Impacts of long-range increases in the fuel economy (CAFE) standard.

Kleit, Andrew N.

I. INTRODUCTION AND BACKGROUND

In 1975 the U.S. government enacted legislation regulating the fuel efficiency of new motor vehicles. The apparent objective of this law is to reduce American dependence on foreign oil. After large increases in the price of petroleum in the late 1990s, and with continued conflict in the Middle East, corporate average fuel economy (CAFE) standards once again became a topic of interest. A number of proposals for changing the CAFE standards were discussed in Congress in early 2002, culminating in a defeat in the Senate of an amendment that would have required a 50% increase in the relevant CAFE standards. In place of that increase, the Senate voted to require the executive branch to examine the impact of further increases in the CAFE standard.

This work evaluates the long-term economic implications of raising the standard by 3.0 miles per gallon (MPG) above current levels. In industry parlance, this approach is sometimes referred to as "technology forcing." I choose 3.0 MPG because it reflects the focus of a May 2001 report by the vice president's task force on energy policy and because it reflects several legislative proposals in Congress. (1) The long term refers to a length of time such that manufacturers can adjust vehicle technologies and powertrain designs to reduce the amount of fuel required to move a given amount of mass or to achieve a given amount of performance or acceleration per gallon of fuel consumed. Previous work on CAFE standards, such as Kleit (1990) and Thorpe (1997), focused on short-term responses to higher CAFE standards, where technology forcing was not an option for manufacturers.

The analysis is conducted under two different scenarios. The first scenario is that CAFE standards are not binding in the current marketplace. The second scenario takes account of the current impact of CAFE standards and then analyzes the costs and benefits of increasing the standards. The costs of CAFE standards are broken down into two areas: the changes in consumer and producer surplus, and the increase in externalities caused by the increased driving that higher CAFE standards induce.

The plan of this article is as follows. Section II reviews the history of CAFE standards and briefly discusses the rationale for the regulation. Section III develops a model in which the current CAFE standard is assumed to be nonbinding. Section IV provides estimates of the impacts for a long-term 3.0 MPG CAFE increase under the assumption that the current standard is not binding. Section V then revises the model to take into account the arguably more realistic assumption that the existing CAFE standard was in fact binding. It then reports estimates for a long-term 3.0 MPG increase. Section VI provides a brief cost-benefit analysis of CAFE increases, and section VII provides a summary and conclusion.

II. BACKGROUND ON AUTOMOBILE FUEL ECONOMY STANDARDS

A Brief History of the CAFE Program

The CAFE program, as enacted in 1975, called for all manufacturers selling more than 10,000 autos per year in the United States to reach the mandated CAFE levels. CAFE levels rose from 19.0 MPG in 1978 to 27.5 MPG in 1985 and later years. A manufacturer's domestic and foreign cars are placed in separate CAFE categories, based on the domestic context of the vehicle. If a car has over 75% American context, it is considered domestic and placed in the domestic pool. Otherwise, it is placed in the foreign car pool (see Kleit 1990 for a discussion).

Light trucks (pickup trucks, sport-utility vehicles [SUVs], and minivans) were placed in a different CAFE pool than cars. When CAFE standards were originally passed, these vehicles represented a small fraction of the relevant market. By 2001, however, such vehicles made up approximately one-half of the sales of personal vehicles. In 2001, light trucks were required to reach 20.7 MPG. (There is no domestic and foreign division in the CAFE regulation for light trucks.)

If a review process finds that a manufacturer has not met the CAFE standard, that manufacturer is subject to a civil fine. The level of that fine is now set equal to $55 per car-MPG for each manufacturer. For example, if a manufacturer producers 1 million cars with an average MPG of 26.5, when the CAFE standard equals 27.5 MPG, that firm could be subject to a fine of $55 * 1,000,000 * (27.5 - 26.5) = $55 million. CAFE standards are calculated using harmonic averaging, as described below.

One important aspect of the impact of CAFE standards is that foreign firms appear to view the CAFE fine as a mere tax. Thus, several foreign firms, such as BMW and Mercedes-Benz, have routinely paid CAFE fines. In contrast, American firms have stated that they view CAFE standards as binding. Were they to violate the standards, American firms claim that they would therefore be liable for civil damages in stockholder suits. Even Chrysler, now owned by Daimler-Benz, has made it clear it is unwilling to pay CAFE fines. CAFE standards thus impose a shadow tax equal to the value of the relevant Lagrange multiplier on constrained domestic producers. Because the shadow tax of the CAFE constraint can be far higher than $55 per car-MPG, this implies that CAFE standards are not terribly binding on foreign firms and far more binding on U.S. firms.

As stated, higher CAFE standards were defeated in the U.S. Senate in early 2002. In addition, in the summer of 2002 the state of California passed legislation limiting the average output (by firm) from new automobiles of carbon dioxide per mile. Because the current method of reducing carbon dioxide emissions from vehicles is to raise fuel economy, California's law is effectively another form of CAFE regulation.

In 2003, the National Highway Traffic Safety Administration (NHTSA) issued rules-raising CAFE standards for trucks by 1.5 MPG by model year 2007. Congressional proposals to raise CAFE standards continue to be discussed in both houses of Congress.

If CAFE Is the Answer, Exactly What Is the Question?

At the margin, consumers equate the price of gasoline (the "internal" cost) with the marginal value of its consumption. In the absence of any externality, the marginal value of the use of a gallon of gas equals its price, and there is no public benefit from reducing the consumption of gasoline. Where externalities exist, economic theory is clear that the optimal policy is to set a level of stringency at which the marginal benefit of consumption of a gallon of gasoline equals the marginal cost plus the level of the relevant externality.

Thus the question becomes one of determining what the relevant externality is. A recent report of the National Research Council (NRC) attempted to quantify this externality. (2) The NRC concluded that the high level externality associated with the consumption of a gallon of gasoline amounts to $0.26 per gallon. (For the purposes of this work, I will assume this amount is both an average and a marginal benefit.)

The NRC divides the estimate into three components: $0.12 per gallon for adverse global climate effects, $0.12 per gallon for oil import effects, and $0.02 for changes in other pollution emissions at the refining level. Each estimate is subject to criticism. For example, there is a wide range of uncertainty about measuring the relevant externality for climate change. Several previous estimates imply that the climate change externality is between 1 and 4 cents a gallon, implying that the NRC may have overestimated this externality by a factor of at least three (see Toman and Shogren 2000).

The $0.12 per gallon estimate for oil import is also subject to criticism. The basis of this estimate is that the United States has market power in the purchasing of petroleum. Thus, if the United States were to reduce its demand for petroleum, the price of oil would decline. This estimate assumes, however, that CAFE changes can have a material influence on worldwide energy supply and demand. Because the United States only has about 26% of world oil consumption, (3) however, and there seems to be significant elasticity to the supply of oil, the United States does not appear to have any significant monopsony power in this market. Finally, it is unclear how reducing domestic consumption increases "oil security." Oil is traded in a world market, implying that it is difficult to insulate the United States from price shocks originating anywhere in the world. Reviewing such factors, Bohi and Toman (1996) conclude that there is no discernible oil import or energy security premium, though this question is subject to serious debate.

The NRC also allocates an externality of $0.02 per gallon for emissions of criteria pollutants from refiners. To the extent refiners are already under emission caps, it is unclear what effect higher CAFE standards would have on refinery emissions. CAFE standards, however, are not likely to reduce the emissions of traditional pollutants, volatile organic compounds (VOCs), oxides of nitrogen (NOx), and carbon monoxide (CO) from automobiles at the street level. These traditional pollutants are regulated by the Environmental Protection Agency on a per-mile basis. Thus CAFE does nothing to change the grams/mile emissions. However, if CAFE standards increase miles driven, via what is termed the rebound effect, they can be expected to increase emissions of traditional pollutants (see Espey 1997). Indeed, the results indicate that higher CAFE standards serve to increase the emissions of traditional pollutants. (4)

In addition, the gains to society from reducing the consumption of gasoline may be reduced or eliminated because gasoline is already a highly taxed good. (5) The question becomes one of how much of those funds are recycled back into funds to build and support roadways and therefore might better be viewed as user fees rather than attempts to combat externalities.

Greene (1997) asserts that a further rationale for CAFE standards is that purchasers of automobiles cannot truly estimate the fuel costs of their vehicles, and this is the "market failure" NHTSA alluded to in its 2003 truck proceedings. Nivola and Crandall (1995, 27) counter that fuel costs are prominently displayed for the consumer to read. Indeed, it is difficult to think of an automobile attribute that is better communicated to consumers. Even if consumers do have trouble obtaining and processing this information, however, it is unclear why the level of fuel economy offered in the market should be biased either above or below the efficient level.

III. ASSUMPTIONS OF THE MODEL

Many of the theoretical details of this model are similar to what I used in my previous work on the impact of CAFE standards in the short run, (6) and I will not repeat that discussion here. The model begins with a set of supply and demand elasticities and initial conditions in prices and quantities. It assumes that demand and supply curves are linear. It then imposes a set of implicit CAFE taxes on each constrained firm such that in equilibrium each constrained firm reaches the relevant CAFE standard. I begin the analysis under the assumption that CAFE is not currently binding.

Base Year and Categories

Given the availability of data, model year (MY) 1999 was chosen as the base year (all dollar figures are therefore in 1999 dollars). Light vehicles were broken down into 11 categories. Cars are broken into five categories (1) small; (2) midsize; (3) large; (4) sports; and (5) luxury. Trucks are broken down into (6) small pickups; (7) large pickups; (8) small SUVs; (9) large SUVs; (10) minivans; and (11) vans.

For convenience, the data are broken down into four firms, General Motors (GM), Ford, Daimler-Chrysler (domestic production), and Other. The other firms consist of several foreign concerns, such as BMW, Honda, Mercedes-Benz, and Toyota. The relevant numbers and the MPGs for each firm/category, are presented in Table 1. (7)

Transaction prices are generated by taking the average price for each category in the GM model supplied by GM economists. Data on MPGs was also supplied by GM.

Demand Side

Elasticities and cross-elasticities between categories are calculated using the internal GM demand model. The GM model starts by using conjoint analysis (similar to, for example, Roe et al. 1996) of different vehicle attributes, based on the responses of about 4,000 "clinic" participants. These results are combined with estimates from market data and other clinics of the interactions between new and used vehicles in different segments to estimate the own-price elasticity for each nameplate. Thus one of the outputs of the model is an estimate of the change in sales for each vehicle nameplate (e.g., Chevrolet Cavalier) as its price changes.

This information is, in turn, combined with survey data on the second choices of about 90,000 new vehicle buyers from all manufacturers to estimate the cross-elasticities among nameplates in a method similar to Bordley (1993). These results are then aggregated into own- and cross-price elasticities for all vehicles in a given market segment. The estimates are updated every year.

The model given to me starts with base quantities and prices for MY 1999. In response to a new vector of auto prices, it will calculate a new vector of quantities sold. I therefore calculate elasticities and cross-elasticities by raising the price of all vehicles in a particular category by 1% and determining the resulting percentage change in demand, not only in that category but for all other categories as well. Because 10.0% of cars are placed in a category designated as Other in the GM model, all elasticities are multiplied by 0.90. (The calculated elasticities are presented in Table 14.)

Supply Side

Consistent with my previous work, I assume that the supply side is competitive with an elasticity of supply in the short run of 2. (8) In the longer run, supply is generally more elastic, as firms have a longer time to adjust to new conditions. Therefore, for the long-run model, I assume an elasticity of supply of 4. Because CAFE standards divide cars into domestic and foreign fleets, this essentially implies for the purpose of this model that (Daimler) Chrysler is two firms, one domestic and one foreign.

A competitive model is used for two reasons. First, the market is becoming more competitive over time. For example, in 1999 the Big 3 American firms had less than 50%, of the small car market. Although the truck market in 1999 was apparently less competitive, all indications are that Asian firms will be entering these segments aggressively. Second, in the context of the 1999 market, where firms own both domestic and foreign production under the CAFE law, creating an Cournot-Nash equilibrium is more difficult. A Cournot equilibrium in this case is usually calculated by assuming that each firm has a fixed marginal cost and solving backward. In this case, however, that is unrealistic. Ford, for example, produces both Lincoln Continentals (domestic) and Jaguars (foreign). With the typical Cournot assumption and CAFE shadow taxes on Lincolns, Ford would simply shift all production out of Lincolns into Jaguars.

The differences between the results assuming a competitive model and using an oligopolistic model depend on the relative demand elasticities between larger and smaller cars (see Kleit 1990, 166-70). CAFE shadow taxes result in an increase in small car production and a decrease in large car production. When the demand for large cars is more elastic than the demand for small cars, this can reduce or even eliminate the deadweight loss associated with CAFE standards. The reason for this is that in such a market, relative to the production of large cars, there are too few small cars produced. In the demand structure employed here, however, the demand for large vehicles is generally less elastic than the demand for small vehicles. (9)

Treatment of Foreign Firms

As discussed in section II, CAFE standards call for a fine of $55 per car-MPG to be assessed to firms that do not meet the standard. Domestic firms have always asserted that for corporate policy and legal reasons, paying a fine is not an option. Therefore, the standard is modeled as binding on them. Foreign firms, however, appear to view the fine as equivalent to a tax. Several foreign firms with relatively small volumes over the years have paid this tax to the federal government. The larger foreign firms, however, have traditionally sold a mix of smaller, more fuel-efficient vehicle mixes and have not been bound by CAFE standards. This model therefore treats the foreign sector as unbound by standards.

The Technology Forcing Model

In my previous work (Kleit 1990), I assumed that manufacturers could not change the technology of their vehicles. This was done because the time period in question was short run, where technology innovation could not reach the market in time. In such circumstances, manufacturers must "mix shift" (sell fewer large cars and more small cars) to meet CAFE standards.

The circumstances evaluated here, however, relate to the long run. In this case, firms can meet higher CAFE standards by either mix shifting or improving their fuel-efficiency technology. Therefore, in this section I present a model of technology forcing, in which firms increase the fuel efficiency of particular vehicles in response to CAFE standards. (10)

According to the method by which the statute defines a firm's average MPG, a firm that does not meet the CAFE standard has total CAFE fine equal to

(1) F = [lambda] ([T.summation over i=1] [Q.sub.i] (S - MPG), S > MPG,

where [lambda] is the shadow cost of compliance, S is the CAFE standard, and [Q.sub.i] is the quantity of each model type i sold by the firm. Under the CAFE standard, a firm's MPG is defined as a harmonic average,

(2) MPG = [T.summation over i=1] [Q.sub.i]/[T.summation over i=1] ([Q.sub.i]/MP[G.sub.i]),

where MP[G.sub.i] is the mileage for each type of car sold by the relevant firm.

In this model, the firm faces total cost

(3) TC = [SIGMA] [C.sub.i]([Q.sub.i], MP[G.sub.i]) + F,

where [C.sub.i] represents the costs of one model and i is an index of models. Here the cost for MP[G.sub.i] is net of consumer demand for MPG. Thus I assume that a firm will invest in fuel efficiency in a world without CAFE standards as long as the firm finds it profitable to do so, that is, consumers are willing to pay for fuel economy increases. Under this assumption, the free market net marginal cost of fuel economy is 0, (11) as the marginal cost of fuel economy will equal the marginal return of fuel economy to the consumer.

I define the cost function for any vehicle type i as

(4) T[C.sub.i] = [C.sub.i]([Q.sub.i]) + [Q.sub.i][D.sub.i](MP[G.sub.i]),

where [D.sub.i] represents the cost of fuel economy. Note that here and in following references [D.sub.i] refers to the net cost of fuel economy. I will discuss how the marginal cost of fuel economy relates to the CAFE standard. Inserting the impact of fuel economy standards, total cost becomes

(5) TC = [T.summation over i=1] (C[[Q.sub.i]] + [Q.sub.i][D.sub.i][MP[G.sub.i]]) + ([lambda] [T.summation over i=1] [Q.sub.i]) x (S - [[T.summation over i=1] [Q.sub.i]/[SIGMA]([Q.sub.i]/MP[G.sub.i])]).

Minimizing total (net) costs with respect to MP[G.sub.i] yields

(6) dTC/dMP[G.sub.i] = [Q.sub.i](d[D.sub.i]/dMP[G.sub.i]) - [lambda]MP[G.sup.2][Q.sub.i]/MP[G.sup.2.sub.i] = O.

If the constraint is binding, MPG = S and

(7) d[D.sub.i]/dMP[G.sub.i] = [lambda][S.sup.2]/MP[G.sup.2.sub.i].

This defines the level of technology forcing undergone by the firm.

Given this and MP[G.sub.i], a firm has marginal cost of production in type i of

(8) dTC/d[Q.sub.i] = dC/d[Q.sub.i] + [D.sub.i](MP[G.sub.i]) + [lambda][(S - MPG) - [SIGMA] [Q.sub.i](1/[SIGMA] (MP[G.sub.i]) - ([SIGMA] [Q.sub.i]/[([SIGMA][[Q.sub.i]/MP[G.sub.i]]).sup.2]) x (1/MP[G.sub.i]))] = dC/d[Q.sub.i] + [D.sub.i](MP[G.sub.i]) + [lambda][S - 2MPG + (MP[G.sup.2]/MP[G.sub.i])].

In equilibrium, S = MPG, which implies

(9) dTC/d[Q.sub.i] = dC/d[Q.sub.i] + [D.sub.i](MP[G.sub.i]) + [lambda]S([S/MP[G.sub.i]] - 1),

This equation defines the CAFE-induced marginal cost of production, which is set equal to price in the next model. It also implies that an important element of the model is an estimate of [lambda], the shadow CAFE tax.

The model requires for both cars and trucks an estimated function

(10) d[D.sub.i]/dMPG = a[DELTA]MPG + b[([DELTA]MPG).sup.2],

where [DELTA]MPG equals the change in MPG above the unconstrained market level. I expected both coefficients a and b to be positive. Consistent with the discussion, in this model, D = 0 at the MY 1999 equilibrium level ([DELTA]MPG = 0), making the assumption that at this point CAFE standards were just nonbinding. Without binding CAFE standards, firms should invest in fuel economy up to the point where consumers are willing to pay for it.

For estimates of the cost of fuel economy (the coefficient b above) I take up the results in from Table 1 of Greene and Hopson (2003), here presented as Table 2. Greene and Hopson assume a formula Total Gross Costs = a([DELTA]MPG/MP[G.sub.0]) + b[([DELTA]MPG/MP[G.sub.0]).sup.2], where MP[G.sub.0] is the initial MPG. Of relevance are two curves Greene and Hopson estimated for the gross cost of fuel improvements. The curves are estimated from data from the Sierra Research Report and from 2002 data supplied by K. G. Duleep. (12) In addition, I take three curves estimated from NRC results, as presented in Greene and Hopson, although these have been subject to significant criticisms in the 2003 NHTSA truck proceedings.

To break these results down into net (of fuel economy) costs I assume MP[G.sub.0] is 27.5 for cars and 20.7 for tracks, and I assume gasoline costs $1.25 per gallon, Assuming a discount rate for fuel economy is more difficult. As Orazio et al. (2000) show, many automobile purchasers are liquidity constrained and therefore face implicit discount rates much higher than the market level. Following along this, Sutherland (2000) suggests that discount rates for these types of purchases should be raised far above market levels. The basic rationale for this is that auto purchases represent irrevocable commitments. Real options analysis, along the lines of Dixit and Pindyck (1994), implies that such commitments should attain higher interest rates. Given this line in the literature, I will take the medium point in Train's (1985) analysis and use a discount rate of 20% for consumer purchases.

Given a discount rate of 20%, I derive that drivers drive about 55,000 net present miles (this is taken from NHTSA's estimate of survival rates and miles driven for trucks) and gasoline costs $1.25 per gallon. (13) Therefore, the cost of gasoline to truck purchasers is 1.25 * 55,000/MPG. The initial cost of gasoline is therefore 1.25 * 55,000MP[G.sub.0], and the change in gasoline costs is (1.25 * 55,000/MP[G.sub.i]) - (1.25 * 55,000MP[G.sub.0]). (Of course, a similar formula applies for cars.)

Consistent with the discussion, I then eliminate all the data where the net marginal cost (changes in the cost of innovation plus changes in the fuel cost of driving) of fuel improvement is negative. I then estimate the net cost of fuel economy as a function of [([DELTA]MPG).sup.2]. The coefficients I derived are in Table 3. (14) For my base analysis, I take the median result from Table 3. However, 1 will also estimate the costs using the low and high cost scenarios from that table.

It should be noted that the long-term model implicitly assumes that the vehicle manufacturers have perfect foresight with respect to the demand for fuel economy several years into the future. With this perfect foresight, firms can reach all of the CAFE-mandated increases in fuel economy through technology forcing without the need to resort to far more expensive short-run mix shifting. Given the uncertainties inherent in the market for energy, which is crucial to the demand for fuel economy, the perfect foresight assumption would appear to result in a conservative estimate of the long-run cost of CAFE standards (see Kleit 1992).

The Gasoline Consumption Model

Once the relevant market equilibrium has been calculated, the impact of that market equilibrium on gasoline consumption must be estimated. Two important factors must be considered here. First, CAFE standards put some or most new car buyers in more fuel-efficient vehicles. This lowers their marginal cost of driving and causes them to drive more, a phenomenon referred to as the rebound effect. A recent study Greene et al. (1999), accepted by NHTSA in its 2003 proceedings and whose results I employ, finds that for every 10% that fuel economy is increased, driving increases 2%.

In addition, several studies imply that changing conditions in the new car market changes the actions of market participants in the used car market. Higher prices in new car markets makes used cars more attractive, reducing the scrappage rates of such cars. Here I adopt the empirical estimates I used in my 1990 article. (Because these estimates are in percentage terms, they are not obviously affected by the improvements in automobile durability.) As in my previous work, a (real) discount rate of 4% is used to analyze the net effect of gasoline savings in later years.

Pollution Impacts

To model pollution emissions, one must know the emissions per mile by model year and vintage. The difficulty here is that although regulators set the standards at one level, emissions over time are generally larger because on-board emission systems deteriorate and automobile users fail to maintain and repair them. Data on emission rates by model year and vintage were obtained from Air Improvement Resources. (15)

Unlike the rest of the model, I use 2004 pollution characteristics for the base year and years 1990-2003 for the stockage years. This is because these levels are set by government regulation, and we can have some confidence at this point in time that this will be the actual emissions from MY 2004 and later vehicles.

IV. RESULTS OF THE MODEL WHERE THE CURRENT CAFE STANDARD IS NONBINDING

Tables 4 and 5 present the results of raising the CAFE standard by 3.0 MPG for a one-year period far enough in the future so that it can be considered long run. U.S. manufacturers between them would lose about $463 million, and U.S. consumers would lose approximately $953 million. (Consumer welfare losses are calculated along the lines of Braeutigam and Noll 1984.) Total losses to producers and consumers therefore amount to $1.415 billion.

It is also necessary to calculate the increase in externalities caused by higher CAFE standards, along the lines of Table 6 CAFE standards lead to more miles driven, which leads to increased accidents and congestion. Edlin (1999, 4) estimates that accidents cost about 8 cents per additional mile driven. Lutter finds that the average congestion cost per mile of vehicle use is about 2.4 cents per mile. This is likely a conservative estimate of the congestion cost of extra driving, because the marginal cost of congestion is expected to be higher than the average cost. (16) On the other hand, NHTSA used a figure of 6.15 cents per mile. Here I will an average of the more conservative figures in the literature with an externality per mile of 10.4 cents (the Edlin estimate for accidents plus the Lutter estimate for congestion) with the NHTSA estimate (6.15 cents), yielding an externality cost of 8.27 cents per mile. (17)

In contrast, the economic value of the increases in pollution are relatively small. The federal Office of Management and Budget (OMB) values VOCs at approximately $0.51 to $2.36 per kg, and NOx at the same level. CO, at least according to the OMB, appears to have no marginal cost impact on the economy. (18) For purposes of this work, I choose an externality cost of $1.43 per kg for both VOCs and NOx.

As Table 6 indicates, miles driven increase 26.3 billion, or 1.62% of MY 1999 fleet levels. Pollution impacts are also presented in Table 6. Emissions of all three traditional pollutants rise between 1.73%, and 1.88%. This increase is due in large part to the rebound effect, which causes more driving and more pollution.

The net externality cost of higher CAFE standards, using the estimates presented in the preceding paragraphs, is $2.24 billion. As Table 6 indicates, almost 99% of the increased externality costs come from accidents and congestion.

In this model, gasoline consumption declines by 5.392 billion gallons, or 7.21% of total fleet consumption. As Table 7 indicates, the average cost of gasoline saved is $0.264 when using only consumer and producer welfare effects and $0.700 per gallon when externality effects are included.

The model does not explicitly generate a marginal cost per gallon saved. To generate such a figure, I ran the model 50 times, for MPG increases of 0.10 MPG at a time, for MPG increases ranging from 0.1 MPG to 5.0 MPG. I then ran a regression of total cost on gallons saved, gallons saved squared, and gallons saved cubed (costs in billion dollars, gallons saved in billions). Taking the relevant derivatives and solving for the amount of gasoline saved with a CAFE increase of 3.0 MPG yields a marginal cost per gallon saved of $0.582 when only producer and consumer effects are considered. The total marginal cost, including externalities, is $0.934.

Table 8 presents the simulations with both the low cost and the high cost of fuel improvement are used. Under the low-cost assumption, the marginal cost of a gallon of gasoline save is $0.761, and with the high-cost scenario it is $1.119. One of the factors these scenarios reveal is that the higher the welfare cost of imposing CAFE standards, the lower the externality effects. This is because as CAFE standards become more expensive for consumers, fewer cars are bought, and fewer automobile miles are driven.

All of the results of sections III and IV assume that the current CAFE standard is not binding at today's standard but would be binding for any increases. The NRC study, however, concludes that the existing standards are in fact binding, and this is consistent with my discussions with industry engineers and economists. I next turn to the case of binding current constraints.

V. THE EFFECT OF RAISING CAFE STANDARDS ASSUMING THE STANDARDS ARE ALREADY BINDING

It is conceptually possible to calculate the impact of increasing CAFE standards given that they are already binding. This is an important consideration. It is a well-known result of public finance economics that the losses due to taxation are a function of the taxes squared, rather than simply a linear function of the taxes. If CAFE standards were already binding in MY 1999, it implies that the approach used underestimates the true loss to the economy of raising CAFE standards. The first part of this section outlines the several-step process for estimating this loss. The second part applies the methodology of the first part to this market.

Modeling the Existing CAFE Shadow Tax

To make the estimation of the losses to increasing an already-binding CAFE standard, I take the following steps. First, I assume that U.S. firms in MY 1999 engaged in mix shifting but not technology forcing as a result of CAFE standards. Second, I obtained input ratios by car type for General Motors (GM) cars (with a Chevrolet Malibu having an input ratio of 1.0). I assume that the marginal costs of production for cars are a linear function of these input ratios. Third, I assume that marketing and other costs (including goodwill) constitute a constant fraction R of marginal costs. (Recall that because a competitive model is being used here, price equals [total] marginal cost.) In this context, assume that the shadow CAFE tax per MPG on vehicles is L. Also assume that the PT equals the pass-through rate, the rate at which changes in taxes are passed through to the final consumer. This implies the equation

(11) (1 + R)M[C.sub.i ] + PT * L(S[(S/MP[G.sub.i]) - 1]) = [P.sub.i],

where [P.sub.i] equals price of car i, M[C.sub.i] equals marginal cost of car i, S is the implicit CAFE standard (here it would be the fleet MPG that actually occurred in MY 1999), MP[G.sub.i] is the miles per gallon achieved by car i, and L(S[S/ MP[G.sub.i]] - 1) is the formula for per-car MPG, derived from CAFE harmonic averaging. Because I only have data on GM models (and only sufficient data on GM car models) I estimate the value of L using least squares across GM car models.

Fourth, the implicit tax L calculated here applies directly only to GM cars. I assumed that Ford and Chrysler have similar CAFE taxes on their cars. Because they currently have CAFE levels roughly equivalent to GM's, their implicit taxes may be similar to GM's. (In fact, Ford and Chrysler had slightly lower fleet MPGs than GM in MY 1999.) I also assume that the CAFE tax on trucks is equal to the tax on cars. Because there is substantial evidence that U.S. manufacturers have had more difficulty reaching their CAFE standards for trucks rather than cars, this assumption serves to underestimate the relevant loss to society.

Fifth, given an estimated CAFE shadow tax L, I ran the 1999 model (the one presented already) backward, setting the CAFE tax at -L, generating a new equilibrium in prices and quantities.

Sixth, the supply curves calculated for the initial model will have the relevant values subtracted from its intercept terms to recalibrate the model for the unconstrained scenario.

At this point I have a new initial no-CAFE or free market equilibrium with demand and supply curves. The model can then be run for firms to reach a particular CAFE standard. Changes in welfare from this equilibrium to the higher CAFE standard equilibrium can then be calculated.

An additional problem comes from the multiproduct nature of the market. This implies that taxes on one type of vehicle will impact prices of other types of vehicle. Given this, it takes some work through manipulation of supply and demand matrices to determine the pass-through rates for each type of vehicles. This work is available on request.

For this model, the results of the impact of a CAFE tax by vehicle type for GM cars are presented in Table 9. For every dollar of CAFE shadow tax, dP/dt represents the pass-through rate for GM. For example, every dollar of CAFE tax reduces the price of small cars by about $0.84, and increases the price of luxury cars by about $0.88.

Table 10 presents the estimation results for the level of the CAFE tax in MY 1999. The dependent variable is the price in thousand dollars of GM cars. The two independent variables are the input ratios and the coefficient on the CAFE tax, as implied by Table 9. The model is run with and without a constant term. However, the estimated constant term in model 1 has a very low t-statistic. Model 2, which is run without a constant, has large t-statistics and a high [R.sup.2] (0.950). The estimated shadow tax from this estimation is $1,652/MPG, and this is the level used in the simulations to follow. (19)

Welfare Implications of Raising CAFE Standards Given that Standards Are Already Binding

Tables 11 and 12 present the welfare changes as a result of raising the long-run CAFE standard 3.0 MPG above the 1999 level, assuming a short-run tax of $1,652 was binding in MY 1999 and using the median cost of fuel technology scenario. As expected, the harm to the economy is greater than that in the previous long-term model.

Total producer and consumer welfare losses to society from the MY 1999 equilibrium of raising the long-run CAFE standard 3.0 MPG are $1.901 billion. Miles driven rise 26.151 billion from the MY 1999 equilibrium. Emissions of VOCs, NOx, and, CO rise between 1.74% to 1.91% from the MY 1999 equilibrium. Total externality costs are $4.089 billion. Consumption of gasoline is reduced 5.240 billion gallons. The average cost of reducing a gasoline externality is $0.363 from the MY 1999 equilibrium including only producer and consumer welfare terms. Including externalities, the average cost of reducing a gasoline externality is $0.780.

Total U.S. producer and consumer losses from the no-CAFE equilibrium of raising the long-run CAFE standard 3.0 MPG are $1.946 billion. Miles driven rise 32.792 billion miles from the no-CAFE equilibrium. Emissions of VOCs, NOx, and, CO rise from 2.12% to 2.26% from the no-CAFE equilibrium. The total cost of CAFE related externalities is $2.743 billion. Gasoline consumption falls 6.6 billion gallons from the no-CAFE equilibrium. The average cost of reducing a gasoline externality from the no-CAFE equilibrium is $0.295, including only producer and consumer welfare losses, and $0.710 when including all losses.

Similar to before, I use the results of Table 11 to estimate the marginal cost of saving a gallon of gasoline. I generate 50 data points, increasing the required fuel economy 0.1 MPG each time. I then regress gallons saved, linear, quadratic, and cubic terms on total cost. I then can estimate the derivative of total cost with respect to gallon saved. Given this, the marginal cost of reducing a gasoline consumption externality is $0.695 in producer and consumer welfare terms, and $1.050 when including externalities.

Figure 1 graphs out the marginal cost per gallon saved as a function of the increase in MPG. When the CAFE increase is 0.0 MPG (form current levels), the marginal welfare cost per gallon saved is estimated to be $0.10, with the externality cost estimated to be $0.440, for a total welfare cost per gallon saved of $0.54. As the CAFE standard increase, total marginal cost increases. However, the marginal external cost decreases. As discussed, this results because as the cost to consumers of CAFE standards increases, fewer automobiles are purchased, and therefore fewer miles are driven.

[FIGURE 1 OMITTED]

Results for low and high cost scenarios are laid out in Table 13. In the low-cost scenario the marginal cost of fuel economy is estimated to be $0.827 per gallon, whereas that marginal cost is estimated to be $1.283 in the high-cost scenario.

VI. COST-EFFECTIVENESS AND COST-BENEFIT ANALYSIS

This section asks two questions. First, do the benefits of CAFE standards exceed the costs? For benefits, I use the NRC figure of $0.26 per gallon of externality, although one could use NHTSA's far lower figure of 8.3 cents. (20) Second, are CAFE standards cost-effective? In this context, this means comparing the cost of CAFE standards to the cost of a gasoline tax that would generate equivalent gasoline savings.

The discussion so far indicates the impact of a CAFE increase of 3.0 MPG. For cost-effectiveness measures, I need to know the level of the tax that would generate equivalent gasoline savings. Pindyck (1979) indicates that the elasticity of demand for gasoline over a five-year period is approximately 0.49, a number that is roughly halfway between short-run and long-run estimates by Dahl and Sterner (1991). I will also assume a base gasoline consumption in the United States of 120 billion gallons at an initial price of $1.25 per gallon and that the demand curve for gasoline is linear in shape. Using these assumptions, it is straightforward to determine the gasoline tax needed to reach the desired level of gasoline savings.

Economic theory indicates under these assumptions that the total loss to society from such a tax equals one-half the tax times the reduction in the number of gallons of gasoline consumption, and the marginal loss equals the level of the relevant tax. (21) Thus the comparison here is between the gasoline savings of a one-year CAFE standard increase of 3.0 MPG (announced credibly several years in advance so that new technologies could be introduced) and an increase in the gasoline tax years in advance that has long-run impacts in the same year as the hypothetical CAFE standard increase.

Assuming that CAFE standards were not binding in 1999, the median cost scenario implies that an increase in the CAFE standard of 3.0 MPG decreased gasoline consumption by 5.392 billion gallons, for an average cost per gallon in the base scenario of $0.70. This is about 2.7 times the $0.26 per gallon benefit estimated by the NRC.

Using estimates for the long-run elasticity of gasoline demand, a tax of $0.115 per gallon would be required to induce savings of 5.392 billion gallons of gasoline. Thus a tax would impose an average cost on society of half of that amount, or $0.05775 per gallon. In other words, the 3.0 MPG increase in the CAFE mandate would cost society 12 times more than a gasoline tax increase saving the same amount of fuel. At the margin, saving a gallon of gasoline costs the economy $0.93 in this scenario, far higher than the $0.26 benefit estimated by the NRC.

Perhaps the more appropriate scenario is the one that compares a mandated CAFE increase to a binding CAFE constraint in 1999. In that scenario, gasoline consumption falls by 5.240 billion gallons per year, for an average cost in the base scenario of $0.78 per gallon.

This is three times the NRC estimated benefits. This same reduction in gasoline consumption could be achieved by a gasoline tax increase of $0.111 cents per gallon, implying social costs of 0.05505 per gallon. Thus, the $0.78 cost per gallon of CAFE standards would be approximately 14 times more costly to society than the tax that would save the same amount of gasoline. At the margin, a gallon of gasoline saved by CAFE standards costs $1.05, four times the NRC estimated benefits.

VII. CONCLUSION

Increases in CAFE standards above current levels are neither cost-effective nor cost-beneficial. Assuming that current CAFE standards are already binding, in the long-run median cost scenario, increasing the CAFE standard by more than 3.0 MPG would impose additional costs of over $4 billion per year and reduce gasoline consumption by about 5.2 billion gallons per year. This amounts to about 12 times the cost of a gas tax increase that would save the same amount of fuel. The long-term marginal costs of the 3.0 MPG mandate would exceed the additional benefits of avoided gasoline consumption externalities by a factor of four to one.

CAFE standards suffer from a wealth of difficulties. They discriminate against American production, they encourage people to drive more, and retain their used vehicles longer, increase automobile accidents and congestion, the emissions of several pollutants, and they have the potential for serious consumer injury. If policy makers desire to reduce energy consumption, it would seem they should focus their attention on raising energy taxes.

ABBREVIATIONS

CAFE: Corporate Average Fuel Economy

GM: General Motors

MPG: Miles Per Gallon

MY: Model Year

NHTSA: National Highway Traffic Safety Administration

NRC: National Research Council

OMB: Office of Management and Budget

SUV: Sport Utility Vehicle

VOC: Volatile Organic Compound

TABLE 1

Initial Conditions--Prices and Quantities (Model Year 1999)

 Initial Totals by Class Initial Quantities by Firms
 (millions of units)
 Prices Quantity
Class ($000) (million) MPG GM Ford Chrys. Forgn.

 1 14.336 2.057 33.53 0.589 0.313 0.096 1.059
 2 18.508 2.921 27.26 1.255 0.640 0.395 0.631
 3 21.710 1.840 26.86 0.267 0.363 0.243 0.968
 4 21.607 0.506 26.03 0.104 0.214 0.004 0.184
 5 30.365 1.102 24.44 0.240 0.117 0.000 0.746
 6 17.345 1.718 22.68 0.328 0.783 0.257 0.350
 7 23.424 1.596 18.83 0.845 0.435 0.316 0.000
 8 26.284 1.463 20.24 0.352 0.429 0.254 0.428
 9 31.296 1.474 18.30 0.331 0.340 0.300 0.503
 10 25.157 1.074 23.49 0.245 0.258 0.328 0.242
 11 20.611 0.969 18.90 0.320 0.202 0.437 0.000

 Initial MPG by Firms
 (miles per gallon)

 GM Ford Chrys. Forgn.
Class

 1 32.52 33.61 31.92 34.26
 2 27.15 26.15 27.29 28.71
 3 26.05 24.65 25.46 28.46
 4 24.84 26.10 22.62 26.75
 5 23.80 22.78 -- 24.94
 6 24.56 22.61 19.25 23.59
 7 19.34 18.43 17.60 --
 8 21.36 19.78 20.85 23.17
 9 16.91 16.36 18.53 20.20
 10 23.72 22.44 23.70 24.46
 11 19.78 17.77 18.04 --

TABLE 2

Gross Cost of Fuel Economy
Improvement from Greene and Hopson
(2003)

Data Source Cars Trucks

Sierra gross a = 1097, b = 7480 a = 2102, b = 6183
Greene-Hopson- a = 16, b = 9025 a = 219, b = 8772
Duleep-Gross
NRC high cost a = 4211, b = 1430 a = 3917, b = 1020
NRC average a = 2461, b = 2359 a = 2529, b = 1588
NRC low a = 1337, b = 2404 a = 1559, b = 1689

TABLE 3

Cost of Innovation Results

 Cars Trucks

Sierra adjusted net 12.2 19.5
Greene-Hopson- 14.2 21.93
Duleep adjusted net
NRC high cost 12.9 11.9
NRC average 5.81 15.35
NRC low 5.25 7.52
Median 10.07 15.24
Low 5.28 7.52
High 14.2 21.93

TABLE 4

Price and Output Effects of CAFE Increase of 3.0 MPG for Both Cars
and Trucks

 Output by Firms
 Totals by Change from (millions of
 Class Initial units)

 Prices Quantity Prices Quantity GM Ford
Class ($000) (million) ($000) (million)

 1 14.290 2.079 -0.046 0.022 0.606 0.329
 2 18.555 2.898 0.047 -0.023 1.237 0.631
 3 21.766 1.829 0.056 -0.011 0.261 0.353
 4 21.680 0.503 0.073 -0.003 0.101 0.213
 5 30.463 1.101 0.098 -0.001 0.233 0.113
 6 17.330 1.705 -0.015 0.013 0.340 0.794
 7 23.592 1.701 0.168 -0.017 0.840 0.429
 8 26.335 1.596 0.051 0.000 0.353 0.424
 9 31.478 1.452 0.182 -0.011 0.320 0.329
 10 25.105 1.082 -0.052 0.008 0.248 0.260
 11 20.755 0.953 0.144 -0.016 0.319 0.196

 Output by Firms
 (millions of Change of Output by Firms
 units) (millions of units)

 Chrys. Forgn. GM Ford Chrys. Forgn.
Class

 1 0.099 1.046 0.017 0.016 0.003 -0.014
 2 0.393 0.638 -0.018 -0.009 -0.003 0.006
 3 0.237 0.978 -0.006 -0.010 -0.006 0.010
 4 0.003 0.187 -0.003 -0.002 0.000 0.002
 5 0.000 0.755 -0.007 -0.003 0.000 0.010
 6 0.248 0.349 0.012 0.011 -0.009 -0.001
 7 0.310 0.000 -0.005 -0.006 -0.006 0.000
 8 0.256 0.431 0.001 -0.005 0.002 0.003
 9 0.299 0.515 -0.011 -0.011 -0.001 0.012
 10 0.335 0.240 0.003 0.001 0.007 -0.002
 11 0.429 0.000 -0.001 -0.006 -0.008 0.000

TABLE 5

Producer and Consumer Welfare Impacts of CAFE Increase of 3.0 MPG for
Cars and Trucks

 U.S.
 GM Ford Chrysler Foreign Total

Change in -0.163 -0.199 -0.101 0.220 -0.463
Producers
Surplus
($ billion)
Change in -0.953 Total U.S. -1.415
Consumer Change in.
Surplus Surplus
($ billion) ($ billion)

TABLE 6

The Impact of Standards on Externalities

 Pollution Impacts
 (million kgs)

 Miles Driven (millions) VOCs

Original MY level 1,652,362 578,722
CAFE-induced change in 26,304 9383
MY level
Change in stockage levels 468 628
Total change 26,772 10,011
Percent change 1.62 1.73
External cost per unit $0.104/mile $1.43/kg
Total external cost $2.214 billion $0.014 billion

 Pollution Impacts (million kgs)

 NOx CO

Original MY level 453,814 4,855,112
CAFE-induced change in 7684 81,621
MY level
Change in stockage levels 717 9748
Total change 8401 91,369
Percent change 1.85 1.88
External cost per unit $1.43/kg --
Total external cost $0.012 billion Total externality
 cost: $2.240 billion
 Total cost:
 $3.665 billion

TABLE 7

Impact of Higher Standards on Gasoline Consumption

MY pre-CAFE gas. cons. 75.045 Average cost of $0.264/$0.700
(billion gall.) gasoline externality
 saved without and
 with externalities
Change in MY gas. cons. -5.425
(billion gall.)
Change in stockage 0.033
consumption (billion
gall.)
Net change in consump- -5.392 Marginal cost of $0.582/$0.934
tion (billion gall.) gasoline externality
 saved (inferred)
 without and with
 externalities
Percentage change in -7.19
consumption

TABLE 8

Welfare Effects--3.0 MPG Increase, Low- and High-Cost Scenarios
CAFE Nonbinding Binding; Model

 Low-Cost High-Cost
 Scenario Scenario

Changes in Producer and Consumer
Surplus ($ billion)
Foreign firms 0.118 0.335
U.S. firms total -0.253 -0.693
Change in consumer surplus ($ billion) -0.513 -1.454
Change in U.S. total surplus ($
billion) -0.766 -2.147
Change in gasoline consumption (billion -5.345 -5.449
gallons)
Externalities costs
Change in miles driven 28.100 24.982
Total externality costs ($ billion) $2.372 $2.092
Total costs ($ billion) $3.138 $4.239
Average cost of reducing gasoline $0.143/$0.587 $0.394/$0.778
externality without and with
externalities
Marginal cost of reducing gasoline
externality (inferred) without and with $0.329/$0.761 $0.854/$1.119
externalities

TABLE 9

Pass-Through Rates by Car Type

Type Description MPG dP/dt

 1 small car 32.52 -0.839
 2 midsize car 27.15 0.040
 3 large car 26.05 0.228
 4 sports car 24.84 0.783
 5 luxury car 23.80 0.876
 6 small truck 24.56 -1.168
 7 large truck 19.34 0.246
 8 small suv 21.36 -0.300
 9 large suv 16.91 1.171
10 minivan 23.71 -1.253
11 van 19.78 0.007

TABLE 10

Estimating the 1999 CAFE Tax
(t-Statistics in Parentheses)

 Model 1 Model 2

Constant 0.725 (0.39) --
Input ratio 15.271 (1.48) 15.835
CAFE tax 1.986 (1.12) 1.652
[R.sup.2] 0.951 0.950
Number of observations 25 25

TABLE 11

Welfare Effects--3.0 MPG Increase
CAFE Already-Binding Model

 Change from Change from
Changes in Producer MY 1999 No-CAFE
Surplus ($ billion) Equilibrium Equilibrium

General Motors -0.309 -0.337
Ford -0.346 -0.385
Chrysler -0.148 -0.154
Foreign firms 0.192 0.16
U.S. Firms total -0.8038 -0.876
Change in consumer -1.097 -1.07
surplus ($ billion)
Change in U.S. total -1.901 -1.946
surplus ($ billion)

TABLE 12

Externality and Gasoline Consumption Effects--3.0 MPG Increase
CAFE Already-Binding Model

 Change from Change from
 MY 1999 No-CAFE
 Equilibrium Equilibrium

%Change in VOC emissions 1.74 2.12
%Change in NOx emissions 1.87 2.23
% Change in CO emissions 1.91 2.26
Change in gasoline consump- 5.240 6.600
tion (billion gallons)
Change in miles driven 26.151 32.792
(billions)
Total externality costs 2.188 2.743
($ billion)
Total costs ($ billion) 4.089 4.689
Average cost of reducing $0.363/$0.780 $0.295/$0.710
gasoline externality without
and with externalities
Marginal cost of reducing $0.695/$1.050
gasoline externality
(inferred) without and with
externalities

TABLE 13

Low- and High-Cost Scenarios, CAFE Binding Model, Chnages from No-CAFE
Equilibrium

Changes in Producer and Consumer Sur- Low-Cost High-Cost
plus ($ billion) Scenario Scenario

Foreign firms 0.084 0.2478
U.S. firms total -0.480 -1.3096
Change in comsumer surplus ($ -0.574 -1.6360
billion)
Change in U.S. total surplus ($ -1.053 -2.946
billion)
Change in gasoline comsumption 6.568 6.638
(billion gallons)
Externalities costs
Change in miles driven 34.098 31.290
Total externality ($ billion) 2.820 2.618
Total cost ($ billion) 3.873 5.564
Average cost of reducing gasoline $0.160/$0.429 $0.444/$0.838
externality without and with
externalities
Marginal cost of reducing gasoline $0.388/$0.827 $1.018/$1.283
externality (inferred) without and
with externalities

TABLE 14

Initial Conditions--Demand Elasticities

 Parameters Used in CAFE Simulation
 Demand Elasticity Table

Class 1 2 3 4 5 6

 1 small car -2.808 0.423 0.063 0.018 0.000 0.036
 2 medium car 0.684 -3.528 1.107 0.027 0.018 0.018
 3 large cars 0.270 1.926 -4.500 0.027 0.216 0.009
 4 sport car 0.549 0.423 0.324 -2.250 0.009 0.090
 5 luxury car 0.045 0.405 1.062 0.009 -1.737 0.000
 6 small truck 0.162 0.099 0.000 0.009 0.000 -2.988
 7 large truck 0.063 0.072 0.018 0.009 0.000 0.234
 8 small SUV 0.216 0.279 0.099 0.027 0.009 0.090
 9 large SUV 0.117 0.243 0.171 0.018 0.018 0.054
10 minivan 0.081 0.171 0.063 0.000 0.009 0.009
11 van 0.027 0.036 0.009 0.009 0.000 0.009

 Parameters Used in CAFE Simulation
 Demand Elasticity Table

Class 7 8 9 10 11

 1 small car 0.027 0.009 0.009 0.009 0.000
 2 medium car 0.018 0.036 0.045 0.054 0.009
 3 large cars 0.054 0.018 0.063 0.054 0.009
 4 sport car 0.198 0.045 0.108 0.018 0.000
 5 luxury car 0.027 0.045 0.189 0.072 0.009
 6 small truck 0.702 0.045 0.054 0.009 0.009
 7 large truck -1.548 0.027 0.090 0.018 0.036
 8 small SUV 0.351 -3.645 0.747 0.108 0.072
 9 large SUV 0.387 0.414 -2.043 0.234 0.108
10 minivan 0.045 0.027 0.135 -2.286 0.180
11 van 0.054 0.036 0.072 0.387 -2.385

(1.) See "National Energy Policy," Report of the National Energy Policy Development Group (May 2001), available online at www.whitehouse.gov/energy, at pp. 4-10.

(2.) See NRC, "Effectiveness and Impact of Corporate Average Fuel Economy (CAFE) Standards," July 2001, online at http://books.nap.edu/html/cafe.

(3.) See Energy Information Administration, online at www.eia.doe.gov/emeu/ipsr/t24.txt.

(4.) In its 2003 rule-making procedure on CAFE standards, NHTSA calculated the relevant externality at a level for below that of the NRC value. NHTSA (2003, 44) estimated a total externality of 8.3 cents per gallon, with 4.8 cents attributed to monopsony power and 3.5 cents to supply disruption. Of course, this is subject to the same critiques made.

(5.) For the extent of such taxes, see www.energy.ca.gov/fuels/gasoline/gas_taxes_by_state.html.

(6.) See Kleit (1990). For a similar model, see Thorpe (1997).

(7.) Categories are taken largely from internal GM documents, but most vehicle types fit easily into particular categories based on size, price, and whether the vehicle is a truck or a car.

(8.) Later I will attempt to account for the likelihood that CAFE standards were already binding. I do this by first running the model "backward," employing the estimated tax as a subsidy and using short-run elasticities of supply. I then take the results as the free-market equilibrium and then run the model "forward" using the long-run elasticity of supply.

(9.) There are some caveats to this, as a review of Table 14 will make apparent. The own demand for medium and large cars is relatively elastic, but this is due to the high cross-elasticity of demand between the two segments. In addition, the own elasticity of demand for vans is very similar to the own demand elasticity for minivans.

(10.) I note that my use of the phrase technology forcing may be slightly different than that generally used in the environmental literature. Here, by technology forcing I refer to manufacturers using technologies that they would otherwise not find profitable, rather than the standards actually inducing the creation of new technologies.

(11.) All of the costs of fuel efficiency used in this section and applied to subsequent sections refer to net costs, that is, the costs of fuel efficiency minus the benefits. The benefits are, of course, the reduced per mile cost of driving. Thus these represent economic rather than engineering costs.

(12.) See www.tc.gc.ca/envaffairs/subgroups1/vehicle% 5Ftechnology%5Fold/study2/Final_report/Final_Report. htm.

(13.) Of course, many consumers will eventually sell their cars in the used car market. But the same reasons why the market for new cars should value fuel economy apply to the market for used cars.

(14.) The cost of fuel economy in Table 3 is not greatly dependent on the choice of interest rate used.

(15.) The emissions measured here are solely from tailpipes. As discussed, there are some questions about whether reduced gasoline consumption would imply reduced refinery pollution.

(16.) This is the average cost calculated as the cost of congestion-related delays and fuel costs, $ 78 billion, divided by aggregate vehicle miles traveled by light duty vehicles. See Lutter (2002).

(17.) The externality costs are a linear function of the costs per mile, so interested readers can choose their own externality cost level for analysis.

(18.) See www.whitehouse.gov/omb/inforeg/ costbenefitreport1998.pdf.

(19.) The resulting changes in MPG because of this negative tax of $1,652 per MPG are -1.05, -1.42, and -0.55 MPG for GM, Ford, and Chrysler cars, and -0.59, -.50, and -0.40 MPG for GM, Ford, and Chrysler trucks.

(20.) Note that in performing a cost-benefit analysis of CAFE standards, the price of gasoline will equal the marginal benefit of consumption. Thus the value of the externality associated with the consumption of gasoline will constitute the net benefit to society from reductions in gasoline consumption.

(21.) I do not subtract from the cost of a gasoline tax the economic impact on accidents and congestion resulting from the decrease in miles driven.

REFERENCES

Bohi, D. R., and M. A. Toman. The Economics of Energy Security. Boston: Kluwer, 1996.

Bordley, R. F. "Estimating Automotive Elasticities from Segment Elasticities and First Choice/Second Choice Data." Review of Economics and Statistics, 75, 1993, 455-62.

Braeutigam, R. R., and R. G. Noll. "The Regulation of Surface Freight Transportation: The Welfare Effects Revisited." Review of Economics and Statistics, 66, 1984, 80-87.

Dahl, C. A., and T. Sterner. "Analyzing Gasoline Demand Elasticities: A Survey." Energy Economics, 13, 1991, 203-10.

Dixit, A. K., and R. S. Pindyck. Investment under Uncertainty. Princeton, NJ: Princeton University Press, 1994.

Edlin, A. S. "Per Mile Premiums for Auto Insurance." National Bureau of Economic Research Working Paper No. W6934, 1999. Cambridge, MA.

Espey, M. "Pollution Control and Energy Conservation: Complements or Antagonists? A Study of Gasoline Taxes and Automobile Fuel Economy Standards." Energy Journal, 18, 1997, 23-38.

Greene, D. L. "Why CAFE Worked." Energy Policy, 1997, available online at www.cta.ornl.gov/Research/teep/ CAFEabs.html.

Greene, D., and J. Hopson. "An Analysis of Alternative Forms of Automotive Fuel Economy Standards for the United States." Transportation Research Board Annual Meeting CD, 2003.

Greene, D. L., J. R. Kahn, and R. C. Gibson. "Fuel Economy Rebound Effect for U.S. Household Vehicles." Energy Journal, 20, 1999, 1-31.

Kleit, A. N. "The Effect of Annual Changes in Automobile Fuel Economy Standards." Journal of Regulatory Economics, 2, 1990, 151-72.

--. "Enforcing Time-Inconsistent Regulations." Economic Inquiry, 30, 1992, 639-48.

Lutter, R. CAFE: The Numbers behind the Story. AEI-Brookings Joint Center for Regulatory Studies, 2002, available online at www.aei-brookings.org/ policy/page.php?id=84.

National Highway Traffic Safety Administration. "Light Truck Average Fuel Economy Standards." 2003. Available online at www.nhtsa.gov/cars/rules/ ruleings/CAFE05-07/index.html.

Nivola, P. S., and R. W. Crandall. The Extra Mile. Washington, DC: Brookings, 1995.

Orazio, A., P. Goldberg, and E. Kyriazidou. "Credit Constraints in the Market for Consumer Durables: Evidence from Micro Data on Car Loans." NBER Working Paper W7694, 2000. Cambridge, MA.

Pindyck, R. S. The Structure of World Energy Demand. Cambridge, MA: MIT Press, 1979.

Roe, B., K. J. Boyle, and M. F. Teisl. "Using Conjoint Analysis to Derive Estimates of Compensating Variation." Journal of Environmental Economics and Management, 31, 1996, 145-59.

Sutherland, R. J. "'No Cost' Efforts to Reduce Carbon Emissions in the U.S.: An Economic Perspective." Energy Journal, 21, 2000, 89-112.

Thorpe, S. G. "Fuel Economy Standards, New Vehicle Sales, and Average Fuel Efficiency." Journal of Regulatory Economics, 11, 1997, 311-26.

Toman, M. A., and J. Shogren. "How Much Climate Change Is Too Much: An Economic Perspective." Climate Change Issues Brief 25, Resources for the Future, 2000.

Train, K. "Discount Rates in Consumers' Energy Related Decisions: A Review of Literature." Energy, 10, 1985, 124-125.

ANDREW N. KLEIT *

* This report was funded by the General Motors Corporation. The views expressed herein are solely those of the author and not those of General Motors or of Pennsylvania State University. I thank General Motors economists Marc Robinson, Tom Walton, and Mike Whinihan and two anonymous referees for helpful comments and data, and graduate students Supawat Rangsuriyawiboon and Tina Zhang for excellent research assistance.

Kleit: Professor of Energy and Environmental Economics, The Pennsylvania State University, 507 Walker Building, The Pennsylvania State University, University Park, PA 16802-5010. Phone 1-814-865-0711, Fax 1-814-865-3663, E-mail [email protected]