OPTIMUM ROAD PRICING UNDER CONGESTED CONDITIONS Abstract: This paper looks at the question “what is the optimal congestion price?” Many published papers attempt to calculate the congestion price by estimating the short-run marginal cost and hence the ‘congestion externality’. This paper asks how that relates to the marketclearing price and shows that the optimum price tends towards the market clearing price1 when congestion is severe. Although the short run marginal cost is the “correct” measure, it is impossible to calculate in practice as it depends on knowing the shape of the speed/density curve close to capacity and the value of time for just those motorists who continue to travel. The purpose of this paper is to re-assert the importance of matching supply and demand as a basis for pricing policy and to demonstrate that this leads to a rule that is both pragmatic and defensible. Key words: Congestion, Pricing, Demand, Externalities 1 INTRODUCTION This paper looks at the question “what is the optimal congestion price?” We are hesitant to venture into a subject which has already been the subject of considerable scholarship. (See for example Dupuit, 1844; Walters, 1961; Vickrey, 1969; Newbery, 1990; Verhoef, 1999a). However it appears to us that the debate is in danger of becoming so esoteric that the basic principles may have been overlooked. Most papers appear to make at least one of four fundamental errors: 1 Treating the value of time as constant across the population. Using a demand curve or a diversion curve implicitly assumes that values of time differ2, but many ex post evaluations value the time savings of those who continue to pay at the same rate as those who are “tolled off”. It is not surprising that these studies find limited benefits – the surprise is that they find benefits at all Treating flow as the exogenous variable3. For low levels of traffic all demand can be met, but at higher traffic levels flow is constrained. Flow is an endogenous variable. It is possible to draw a graph showing speed, density and flow and to relate each to each other (Haight 1963) but that is irrelevant. The mistake it leads to is calculating the externality by differentiating with respect to the flow. This results in the preposterous conclusion that if a road is operating at capacity, the externality is infinite. We define the externality as those costs consequent on a decision that are not taken into account by the decision maker. The decision in this case is to travel or not. This results in a unit change in demand, but may have no impact on flow. Treating the cost curve as the supply curve and averring that there is equilibrium when demand equals supply. While the cost curve will be coincident with the supply curve under perfect competition, there is no reason to expect this to be the the price(s) at which the demand for road space during each time interval equals the road space available. 2 It might be argued that a diversion curve could result from benefits being different while the value of time is constant across the population. We show later that the benefits and the value of time must be related. 3 There is a long-running controversy in the literature on whether it makes sense to define supply and demand for trips in terms of flow. We are absolutely on the side that says that it doesn’t. 1 case for a monopoly or oligopoly. The problem we are addressing arises because at the point where the demand and the marginal private cost curve cross, the demand exceeds the capacity of the road (ie demand exceeds supply). Some authors regard the situation under hypercongestion as an equilibrium by regarding the time spent queuing as part of the generalised cost of travel. This is again confusing cause and effect. On this definition, the soviet union was a market economy4. Assuming a fixed demand. The purpose of road pricing is to change behaviour. If we assume behaviour is fixed it is not surprising that we see some strange consequences. Most published papers attempt to calculate the congestion price by estimating the short-run marginal cost and hence the ‘congestion externality’ (following Pigou, 1920). There appears to be an assumption that this will result in demand being constrained to capacity, but no proof that this is the case5. In this paper, we investigate the relationship between the concept of matching supply and demand as a basis for pricing policy and the maxim to price at social marginal cost. Unlike some authors, we are interested in the situation where “hypercongestion” occurs, or would occur in the absence of road pricing6. 2 BASIC CONCEPTS Road vehicles interact with one another. Addition of one more vehicle to a traffic stream has a tiny but quantifiable effect on other vehicles, which are slowed by a small amount. This economic externality is of little significance when traffic is sparse. When traffic is dense, the effect of one more vehicle is bigger, although still tiny, but the additional vehicle now affects a large number of other vehicles. The aggregate externality (summed over all vehicles) can be large. As traffic flow increases, travel speed reduces. As traffic becomes congested, speed falls off at an increasing rate. Speed cannot keep falling monotonically to zero because at zero speed there is zero flow.7 Hence there must be a point, call it ‘capacity’, at which the flow rate is a maximum. If flow is a maximum at capacity, then at capacity, the incremental vehicle slows the traffic stream by an amount that leaves the flow rate unchanged. What happens beyond capacity? If we plot speed against the actual vehicle flow rate, a level of flow exceeding capacity cannot be represented analytically. Yet, operating at capacity is no barrier to more vehicles attempting to use the road. When more vehicles attempt to use the road, speed continues to fall. What is more, the flow rate falls. Vehicles travel increasingly close to each other, increasingly slowly.8 A speed-flow relationship is 4 If you think of regular commuters adjusting their start times to ensure that they arrive at their destination within a particular time range, you might describe this as an equilibrium. This is not what we mean by supply equalling demand – rather it is a response to market failure. 5 Of course it will be if you treat flow as if it were demand. Flow can never exceed capacity. Some authors claim hypercongestion is a temporary phenomenon and don’t study it, or even claim it doesn’t exist. This seems to assume away the problem we are trying to solve. Since the speed does not fall significantly until close to capacity, the optimal congestion charge for a road operating below capacity is near zero. This is a non-problem. 6 7 Mathematically there is another possibility, which is that the speed falls asymptotically to a limiting value, but this is contrary to observation. 8 If this situation were to continue indefinitely, queues of traffic would increase without limit. Since peak periods are of finite duration, this does not happen. 2 not very useful when demand exceeds the road capacity. It is better to use demand, not flow, as the independent variable. Demand for what? Hills and Gray (1999) adopt vehicle trips as the basic unit of demand. We will call this aggregate demand. These trips will have a preferred arrival time and associated travel time and schedule delay costs. We are particularly interested in the situation where not all trips can be accommodated at the preferred travel time, so that some vehicles arrive at their destination early or late and thus incur schedule delay costs. Drivers will select their departure time in order to minimise their expected total travel costs. Now consider a time interval of some (arbitrary) length during this period. Then the number of vehicles wishing to start their journey during this time interval is the interval demand (referred to as “demand” unless it is not clear from the context). This demand is influenced by the expected travel time and schedule delay costs, plus tolls etc if any. The “flow” is the actual number of trips that commence in that time interval, while the “capacity” is the maximum achievable flow. All three measures are in units of vehicles per hour. An increase in aggregate demand would result in the demand in each time interval increasing. This may be accommodated by an increase in flow if demand in the interval is less than capacity and/or by some trips being displaced to later or earlier time intervals. Many practitioners refer to “speed-flow” or “volume-delay” curves as if the independent variable was the flow (volume). The curves commonly used for transport modelling are generally speed (or time) vs demand curves. This is a common source of confusion. Hills and Gray (1999) overcomes this by referring to speed flow curves as “performance curves” and time vs demand curves as “supply and demand curves”. DIGRESSION: The notion of capacity is ill-defined. There is no identifiable point at which an increasing traffic flow reaches a limit, viz the road’s capacity. It was once thought that 1800veh/h was the capacity of a single lane. That increased to 2000veh/h. But supersaturated flow rates of 2700veh/h per lane have been observed. Such supersaturated traffic flows are common on freeways with ‘ramp metering’, which injects vehicles into gaps in the approaching traffic stream. Supersaturated flows are unstable. A perturbation such as one vehicle braking a little too hard can surprise the following driver who has to brake harder still. This set ups a shock wave that causes all following vehicles to come to a standstill before carrying on, an effect which persists throughout the remainder of the peak period. Vehicles come to a stop at some point on the road for no apparent reason (that reason having long gone). After the perturbation, the traffic flow cannot recover to its former high rate (until the next peak period). 3 3.1 PRELIMINARY RESULT Simple Model We assume a one kilometre section of road operating under congested conditions (level of service “D” or worse). In these conditions all traffic moves at approximately the same speed. Let speed be v, in km/h, flow be F, in vehicle/h, and density be D, in vehicles/km. Speed is a function of the density ie v = f(D) Then the time to traverse a section of road t = 1/v = 1/f(D) The cost to the ith user is μi t = μi/f(D) +vehicle operating cost Ignoring the variation in vehicle operating costs the externality is E = Σ μi . dt/dD = μiD.(-f’(D)/f(D)2) where μ is the average value of time 3 The flow F =f(D). D If we introduce the concept of capacity as the demand at which flow is a maximum, then at capacity: dF/dD = f(D) + D.f’(D) = 0 ie at maximum flow, f(D) = - D.f’(D) Thus at maximum flow, E = μ D.(-f’(D)/f(D) 2 = μi/f(D) = μ t In other words, at capacity, the additional cost imposed by the marginal vehicle is exactly equal to the travel time valued at the average value of time. 3.2 Illustration Early work (e.g. Greenshields (1935), Almond, 1965, Homberger, 1982) on the relationship between speed and traffic flow assumed that speed decreased linearly with increasing density of vehicles. ie that the function is of the form v=v0 – k*D Then F = v*D = (v0v - v2)/k (a parabolic speed flow curve) or F = v0D –k D2 F is maximum when dF/dD = v0 – 2k D =0 ie when D= v0/2k Thus capacity = Fmax =(v0 – v0/2).v0/2k = v02/4k Travel time (in hours) for one kilometre is 1/v = 1/(v0 - kD). Differentiating gives the marginal cost when the density is increased by one veh/km, namely, k/(v0 - kD)2 hours per vehicle for one kilometre of travel. Since D vehicles are affected, the congestion externality is Dk/(v0 - kD)2. Table 1 displays these quantities with v0=50 and k=5/8. Flow reaches a maximum of 1000 veh/h at a density of 40 veh/km. At this point, the externality cost and the travel time are equal, at 2.40 minutes. The speed-flow curve has the property that when the density exceeds 40 vehicles/km, both the speed and the volume drop. This is a consequence of assuming that there is a maximum flow, and results in the phenomenon of the “backward-bending speed-flow curve” Table 1. Parabolic Speed-Flow Curve Density (veh/km) Speed (km/h) 5 10 20 30 40 50 60 70 75 Flow (veh/h) 46.9 43.8 37.5 31.3 25.0 18.8 12.5 6.3 3.1 234 438 750 938 1000 938 750 438 234 4 Externality (min/km) 0.09 0.20 0.53 1.15 2.40 5.33 14.4 67.2 288.0 Travel Time (min/km) 1.28 1.37 1.60 1.92 2.40 3.20 4.80 9.60 19.20 Figure 1. Typical Speed/Volume Relationship This backward-bending phenomenon is illustrated in Figure 1, for data observed on the northern motorway in Auckland, New Zealand. Note that rather than truly parabolic, the speed actually only declines slowly until the flow breakdown zone. The backward-bending phenomenon arises because flow is plotted as the dependent variable. In fact, the flow is not independent. The true independent variable is the demand. In this analysis we use density as a proxy for demand. This is equivalent to assuming that all vehicles wishing to travel enter the link and queuing occurs on the road9. 3.3 Benefit Maximisation We now introduce the concept of user benefit. For simplicity of exposition, suppose congestion is the only externality. This simplification can be removed later. We assume all vehicles travel at the same speed t. While this is not true in general, it will be true under the congested conditions that interest us. If there are no other externalities, then the net benefit to society will equal the sum of the net benefits to vehicle users. Net Benefit per hour Then = D.(B - t)* v = B.F - D where B is average benefit per user dN/dD = B.dF/dD + F.dB/dD - 1 Verhoef has shown that density is not an ideal proxy for demand. I don’t have this article, so am not clear whether this has any real impact on my conclusions. I suspect not. 9 5 Since dN/dD > 0 when D is small and dB/dD <0, this says that the net benefit is maximum before flow is maximum. In other words for benefit maximization, demand must be kept within the capacity of the road. Introduce a price p such that flow is at a maximum. Now consider an increase in price of dp that results in a reduction in demand of dD and a consequential reduction in time of dt. Then the benefit will change by μD.dt – p.dD This is an increase if p<μD.dt/dD Thus the price should increase if p < E (or if p <μt at capacity) The optimum price is (as expected) p = E. 3.4 Revenue Maximisation As an aside, we can also show that revenue is maximized when or D.(dp + μdt) = p dD p = D.dp/dD + μD.dt/dD define the elasticity e as e= -dD/dp * p/D Then p = - p/e + μD.dt/dD The revenue maximizing price is thus p= (E /(1+1/e) where E is the externality Hence if e is large (demand is price sensitive) p will be close to the optimum. If e is small (demand is price insensitive) p will be significantly higher than optimum. 3.5 Relationship between value of travel time savings and benefits Let the occupiers of vehicle i have a value of time μi and a benefit Bi Assume that there is a price p that ensures that the road operates at capacity Then p = μt = μdt/dD (μ = Σ μi/ n) But the benefit to the marginal user Bm = p + μmt = μt + μmt In other words Bi and μi must be related. Now assume an exogenous increase in demand (a new subdivision opens) but the benefit to former users is unchanged. The price must increase to keep demand to the same level. The former marginal user is “tolled off” and there is a new marginal user. The travel time and number of users is the same. For the new marginal user n: Bn = p + μnt = μ’t + μnt (p’ = μ’t where p’> p and μ’ > μ) The average value of time must be higher. [I think we can take this further …] 6 4 GRAPHICAL ILLUSTRATION We now present a simplified graphical representation of the problem. This is not intended to be rigorous, but to provide an intuitive illustration.10 P cost Fig 2 shows the demand for road space and the private user cost of travel. In the absence of road charges, the user cost is made up of time and vehicle operating costs. If the road is operating at less than capacity, an equilibrium is reached. What happens if the demand for road space exceeds the space that can be provided? Fig 3 shows the situation where supply of road space is fixed (independent of price). Drawing the supply curve this way may be controversial, but note that Q is demand not flow. There is no reason why demand should not exceed the quantity offered. However roads differ from simple goods in that people continue to attempt to consume (and therefore impact on other consumers) more than the amount supplied (ie the capacity) The point where the cost and demand curves cross is no longer an equilibrium. Because each user increases the travel time for all other users, the social cost of travel exceeds the cost perceived by the individual. Following Pigou, we assert that the benefit will be maximised if the user is charged the marginal social cost Fig 4. Equilibrium occurs where the social cost curve crosses the demand curve. (these are sometimes misleadingly labelled “average cost” and “marginal cost”) But what if following this rule results in demand exceeding the supply as appears to be happening in Fig 5? Increasing the price from p0 to p1 would enable more vehicles to pass at higher speeds. p0 therefore cannot be the optimum price. In fact since at p1 the externality equals the value of the travel time, and at p0 it must be greater than the travel time, p0 must be greater than p1. In other words demand at the point where the demand and marginal cost curves cross must be less than capacity. The figure as drawn must be incorrect. demand Q P cost demand supply Q P social cost private cost Popt Q P social cost P1 Po private cost Q 10 I only put these in to illustrate some of the points, I know using graphs can be misleading as you can draw things that are impossible in practice. If I do use them I must label the figures. 7 social cost P We conclude that charging a price equal to the externality (ie ensuring that the user pays the marginal social cost) will result in the demand being less than the capacity of the road as shown in fig 6. P1 Po private cost This confirms the approach of most authors, who assume that this will be the case. Q P But what happens if there is an exogenous increase in the demand for travel (eg a new subdivision is opened. This will result in a shift right of the demand curve. This would appear to push the ‘equilibrium’ point to the right so that demand exceeds capacity (Fig 7). We know this cannot happen What must happen is that the cost curves move anticlockwise as shown in Fig 8. Why? Because the cost curve is the travel time valued at the average value of time of those who remain travelling. As demand increases, a smaller proportion of the potential demand can travel and the average of their time values will consequently be higher. So both the private cost and the social cost are higher. social cost private cost Q1 Q2 Q SC2 SC1 P PC2 PC1 P2 P1 D2 D2 At p1 the demand will be greater than it was at p0, but it still will be less than C, the capacity. We conclude that as demand increases, flow will increase asymptotically towards a value of C. 5 OPTIMUM PRICING RULE We can now state our optimum pricing rule. Theoretical Rule We have confirmed the validity of the conventional congestion pricing rule, which is to set the price equal to the marginal social cost. What we have done is shown that this results in a price that is higher than the market clearing price, but that the optimum price tends towards the market clearing price as demand increases. We note that this relationship between the optimum price and the market clearing price implies that the value of travel time is not constant. Practical Rule Applying the theoretical rule requires knowledge of The shape of the cost curve (and thus the marginal cost curve) close to capacity The average value of travel time savings for those who actually travel, noting that this will depend on the ratio of satisfied to potential demand. 8 Neither of these can be determined with any certainty. Fig 9, taken from the Highway Capacity Manual, shows sample speed flow (not demand) curves estimated in practice. These show the relationship between the flow rate and the average speed for different free flow speeds (these would reflect different road conditions, roadside hazards, etc). The most notable feature is that they are almost flat until the last few hundred vehicles. Beyond the dotted line, the flow breaks down and the relationship is indeterminate. This suggests that rather than the gently sloping curves of our examples, real marginal cost curves are flat until close to capacity and then turn up sharply. Which in turn implies that the optimum price will be close to the market clearing price. The market clearing price is (again in theory) much easier to determine. The price can be adjusted in small increments up or down to find the price that maximises the flow. Unfortunately flows are unstable close to capacity, and small perturbations in the flow can cause the flow to collapse. It is thus better to price a little higher – the aim is the maximum consistently sustainable flow. Fig 9 shows five zones labelled LOS (level of service) A to E which depend on the density of the traffic. LOS C represents a range in which the influence of traffic density on operations first becomes marked. The ability to manoeuvre within the traffic stream is affected by the presence of other vehicles and average travel speeds show some reduction. LOS D represents a range in which ability to maneuver is severely restricted and speeds are curtailed because of the traffic volume. Only minor disruptions can be absorbed without the formation of extensive queues and the deterioration of service to LOS E and LOS F. LOS E represents operations at or near capacity and is quite unstable. At LOS E, vehicles are operating with the minimum spacing at which uniform flow can be maintained. Thus disruptions cannot be damped or readily dissipated, and most disruptions will cause queues to form and service to deteriorate to LOS F. Speeds are highly variable and unpredictable. LOS F (not shown) represents forced or breakdown flow (the backward section of Fig 1). The practical prescription is thus to price to ensure a LOS of somewhere between C and D. Pricing at this level would appear likely to be both economically and technically efficient. 6 CONCLUSION We have confirmed the validity of the standard prescription to price at social marginal cost, but demonstrated that calculating such a price would be a practical impossibility. On the other hand pricing to ensure that the level of service (LOS) is between C and D (as defined in the Highway Capacity Manual) provides an operationally implementable rule that will ensure both economic optimality and technical efficiency. 9 REFERENCES To come 10