extensive congestion in the city periphery (Clarke and Hawkins, 2006) – then there is a need to restrict traffic follows onto minor unpriced roads by using traffic architecture and perhaps traffic constraints.  These issues again reduce the gains from congestion pricing.

An important issue in large Australian cities is the congestion arising on major arterial roads and cross town roads at a considerable distance from the city.  These are intractable issues since it is difficult to eliminate ‘second-best’ issues on roads that run through low-population density areas on a city’s periphery and difficult to provide public transport infrastructure that will encourage modal shifts that are sought by pricing private vehicle use.  The complication is that many journeys i9n these areas are not along roads directed radially towards the city CBD.  Many journeys are cross-town to workplaces, schools and shops located elsewhere in the city periphery.  Clarke and Hawkins (2006) argue for liberalising the provision of a wide range of bus and mini bus services in such areas as a palliative but there remain difficult issues of transport planning here.

Implementation Costs. The feasibility of peak-load pricing has improved recently with developments in electronic metering technology. Fees can be collected electronically by in-vehicle transponders or by direct billing with global positioning systems.

Nevertheless proposals to price congestion face cost obstacles and vary in accord with the extent and complexity of the pricing.  Assessing these costs is difficult since congestion charging schemes are in their infancy and evolving technologically.  Even providing ballpark estimates of the costs of introducing comprehensive or piecemeal congestion pricing in Australia is complicated by the distinctively lower population densities of Australian cities (BITRE, 2008a).  It is not clear how transferable cost estimates are. Congestion problems are less severe but fixed transaction costs of observing and pricing traffic flows are spread over a smaller population base.  There are also greater costs of providing public transport to encourage sought-after modal shifts.

The costs of comprehensive electronic pricing involving charging by location and time using on-board units are large and uncertain. For Britain the Eddington report (DOT, 2006) estimated set-up costs in the wide range £10-62b with annual running costs £2-5b.  This wide range reflects uncertainties over the scope of charges and technology costs. There are also costs of compliance and enforcement.

In terms of partial approaches to congestion pricing much recent attention has focused on the London area pricing scheme.  The experience of this scheme in terms of costs has important implications for those seeking to imitate it. Set-up and operational costs of the scheme were considerably higher than initially expected. Implementation costs over the first two years averaged were twice what were expected during the planning phase partly because of much higher than expected compliance costs.   The annual costs of the scheme are estimated to be £163m (£143 million if establishment costs are included) compared to total annual benefits of £230m (Leape, 2007).  Thus costs comprise more than two-thirds of benefits.

In addition there are the costs of providing enhanced public transport services to meet demand created if the costs of private vehicle use are increased.

An interesting aspect of the scheme of relevance to Australia is that the London area scheme might be seen as an intermediate policy in a longer-term move to full-scale electronic pricing of all roads in the UK.  The success of the London scheme suggests that such a move may, in fact, be uneconomic since congestion pricing of London accounts for 80 per cent of the benefits from a national scheme (DOT, 2004).  In fact on interurban roads Newbery (2005) estimates that national distance-based pricing would be uneconomic and that a better scheme would be to simply rely on existing fuel taxes which would adequately capture many of the congestion costs.

Reasons for the success of the London scheme include the existence of a well-functioning public transport system and the presence of very high congestion levels.  Moreover, the existence of a ring road around London provided a convenient boundary.   It cannot be assumed that Australian cities with milder traffic congestion, less clearly defined boundaries and an already overtaxed public transport system would have an analogous cost experience.

In Sydney, for example, in 2004  there were 15.8 million trips per day of which 70.5 per cent were  by private vehicle, 17.2 per cent were walked and where only 9.9 per cent were by public transport.  Public transport trips are split about equally between bus and rail.  Achieving substantial reductions in traffic congestion would involve huge costs and hefty road user charges.  Reducing traffic volumes by one third would mean diverting 846,000 people from car to other modes which would place huge expansion pressures on the bus and rail system   (Stopher and Fitzgerald, 2008).   


R. Arnott, A. De Palma & R. Lindsay, ‘A Structural Model of Peak-Load Congestion: A Traffic Bottleneck with Elastic Demand’, American Economic Review, March 1993, 161-179.

R. Arnott & K. Small, ‘The Economics of Traffic Congestion’, American Scientist 82, 1994, 446–455

Bureau of Infrastructure, Transport and Regional Economics (BITRE), Australian Transport Statistics June 2008, June, 2008c.

Bureau of Infrastructure, Transport and Regional Economics (BITRE), Moving Urban Australia: Can Congestion Charging Unclog Our Roads?, Working Paper 74, BITRE, Canberra, 2008a. 

Bureau of Transport and Regional Economics (BTRE), Estimating Urban Traffic and Congestion Cost Trends for Australian Cities, Working Paper 71, BTRE, Canberra, ACT. 2007.

C. Choe & H. Clarke, ‘Pricing and Investment Decisions for a Tollway with Possible Route Substitution onto Alternative Congested Roads’, Australian Economic Papers, 39, 1, 2000, 84-91.

H. Clarke, ‘Targeting Urban Congestion: Equity and Second-Best Issues’, The Australian Economic Review, 41, 2, 2008, 177-186.


H. Clarke & A. Hawkins, ‘Economic Framework for Melbourne Traffic Planning’, Agenda, 13, 2006, 63-90.


Department for Transport, Feasibility Study of Road Pricing in the UK, London DFT, 2004.

Department for Transport, Transport Demand to 2025 and the Economic Case for Road Pricing and Investment, London, DFT, 2006.

A. Downs, Still Stuck in Traffic, Brookings Institution, 2004.

D. Fullerton, A. Leicester & S. Smith, ‘Environmental Taxes’, NBER Working Paper 14197, National Bureau of Economic Research, July 2008.

T.D. Hau, Economic Fundamentals of Road Pricing: A Diagrammatic Analysis, World Bank Policy Research Working Paper Series WPS 1070, Washington, DC: The World Bank, 99 pages.  1992.

I. Parry, ‘Pricing Urban Congestion’, Resources for the Future, Discussion Paper, 08-35, November 2008.

J. Leape, ‘The London Congestion Charge’,  Journal of Economic Perspectives, 20, 4, 2006, 157-176.

T. Litman with E. Doherty, Transportation Cost and Benefit Analysis: Techniques, Estimates and Implications, Victoria Transport Policy Institute, Victoria, BC, Canada, Second Edition, 2009. (Online at

Y. Kidokoro, ‘London-Type Congestion Tax with Revenue Recycling’, Economics Bulletin, 18, 1, 2005, 1-6.

P. Stopher & C. Fitzgerald, Managing Congestion – Are We Willing to Pay the Price?, Institute of Transport and Logistic Studies WP-08-03, University of Sydney, January 2008

D. Van Liet & M. Hall, Saturn 9.3 User Manual, The Institute of Transport Studies, University of Leeds, Leeds, 1997.

W. Vickery, ‘Congestion Theory and Transport Investment’, American Economic Review, Papers and Proceedings, 59, 1969, 251-261.

C. Viauroux, Structural Estimation of Congestion Costs, European Economic Review, 2006.

3 comments to Traffic congestion externalities

  • Uncle Milton

    Good post Harry.

    Something else governments could do is to encourage a greater spread of working hours, so that a standard working day could from 7 to 3, or 11 to 7. This isn’t possible for all people or all organisations but it must be possible for some.

  • hc

    Downs argues that in a ‘bottleneck model’ this will allow people to move closer to their desired departure times. Preexisting congestion levels will be re-established though some commuters will enjoy welfare gains since they travel closer to when they would ideally want to.

  • […] Urban Self-Sufficientist placed an interesting blog post on Traffic congestion externalitiesHere’s a brief overview…travel since then those with a high value of travel time (generally those on high incomes) derive net benefits that exceed the costs to those… […]