The Case for a Congestion Toll

In the previous section of this essay it was demonstrated that expansion of network capacity dos not eliminate the congestion externality and in some cases may not even reduce it. Furthermore, it was suggested that if the expansion in network capacity increased the network size, that this might worsen the congestion externality. An alternate approach therefor needs to be evaluated. One such approach may be the introduction of a 'congestion toll' on traffic using the network during congested times.

As highlighted earlier in this essay, if the price of internet traffic is based on the marginal private cost of the traffic, an inefficient outcome will result because of the presence of the congestion externality. In seeking a solution to the negative congestion externality it may be constructive to consider how other negative externalities are dealt with.

A common approach to dealing with negative externalities is through the introduction of corrective taxes. (Katz & Rosen (1994:624-627); McTaggart, Findlay & Parkin (1993:468-469);Kniest & Burgess (1998:116-117)) This is usually presented in the context of pollution, however it is possible to consider a congestion toll as a form of Pigouvian tax.¹ Let us examine Figure 10 below. Figure 10 represents a typical internet congestion experiencing congestion.

As stated previously, the market outcome occurs where the MPC curve intersects the demand curve. In Figure 10 this is at quantity Q1 and a price of P2. However the optimal quantity in Figure 10 is where price = MSC, a quantity/price combination of Q2P3. To bring about this outcome it is necessary to introduce a corrective 'toll', once congestion occurs ² . The optimal toll is of an amount equal to the size of the externality at the socially optimal outcome. That is, a toll of P3-P4. This has the effect of shifting the portion of the MPC curve located to the right of Q#, upwards by an amount equal to P3-P4. This is shown in Figure 10 by the MPC +toll curve. With the new curve MPC +toll , the new price of P3 will bring about a contraction in demand to Q2, the socially optimal level of output.  

In the above analysis, we have based the congestion toll on the size of the congestion externality at the point where demand = MSC. However, let us now consider the impact of changes in both demand and the application mix.

Firstly, let consider the impact of changes in demand for internet usage. In Figure 10, a congestion toll of P3-P4 was implemented to bring about an optimal allocation of resources. However, let us the consider the outcome if the toll (P3-P4) is implemented and there is an increase or decrease in demand. In Figure 11, demand curve D2 represents an increase in demand for usage and curve D3 represents a decrease in demand.

Figure 11 illustrates that with either an increase in demand to D2, or a decrease in demand to D3, a fixed toll will no longer produce the optimal outcome. In the case of D2, the price is too low and too much usage will be demanded. With D3, the price is too high and too little usage is occurring.³ Therefore in order to ensure that the socially optimal outcome is achieved, the congestion toll must be able to vary in order to match the changes in demand.

The analogy between internet and highway traffic has been introduced earlier in this essay. The analysis above suggests another similarity. In the same manner which highway traffic is dynamic and produces 'peak hour traffic jams', changes in demand for internet usage may produce periods of heavy congestion, as well as periods of little or no congestion. Time varying flows have been considered in the context of traffic flows by Carey & Srinivasin (Carey & Srinivasin(1993)) . Their conclusion was that an optimal toll must take account of changes in traffic flows during the day. While congestion is rising, the toll should be rising, and when congestion is falling, so should the toll.

In California, this theory has been put in to practice in the express lanes of California State Route 91.(Schwartz (1993)) Motorists using special express lanes pay a variable toll based on the level of congestion. The toll at peak times, for example, is up to ten times the toll during the hours of 10.p.m. and 4.a.m. This variable toll has yielded significant reductions in delays. It may be argued therefore that tolls on internet usage during congested periods should also take account of the variations in demand for usage which may occur⁴.

The ability to vary the toll is also important when one considers the possibility of changes in the application mix. Earlier in this essay it was argued that the shape of the MSC curve, and hence the size of the congestion externality, was influenced by the application mix. When more users are utilising applications with large ADU's or poor tolerance of delay, ceteris paribus, the congestion externality will be larger. Therefore altering the application mix will have a similar impact on the optimal congestion toll to changes in demand. If the proportion of application with large ADU's increases, then the size of the congestion externality at the socially optimal point will also increase and hence the optimal toll will also need to increase. If the application mix shifts to include a greater portion of applications with smaller ADU's, then the optimal toll will need to be decreased. 

As the above analysis has shown, the socially optimal outcome on the network can be achieved by the introduction of a congestion toll. However before considering the implementation of a congestion toll the potential problems and pitfalls should also be examined.

The implementation of a congestion toll has a number of problems. Firstly, such a system is usage-based ⁵. It therefore requires detailed recording of the amount of traffic attributable to different users and the different prices attached to that users traffic. The cost of implementing usage based systems, especially those which vary with congestion, is recognised as being a drawback to implementing such schemes.(Hazlett (1994:6);Kelly (1997:276)).

Another issue arising from the implementation of a congestion toll is the impact on growth and innovation which usage-based pricing schemes, such as a congestion toll, may have. For example, Kelly (1997:276) argues that the impact of a poorly designed pricing system on innovation could be more significant than the accounting costs. It is therefore also important to consider the impact of such a scheme on innovation and the development of new applications of the internet.

Next, determining the exact size of the congestion toll creates its own difficulties. This requires information such as the application mix at the time of the congestion, and detailed information on the nature of the demand curve and cost curves. This information may be difficult to obtain or may require that new technology be developed to gather the required information.

The introduction of new systems and technologies can create further difficulties. These problems arise because although "pricing can be changed quickly, the deployment of some mechanism inside the network (involving changes to the packet switches, or routers) can take years to accomplish."(Clark 1997:216) This further compounds the problems associated with the implementation of a congestion toll

Congestion tolls also raise the issues of who will pay the toll and who will collect it. For example, will the toll be paid by consumers to IAP's? Will it be paid by IAP's to an external body? Or will it be paid directly be consumers to an external agency? If paid to an external agency, what will the revenue from the toll be used for and how ill it be distributed? These are further issues which would need to be resolved prior to implementing a congestion toll.

A final problem with congestion tolls is that where there is a positive externality associated with a particular network size, the optimal congestion toll must be adjusted to take account of the positive externality. This situation will be explored more thoroughly in the next section of this essay.

This section of the essay has introduced the concept of a congestion toll. It is argued that a congestion toll provides a mechanism for reducing the congestion externality and bringing about the socially optimal usage of the internet. The possibility for variation in internet demand and application mixes, highlights the importance of ensuring that the toll is dynamic and responsive to changes in these factors. A congestion toll however has a number of potential problems which need to be evaluated before implementing such a scheme. Problems include accounting for traffic, the impact on innovation, who pays the toll and to whom, determining the optimal toll and the introduction of new technologies required to implement a congestion toll. In the next section of the determination of the optimal toll in the presence of positive externalities will be considered.

Endnotes 

1. A Pigouvian tax is "a tax levied on each unit of pollution in an amount just equal to the marginal damage it inflicts upon society at the efficient level of output" (Katz & Rosen (1994:624)).

3. In Figure 11, the red dashed lines shows the price quantity combination where the price is too low and too much usage is being demanded. The blue dashed line shows the price quantity combination where the price is too high and too little usage is being demanded.

4. To illustrate the variability in internet demand, consider the five minute average link utilisation statistics for Newcastle University (http://eccles.newcastle.edu.au/router-stats/). Since individuals users of the universities internet services pay a constant price for usage (P=0 in most cases), the observed variations in traffic can be attributed to changes in demand for internet usage. This would appear to suggest that demand for internet usage experiences significant fluctuations.

• Return to Contents Page
• Return to Previous Section
• Forward to Next Section

Last update 12-Jun-98. To visit my personal web site click here.