Is demand for polluting goods manageable? an econometric study of car ownership and use in Mexico

POLICY RESEARCH WORKING PAPER 1309 Is Demand for Polluting Thismodedisplas immediate adjustment and a Goods Manageable? pnce asdcily for gaoline of -0.8. Demand for gasodine and the subsequent An Econometric Study of Car generadon of polluton, are Ownership and Use in MexicorshpindUseitoiodn Gunnar S. Eskeland Tarhan N. Feyzioglu The World Bank Policy Research Departnent Public Economics Division June 1994  POLICY RESEARCH WORKING PAPER 1309 Summary findings Charging for social marginal costs is efficient regardless correlation in the residuals to model the dynamics of price elasticities, but the importance of getting prices properly. The resulting model is one of almost "right" is grearer the more manageable, or elastic, the immediate adjustment, wirh a short-term price elasticitry demand. In efficient pollurion control programs, options for gasoline close ro the long-term estimate of -0.8. to make cars cleaner are combined optimally with The model displays elasticities that are lower (for demand conservation. The roles plaved by -cleaner cars" income) and higher (for price) than those Pindvck as compared with 'iewer trips" are determined-b hypothesized, and are within the range of elasticities empirical parameters: cheap, clean technologies would found in industrial countries.. imply a grear role for cleaner cars, while high demand Byproducts of the model: The clascicity of car elasticities lead to a greater role for demand reduction. purchases with respect to gasoline prices is positivc. In seminal research, Pindyck found evidence to. Scrappage decisions are affected by income and by car supporr his hypothesis that demand for commodiries and gasoline prices. And these elasticities are nor such as gasoline should have lower price elasricities and significantly different in the richer states. higher income elasticiries in developing than in industrial For policy purposes, these findings do not svpport countries. Eskeland and Feyzioglu estimate a model of "elasticitv pessimism." The use of car senrices is sensitive gasoline demand and car ownership in Mexico, using a to pricing, which suggests that consumers. for some of panel of annual observations by state. KCey features they their demand, have reasonably good alternatives ro car introduce are instrumental variables on differenced data services. Consideration of external cosrs - such as and the treatment of (1) possible dynamics, (2) accidents, congestion, air pollution, and road damage- measurement errors in the data, and (3) unobserved thus involve considerable demand conservation. characteristics in individual states. They use tests of serial This paper -a product of the Public Economics Division, Policy Research Department is part of a larger effort in the department to scudy environmental policy problems in developing countries, emphasizing fiscal pelicy insrruments. The study was funded by the Bank's Research Support Budget under the research project 'Pollution and the Choice of Policy Instruments in Developing Countries" (RPO 67648). Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contacr Carlina Jones, room N10-063, extension 37699 (31 pages).June 1994. The Poliy Reserh Working Paper Series disseminates the findigs of wok in progress to encowage the exchafnge of ideas about development is5e An ob/ective of hesriesivs roger thefindingsourquwikly en if the presentationsarrkss thanfudff polsbed The papers cany the names of the authors and shou bE used and cited accordingly The findings, interpretations, and condusons are the autors own and should no be attributed to the World Bnak its Executive Board of Direcrors, orany of its nenter couarie Produced by the Policy Research Dissemination Center.  Is Demand for Polluting Goods Manageable? An Econometric Study of Car Ownership and Use in Mexico Gunnar S. Eskeland Public Economics Division Policy Researcn Department: World Bank 1818 H Street, NW Washington, DC 20433 and Tarhan N. ]Feyzioglu Department of Economics Georgetown University Washington, DC 20057 We would like to thank colleagues in Me3xco, particularly Jorge D. Balmer at Instituto Nacional de Estadistica Geografia e Informatica, and colleagues at the World Bank for help with data and comments. We wouId also like to thank the partipants of our seminar at the World Bank.  Table of Contents page 1 Introduction I 2. Economic Model 7 3. Dynamics: Short Run and Long Run Elasticities 12 4. Data 14 5- Econometric Issues 15 6e Results 17 7. Summary and Conclusions 24 References 27  I Introduction This econometric study is a part of a:broader research effort on the economics of pollution control policies in developing countries. Our motivation for studying the determinants of demand for cars and gasoline is one of the many possible ones: we want to know the extent to which demand for polluting goods and services is sensitive to prices and income developments. The study and its results, however, should be of more general interest. It may be of interest from the point of view of an applied econometrician -- using some new techniques and data to examine an old problem with real-world data. constraints. It may* also be of interest to those interested in demand for cars and gasoline, either because they may find good ideas on how to study the topic, or because they may find use for our actual results. Our research project on the economics of pollution control policies in developing countries has emphasized two major lines of inquiry that has caused our interest in the demand for polluting goods and services.1 One is that a potential, least cost program can deliver pollution reductions either by making each activity "cleaner" per unit of input or output (illustratively, we may call this cleaner cars and fuels, or technical controls), or by scaling down the level of polluting activities (we may call this fewer polluting trips). Such a least cost program could, theoretically at least. be induced by first best instruments such as tradeable emission permits or emission taxes, based on monitoring of individual emissions. If such programs had been in place, then one could estimate the emission reductions provided at different tax rates, at least in reduced form, and perhaps even recover the relative roles of cleaner trips and fewer trips. Given I Halvorsen and Ruby (1981), and Freeman (1982) present broader treatments of the costs and benefits of air pollution control. Harrison (1975) covers the same field, with an emphasis on vehicular emissions, and the distribution across households of costs and benefits. - 1~~~~~~~~  that such programs are not in place, however, one needs to go the indirect route of estimating the costs at which emission reductions can be provided through trip reductions and technical controls respectively. Such estimate can then be used to estimate both the control costs for potential least cost programs, and the excessive .costs associated with programs that do not combine optimally the various ways by which emission reductions can be provided. Another reason for -inquiring about the demand for polluting goods and services such as cars and gasoline is our belief that costs of monitoring and enforcement often will make the use of first best instruments such as emission taxes difficult or impossible. When that is the case. the policy maker may need to evaluate the various ways by which emission reductions can be provided, in order to stimulate them separately-. For instance, for cars and trips, we may think of fees or sanctions associated with initial and periodic tests of emission factors as stimulating cars and fuels to be cleaner, while gasoline and road taxes, mass transport policies and parking fees are used to manage demand for polluting urban transport.3 Perhaps for many reasons there is a rich body of econometric studies of demand on vehicles and fuels. General studies of demand for energy, and specific fuels among them. bloomed in the years following the first oil price shock in 1973, when 2 In Eskeland and Jimenez (1992), this point is elaborated in their distinction between direct instruments (based on monitoring of individual emissions) and indirect instruments [based on indicators of emissions, such as the characteristics of cars and other machinery as a proxies for "dirtiness". and fuel use or other measures as sroxies for throughput). The case for cleaner cars and fewer trips is examined in detail in Eskeland (1993), and Eskeland (1992). A program optimally combining instruments economizing on trips and making cars and fuels cleaner was found to cost 65 million dollars more per year, if the demand management instrument, a gasoline tax, was excluded, in favor of more aggressive technical controls. Since the 65 million dollar estimate was based in a more conservative price elasticity estimate than found in this study, we may now conclude that the costs of excluding demand management is higher. Han (1992 a and b), and Newbery et al. (1988) discuss charging road users, but to discourage road damage and congestion, rather than pollution. McConnell and Harrington (1992), Hahn (1993), Anderson (1990), Faiz et al. (1990) are exarples of detailed studies of technical control options and costs. 2  importing nations became concerned about their vulnerability to disruptions in imports and/or price increases. The costs to an importing nationL facing exogenous price shock will be higher the lower is the price elasticity of demand for the commodity in question- Among studies focusing on demand for energy, Fuss' 1975 study of energy use in Canadian manufacturing and Pindyck's book "The structure of world energy demand" probably are the most important: Fuss for demonstrating methodological breakthroughs concerning inter-fuel substitution. and Pindyck for a broad inquiry based on data from many countries, including developing countries. Pindyck points out that there are reasons to believe that previous results based on the studies of developed countries should not be generalized to developing countries. It is conjectured that "as incomes rise, additional expenditures are not allocated proportionally to larger homes or to more heat or light in existing homes"; therefore, lower income countries should have higher income elasticities. In the same vain, at low income levels, most energy use is a necessity, whereas at high income levels, energy use "becomes more discretionary, allowing for greater substitution away from energy if prices rise"; therefore, lower income countries should have lower price elasticities- Pindyck compares results he obtairns from a developing country sub sample (Mexico and Brazil) with those from developed countries, and finds the results to be consistent with his expectations of lower price elasticities and higher income elasticities for gasoline: "The estimated price elasticity of demand is -_55 as compared to the estimate of about -1.3 obtained for the developed countries"--the income elasticity is 1.22 as compared to .8 for the developed countries". Comparing his results with those of many others, with lower elasticities, he concludes that (their) "use of data for a single country is more likely to elicit short- or intermediate-run elasticities" (page 233]. 3  Another, not entirely independent development of the 1970s was regulatory changes to enhance environmental quality and fuel efficiency. With relevance for our topic, these developments gave emphasis to the distinction between fuel efficiency, measured for instance by liters consumed per kilometer and vehicle kilometers traveled by the average household. Manski (1983) proposed an elegant model of vehicle scrappage, and Bercovec (1985) estimated a model of vehicle demand by type including such a scrappage model. Using this model, he could estimate the likely effect of vehicle regulations and the associated price increases for new vehicles, on the turnover and properties of the vehicle stock- Broader studies of the behavior of auto ownership and use, are found in, inter alia, Winston et al. (1987), Crandall et al. (1986), Ingram et al. (1975) and Grad et aL (1975). General equilibrium treatments of the effects of energy price increases and environmental regulations, with less emphasis on transportation and a particular fuel, are found in Jorgenson and Wilcoxen (1990), and in Hazilla and Kopp [1990]. There is also a literature of empirical studies based on discrete choice models and micro-data, emphasizing the sensitivity of mode choice for individual trips to, inter-alia, pricing and travel times (see, for instance.- Ben-Akiva and Lerman (1985))3 Results from this literature are not generally comparable to those from aggregate data -- one of the most obvious reasons for this is that the mode- choice models usually assume many variables as given in the outset (residential location, work-place location, car ownership). Due to these and other important differences between the two empirical bodies of literature, one should not be surprised that estimates of such parameters as the elasticity of car use to car operating costs will usually be much lower in these models than in aggregate models. Two recent reviews that highlight findings in empirical models are Oum, Waters and Young (1990) and Krupnick (1992)] Another recent study with both a review of 4.  results and empirical estimates is Sterner (1990). Sterner (1990) surveys close to a hundred different papers with 360 different estimated demand equations, and re- estimates the models *using a larger data base than those used in the studies he summarizes. He points to differences in results that may be seen as discouraging, but concludes that there is consistency in the results and that demand does "adapt to changes in both income and prices". For OECD countries, the sl-jrt run elasticities from the dynamic models "appear to be around -0.2 to -0.3 and 0.35 to 0.55 for price and income respectively". The long run elasticities were around -1.0 to -1.4, and 0.6 to 1.6 for price and income respectively. For OECD countries, the results on price elasticities are consistent with those obtained by Pindyck, but the wide range for income elasticities cast a doubt on the claim that they should be systematically higher for developing countries. Of special interest is,- of course, Berndt and Botero (1985), who obtain elasticities for Mexico that are much closer to those reported for developed countries in Sterner.4 They present estimates from Mexico of a model of vehicle stock adjustments and gasoline demand, thus very similar to the objective of our study. They utilize a pooled cross-section time series data set and use the dynamic gasoline demand model discussed in Drollas (1984). For the short run, they find -0.23 for price elasticity and 0.31 for income elasticity. Long run price and income elasticities they find are -0.96 and 1.25 respectively. There are several key issues that Berndt and Botero do not address. First, they use pooled cross-section time series data, aggregated to 14 regions; however, they do not consider possible differences between these regions, like geography and 4 The higher income elasticity for residential electricity demand in Mexico is explored by Berndt and Samaniego (1984). They point out -that once the increase in demand for electricity do to increase in accessibility is taken into account, the remaining demand behaves similar to electricity demand in industrialized countries. Our study goes in the same direction for gasoline, distinguishing demand given the number of cars from the determinants of car ownership- 5  infrastructure. Omission for possible differences between regions may lead to biased and inconsistent results. Second, they do not test whether the dynamics are adequately taken into account. In consequence, there could still be important dynamics left as residuals in their model. Third, they do not consider the effect of gasoline prices on new car sales. This results in the omission of the indirect effect of gasoline prices on gasoline consumption, thus ignores an empirical effect of interest in policy. We address these and other issues that arise due to the nature of the data. We utilize a pooled cross-section time series data set with annual observations from the 31 states and the federal district in the Mexican Federation. We solve unobservability problem af the state specific effects by differencing the data. We explicitly take into account the possible dynamics in behaviors by incorporating it into the model, and by testing the residuals. We also deal with measurement error problems, specifically in state-wide GDP. by using instrumental variable estimation. Section II introduces the economic model, and III presents dynamics and the relationship between short and long run elasticities. IV and V discuss data and econometric issues, respectively, and VI presents empirical findings. Summary and conclusions are found in a brief section VII: 6  II. Economic Model We start with the identity that the total gasoline consumption in Mexico is equal to the average consumption times the total number of vehicles, for each state and time period: : -- it it (1) where, G is the total gasoline consumption, C., is the total consumption divided by it IL the number of vehicles, and S. is the stock of vehicles (number or cars), at tine t, it for state i. There are two reasons for using this form. First, this decomposition lets us analyze the role of the car stock and the average utilization rate separately. We have in mind a model in which there is a fixed cost of having a car available, associated with car prices, and a variable utilization cost, associated with gasoline prices. In such a model, gasoline consumption as. well as car ownership will depend on car prices, gasoline prices and income.5- Second, this form lets us calculate thie price and income elasticities easily. Elementary calculations show that the gasoline price elasticity of consumption is the sum of price elasticities for per car consumption and total number of cars: 71 t 71c + 71s (2) where, t is the gasoline price elasticity of total consumption, 71 is the gasoline price elasticity of gas consumption per car, and iis is the gasoline price elasticity of car stock. Next, we turn to modeling each component of this identity. First, we model gasoline consumption per vehicle, for a given number of vehicles. We assume a representative consumer with a utility function separable in services rendered by a 5 Throughout, we shall work with three market goods and their prices; gasoline, cars and other goods and services. We normalize each price by the price of other goods and services, thus reducing the analysis to two prices only. 7  car and other goods and services. We assume the services from the car are proportional to gasoline usage. In addition to prices and income, there may be differences between states, like geographies and infrastructure that affect gasoline consumption. These additional effects are not observable to us, but we can summarize them in a state specific variable, a., that is constant through out years, but varies across states. We also incorporate habit persistence by considering the lagged values of the dependent variable, and write the consumption function in the following form: C. f fC GASPRVY. ;Mi.e) . (3) it it-I' it where, GASPRt is the gasoline price. Yi is income, C. is the vector of past t ~~ ~~~it it-1 consumption rates, a. is a scalar that allows for the state specific characteristics, I~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ e is the vector of parameters dictated by the functional form, and f C ) is. the function implied by the first order conditions.6 Second, we model the car stock. We assume that there is an optimal car stock level for each state. Consumers in each state would calculate how much car services they want, given the prices and their incomes. As relevant prices for this choice, they consider car prices and gasoline prices. The reason why we include gasoline 6 Note that we assume that car prices do not effect gasoline consumption per car. To examine this assumption in detail, assume that consiumption per car is average fuel- efficiency of the car stock times the average miles driven: C = et(IlCARPR,GASPR) where, e is the overall stock fuel-efficiency rate and t is Smiles driven. Then we can decompose the change in utilization rate into two, one due to change in efficiency, the other, due to change in number of trips made: B-/CBCARPR = C(e/8CARPR)'t + (at/aCARPR)le Change in average fuel-efficiency due to an increase in car prices should be zero or negative since less old cars will be scrapped and fewer new purchased. A nice piece of evidence to this effect is found in Kahn (1986). On the other hand, car prices will reduce or have no effect on number of cars in the stock, and this in turn will, if anything, increase the trips taken by each car. Our assumption that gasoline consumption per car is independent of car prices thus amounts to assuming the sum of these two effects equal zero. - 8 X :  prices is that we believe consumers do take into account the marginal cost of running a car in their purchasing decisions. They may calculate the total discounted cost of gasoline consumption into the car price. This implies that as gasoline prices increase, we should expect a decrease in the car stock. However, if new cars are more fuel efficient, then new car sales might increase as gasoline prices increase. Considering both arguments, we do not know a priori which direction the gas prices may affect car stock. We can summarize these in the following optimal stock level equation:7 - Sit = s(CARPRt,GASPRt (it4]i;4 where S is the optimal car stock level, Y. is the income level, CARPRt is a price t it index for new cars, ai is a scalar for each state, representing state specific characteristics, C is a vector of unknown parameters. If there are adjustment costs in the car stock, the car stock may deviate from the optimal stock. To allow for this possibility, we decompose the current actual car stock into the depreciated car stock that remained from the previous year and the new car purchases: sit =(l-)Sit + Iit(5) where, St is the stock of cars, I is the new car purchases, and 6 is the t ~~~~~~~~~t depreciation factor. For depreciation, we shall consider two alternatives. The first one is the constant depreciation rate that does not change across the states or through out the years. While this is a commonly used assumption, we believe it should be tested. It can be argued that the higher the new car prices, the higher the value of the used 7 The optimal stock decision depends *on the expected value of the service flow, and the costs of holding the car for one period (where a "period' should be long enough that the transaction costs are not overwhelming]. The relationship between holding costs and car prices may not be one to one, so our estimate of aSt/8CARPRt is a reduced form estimate. 9  cars would be. and the less the number of cars that would qualify to be scrapped. An elegant model is given by Manski (1980). and applied successfully to the US market by Berkovec (1985). In addition. as gas prices go up, if older cars have lower fuel efficiency, scrappage should increase. Similar reasoning goes for an income increase: the higher incomes are, the less they would be willing to use and repair old cars. We may test a model allowing the depreciation rate to depend on these factors: &3 d(CARPR .GASPR Y. I (6) it t 1: it where d(-) stands for the functional form for the depreciation rate. The second component in equation (5), new car purchases, should be a function of the optimal car stock. Higher incomes, or lower car prices would increase the optimal car stock which in turn would increase new car sales. But there is also a mechanism through which new car-purchases depend on the car stock from the previous year. If a partial adjustment model is assumed, this dependence would be reflected in the adjustment factor, which describes the extent to which a difference between the optimal stock and the previous year stock is closed within a year.8 A linear partial adjustment model implies a negative, close to umity relationship between the new car purchases and the previous year stock, simply because the new car purchases are equal to a given fraction the optimal stock, less whatever remains from the previous year. However, Pindyck (1979) and Berndt and Botero (1985) have found evidence against these models; therefore we are not going to 8 The simple stock adjustment model assumes that, each period, consumers buy new vehicles to close some fraction of the gap - between the optimal stock and the depreciated stock from previous year: S -(1 -SIS =Y(S -(1-S)S 1 0 c 5 1. t t-1 t t- It= tS- (l1-)St_) where I is the new car purchases, S is the optimal stock, 5 is the depreciation t t factor, and y is the adjustment factor. 10  restrict the functional form to strict linear partial adjustment model. A less restrictive stock adjustment model is allowed by the following formulation: I. = 1( S .;r) (7) it it' it-lV'i where a is the parameter vector, and i(-) indicates the functional form. Several caveats are in order. First, we should emphasize that we assume that gasoline prices and car prices are exogenously determined. Apart from believing these assumptions are plausible, testing them would require more supply side information, and is outside the scope of this study. Second, we assume that gas consumption per car is independernt of the number of cars- It can be argued that in metropolitan areas, if road space does not increase commensurately with the number of cars, traffic congestion may increase. This in turn would i