Optimal gasoline tax in developing, oil-producing countries:The case of Mexico$

Arturo Antón-Sarabia n, Fausto Hernández-TrilloCentro de Investigación y Docencia Económicas (CIDE), Carretera México Toluca 3655, México DF 01210, Mexico

H I G H L I G H T S

� We estimate the optimal gasoline tax for a typical less-developed, oil-producing country like Mexico.� The relevance of the estimation relies on the differences between less-developed and industrial countries.� The optimal gasoline tax is $1.90 per gallon at 2011 prices.� Distance-related pollution damages, accident costs and gas subsidies account for the major differences.� Gasoline tax incidence may be progressive in less developed countries.

a r t i c l e i n f o

Article history:Received 18 July 2013Received in revised form26 November 2013Accepted 28 November 2013Available online 21 December 2013

Keywords:Gasoline taxGasoline subsidyTax incidence

a b s t r a c t

This paper uses the methodology of Parry and Small (2005) to estimate the optimal gasoline tax for aless-developed oil-producing country. The relevance of the estimation relies on the differences betweenless-developed countries (LDCs) and industrial countries. We argue that lawless roads, general subsidieson gasoline, poor mass transportation systems, older vehicle fleets and unregulated city growth make thetax rates in LDCs differ substantially from the rates in the developed world. We find that the optimalgasoline tax is $1.90 per gallon at 2011 prices and show that the estimate differences are in line with thefactors hypothesized. In contrast to the existing literature on industrial countries, we show that therelative gasoline tax incidence may be progressive in Mexico and, more generally, in LDCs.

& 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Optimal environmental taxation, particularly gasoline taxation,has received a great deal of attention in the academic literature(Sandmo, 1975; Bovenberg and de Mooij, 1994; Bovenberg and vander Ploeg, 1994; Bovenberg and Goulder, 1996; Parry and Small,2005; West and Williams, 2007; Lin and Prince, 2009, amongothers). With a few exceptions (Parry and Timilsina, 2008; Parryand Strand, 2012), empirical estimates of the optimal gas tax rateare generally calculated for developed economies. However, inless-developed countries (LDCs) that produce oil, this naturalresource is usually extracted by a state company, which makesgovernments believe that the oil yields should benefit people

through subsidized, cheap gasoline, instead of believing that theyshould levy an optimal tax on this good.1

There are many factors that may lead to differences in theoptimal tax rate between developing and advanced economies.First, the growth of cities in LDCs is often anarchic, with deficient,narrow, and pot-holed road systems; critical congestion areas;deficient intermodal passenger transport; and urban sprawl thatprovokes higher congestion costs (Gakenheimer, 1999).2 Second,poor regulation and a lack of traffic rule enforcement increase theprobability of accidents.3 Finally, LDCs usually have a much older

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/enpol

Energy Policy

0301-4215/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.enpol.2013.11.058

☆The authors thank the Economic Commission for Latin America and theCaribbean (ECLAC) for financial support, three anonymous referees, Luis MiguelGalindo and David Heres for useful comments, and Alejandra Pérez for excellentresearch assistance.

n Corresponding author. Tel.: þ52 55 5727 9800�2743.E-mail address: arturo.anton@cide.edu (A. Antón-Sarabia).

1 The International Monetary Fund (IMF, 2013) reports that pre-tax petroleumsubsidies are systematically higher in oil-exporting countries. Among thesecountries, Middle Eastern and North African countries allocate approximately4.5% of GDP to petroleum subsidies. This rate is approximately 1.2% in sub-Saharan Africa and Latin America.

2 Based on a study by the International Association of Public Transport, Parryand Timilsina (2008) report that 10 of 12 of the largest megacities with the lowestaverage travel speed are in developing countries. See also “Lawless Roads: RoadSafety in Mexico,” The Economist, Oct. 8, 2011.

3 According to the World Health Organization (2013), middle-income countrieshave the highest annual road traffic fatality rates (20.1 per 100,000 population),

Energy Policy 67 (2014) 564–571

motor vehicle fleet than advanced countries, which increasespollution (Harrington and McConnell, 2003). All of these featurescall for the calculation of an appropriate gasoline tax for LDCs.

The objective of this paper is to estimate the optimal gasolinetax for a representative middle-income country. For that purpose,we apply the methodology of Parry and Small (2005) to Mexico, aprominent oil-producing LDC that heavily subsidizes gasolineconsumption. The advantage of this method is the decompositionof the second-best optimal fuel tax into several components,including those related to congestion, accidents, and air pollution.As previously mentioned, these negative externalities may in factbe more severe for the economies of LDCs than for developedeconomies.

Our results suggest an optimal gasoline tax of $1.90/gallon at2011 prices. The (adjusted) Pigouvian tax is the largest portion ofthe tax, amounting to $1.62/gallon. The accident componentexplains approximately one-third of the Pigouvian tax, followedby distance-related pollution damage and congestion externalities.The Ramsey tax component, arising from a relatively inelastic fueldemand, contributes another $0.28/gallon.

The optimal gas tax in Mexico is larger than the estimatereported by PS (2005) for the US (even after updating their resultsat 2011 prices), and that of Lin and Prince (2009) for California, butlower than that estimated by Parry and Strand (2012) for Chile. Inparticular, PS (2005) report an optimal tax rate of $1.01/gallon forthe US at 2000 prices. This estimate increases to $1.43/gallon at2011 prices. Lin and Prince (2009) obtain a rate of $1.37/gallon forCalifornia at 2006 prices. Finally, Parry and Strand (2012) calculatea corrective fuel tax for Chile of $2.35 per gallon at 2006 prices.

To understand what accounts for the differences with respectto PS (2005), we change each one of the parameters that isdifferent in Mexico than in the US, one at a time. We find thatdistance-related pollution damage and accident costs explain themajority of the differences. The presence of fuel subsidies (typicalof oil-producing countries) explains approximately 20% of thedifferences. Perhaps surprisingly, the lower fuel efficiency attrib-uted to an older vehicle fleet does not explain a significantproportion of the differences in tax estimates.

Using the optimal gas tax estimate, we address the effects ofsuch a tax across income deciles in Mexico. Contrary to theconventional wisdom, we find that the fuel tax is progressive.The intuition is simple: only 9% of the poorest households demandfuel because the majority of these households (87%) do not own acar. Conversely, 85% of the wealthiest households demand fuelbecause 91% own at least one car.4

This paper is structured as follows. Section 2 briefly describesthe fuel pricing policy in Mexico. Section 3 outlines the model, andSection 4 presents the results. Section 5 compares the results to

those obtained for advanced economies and includes a sensitivityanalysis. Section 6 provides concluding remarks.

2. How is the gasoline price set in Mexico?

Mexico is an oil-producing country but currently imports abouthalf of the gasoline demanded in the country, because of capacityconstraints. Because of an inappropriate gasoline pricing policy (apre-determined crawling gas price fixed by the government at thebeginning of each year), a subsidy emerges when the internationalpetroleum price is above its pre-determined level, as has been thecase most of the time since 2006 (see Fig. 1).5 This subvention,which may produce a final price below its market price, isfrequently justified on political, i.e., populist, grounds.

On average, Mexico registers the second-lowest excise tax rate,after the USA, among a sample of representative OECD countries forthe period 2001–2011 (see Fig. 2).6 For the period 2007–2011, the taxis in fact negative (a subsidy). This subsidy has cost the governmentan average of 1.2% of GDP over the period 2007–2011, an amountequivalent to the expenditures on poverty alleviation and publichealth care programs in the country. These large fiscal subsidiesmerit careful re-evaluation, especially in this present time of fiscal

5

10

15

20

25

30

35

40

45

50

Jan/2

000

Jan/2

001

Jan/2

002

Jan/2

003

Jan/2

004

Jan/2

005

Jan/2

006

Jan/2

007

Jan/2

008

Jan/2

009

Jan/2

010

Jan/2

011

Jan/2

012

Jan/2

013

MexicoUS

Fig. 1. Gasoline price, 2000–2013 (Mexican pesos per gallon).Sources: US Energy Information Administration and Mexico's Energy InformationSystem. The US price is the weekly Gulf Coast regular conventional retail price.The Mexican price is the regular conventional retail price.

Fig. 2. Excise tax rate (pre-tax fuel price), average 2001–2011.Source: Author calculations from OECD-IEA, Energy Price and Taxes, QuarterlyStatistics (2012).

(footnote continued)whereas the rate in high-income countries is the lowest (8.7 per 100,000). Inaddition, 80% of road traffic deaths occur in middle-income countries, which have72% of the world population but only 52% of the world's registered vehicles. In fact,nearly 70% of road deaths occur in 13 countries, 12 of which are developingcountries (including Mexico).

4 The source is Mexico's Household Income and Expenditure Survey (2010).Unfortunately, the survey does not provide further information that would help toexplain the gap between fuel consumption and car ownership across households.The survey only registers whether the household owns a vehicle and monthly totalexpenditures on fuel; it does not provide information on vehicle use or drivingpatterns. This gap might be due to a combination of statistical errors and to peopleeither deciding not to use a car (car pooling) or not being able to use a car becauseit is not working. It is common, especially for people in lower deciles, to own oldbroken-down cars with the expectation of putting them back into service in thefuture.

5 Since 2010, the Mexican government has increased the fuel price by $0.016/gallon every month, which corresponds to an increase of $0.19/gallon per year.

6 US data are not included in Fig. 2 because the OECD-IEA, Energy Price andTaxes, Quarterly Statistics (2012) study does not report excise taxes for this country.Only total gasoline taxes (i.e., excise plus sales taxes) in the US of $0.11/l (theaverage for the period 2001–2011) are reported.

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571 565

stress. In particular, the current pre-determined crawling pricesystem requires major revision, including the possibility of introdu-cing a gas tax estimated in an optimal manner.

3. The model

In this sectionwe briefly describe the PS model (2005) and presentan equation for the optimal gasoline tax. For further details, the readeris referred to Appendix. In addition to PS (2005), we derive ananalytical expression for the effect of the optimal gasoline tax onlabor and output. This analysis may be relevant, as policymakers areusually concerned about the effects of taxation on economic activity.

Consider a static, closed economy with a representative agent. Theutility function depends on a series of variables expressed in per capitaterms. Some of these variables are under the agent's control, but someothers are not. The agent chooses consumption (C), vehicle-milestraveled (VMT), time spent driving (T), leisure (N), fuel consumption (F)and themonetary costs of driving as a function of the vehicle price andattributes (H). On the other hand, variables such as governmentspending (G), the amount of pollution (P), and severity-adjusted trafficaccidents (A) are external to the agent. In particular, pollution andtraffic accidents generate disutility to the agent. The time endowmentmay be allocated to labor L, leisure and driving activities.

The agent derives income from labor at the wage rate w.In addition, he/she pays a tax rate tL on labor income and a taxtF on gasoline consumption (or alternatively receives a subsidy iftF is negative). These taxes finance the exogenous governmentspending G. Let qF denote the producer price of gasoline. Hence,the total price of gasoline pF is simply qFþtF. Finally, goods areproduced by perfectly competitive firms. The single input is laborand the technology exhibits constant returns to scale.

In such a context, the optimal gasoline tax tnF is obtained bymaximizing the agent's utility with respect to the gasoline tax tF ,once changes in the labor tax, labor supply, fuel consumption,VMT, and disutility from external costs are taken into account.7 Asdiscussed in PS (2005), the optimal tax may be decomposed intothe following three terms: an adjusted Pigouvian tax, a Ramseytax, and a congestion feedback. The full expression is as follows:

tnF ¼MECF

1þMEBLþð1�ηMIÞεcLL

ηFF

tLðqFþtF Þ1�tL

þEC εLL�ð1�ηMIÞεcLL� �βM

FtL

1�tL;

ð1Þwhere

The adjusted Pigouvian tax¼ MECF

1þMEBL; ð1aÞ

The Ramsey tax¼ ð1�ηMIÞεcLLηFF

UtLðqFþtF Þ1�tL

; ð1bÞ

The congestion feedback¼ EC εLL�ð1�ηMIÞεcLL� �βM

FU

tL1�tL

ð1cÞ

MECF � EPF þðECþEAþEPM ÞðβM=FÞ; ð1dÞ

β� ηMF

ηFF; and ð1eÞ

MEBL ��tL ∂L∂tLLþtL ∂L∂tL

¼tL

1� tLεLL

1� tL1� tL

εLLU ð1fÞ

In the expressions above, ηMI is the elasticity of demand for VMTwith respect to disposable income, ηFF is the own-price elasticityof demand for fuel, ηMF is the elasticity of VMT with respect to the

consumer fuel price, and εLL and εcLL are the uncompensated andcompensated labor supply elasticities, respectively. The termsEPF ; EC ; EA and EPM represent the marginal damage from carbonemissions and the marginal congestion, accident, and distance-related pollution costs, respectively.

It is also of interest to estimate the effect of the optimal gas taxon labor, an analysis not provided in PS (2005). Given the lineartechnology assumption, this procedure amounts to simultaneouslyestimating the effect on output. As detailed in Appendix, such anexpression may be written in terms of the optimal tax tnF as follows:

1LdLdtF

¼ 1tLL

� �FηFFpF

� �MEBLðtnF�tF ÞþðtnF�MECF Þ� �

U ð2Þ

To estimate the effects of the optimal tax on fuel consumptionand fuel economy, we follow PS (2005) and consider the followingempirical function for fuel economy M=F:

MF¼ M0

F0

!qFþtFqFþt0F

!�ηMFF

; ð3Þ

where M0 ¼ ðqFþt0F Þ�ηMF and F0 ¼ ðqFþt0F Þ�ηFF denote the initialVMT and fuel consumption, respectively, and t0F is the initialtax/subsidy on gasoline. In expression (3), ηMFF � ηMF�ηFF is theelasticity of fuel economy with respect to the price of fuel.The after-tax fuel consumption F is calculated as F ¼ ðqFþtF Þ�ηFF .

4. Calibration and results

4.1. Parameter values

Despite the lack of information typical of an LDC, it is possibleto construct good proxies for parameters for Mexico. A detaileddiscussion of the way in which each parameter is calculated maybe found in the extended version of this paper (Anton-Sarabia andHernandez-Trillo, 2013). For ease of reference, central values arepresented in Table 1 at 2011 prices; a sensitivity analysis ispresented in the next section. Parameters for which there aresmall differences or for which there are no available estimates aresimply taken from PS (2005).8 With respect to the remainingparameters, it is worth noting the salient differences between thecentral values for the US and UK (countries considered by PS) andCalifornia (the US state studied by Lin and Prince, 2009).

First, the vehicle fleet in Mexico is 16.5 years old on average(Melgar Consultants of Mexico, 2011), well above the average agein industrialized countries (e.g., approximately 11 years in theUnited States as of 2013).9 The older vehicle fleet negatively affects

7 A detailed derivation of the optimal tax is provided in Parry and Small (2004).

8 It is worth mentioning the parameters related to labor supply elasticity inMexico. Cragg and Epelbaum (1996) report values for the uncompensated laborsupply elasticity between 0.04 and 0.24, depending on the worker's educationallevel. Arceo and Campos-Vazquez (2010) report an elasticity value of 0.21 on thebasis of household data. Campos-Vazquez (2010) report that (uncompensated)labor supply elasticity is between 0.5 and 1 for formal workers and near zero forinformal workers; as a reference, formal workers account for nearly 40% ofoccupied workers in the country (Levy, 2008). Overall, the estimates from Craggand Epelbaum (1996) and Arceo and Campos-Vazquez (2010) are similar to thoseobtained for developed countries (see, for example, Blundell and MacCurdy, 1999)and the value used by PS (2005). On the other hand, there are no availableestimates for the compensated labor supply elasticity in Mexico. However, such anumber may be constructed using information from the uncompensated laborsupply elasticity and an estimate of the income elasticity of labor supply (see Parryand Small, 2004). Gong and Soest (2002) report an income elasticity of �0.17 forwomen in Mexico City. Similarly, Camaal (2007) reports income elasticitiesbetween �0.07 and �0.11 for men and women, based on Mexican data at thenational level. Interestingly, these values are similar to those reported for devel-oped countries (Blundell and MacCurdy, 1999) and also to the implicit value of�0.15 used by PS (2005).

9 The average age of all cars and trucks in the United States stood at 11.4 years in2013, up from 11.2 years in 2012. A decade ago, the average age was 9.7 years,

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571566

average fuel efficiency, which is set at 18.2 miles/gallon, lowerthan in industrialized countries. Distance-related pollutiondamage is set at 5.5 cents/mile, based on the estimates of Smalland Kazimi (1995) and Parry and Timilsina (2008) for Mexico City.Notably, this cost is more than twice the estimate from PS, evenafter updating to 2011 prices, but lower than the estimate of6 cents/mile at 2006 dollars for Santiago de Chile (Parry andStrand, 2012).10 The cost of fuel-related pollution damage inMexico is taken from Johnson et al. (2009). Again, the estimateof 18.2 cents/gallon is substantially higher than those in thereference studies.11

External congestion costs are calculated using the method ofParry and Timilsina (2008). According to this method, the costof congestion is inversely related to the average car speed.The estimate of 5 cents/mile is slightly larger than those reportedin PS at 2011 prices. There are a number of possible reasons forthis higher estimate. Most automobiles in Mexico circulate inmajor cities.12 In addition, city roads are typically in poor shape

and full of speed bumps. Intermodal passenger transport in citiesis deficient, which, among other factors, increases the dependenceon vehicle use.13 These elements contribute to a lower averagespeed and thus to a higher congestion cost.

External accident costs are estimated at 6.4 cents/mile. Thisfigure, which is also large relative to those reported in thereference studies mentioned above, may be explained by a numberof factors, including the following: the ease of obtaining a driver'slicense without even taking a driving test, poor regulation andweak enforcement of traffic laws, and a lack of traffic road signs.14

Medina (2012) estimates that accident costs account for 1.3% ofGDP. This number is used to calculate the accident costs in Table 1.

The government expenditure share αG � G=L and the gasolineproduction share αF � qFF=L are taken from national accounts andofficial data sources. The gas price of $2.52/gallon is the estimatedaverage for the producer's price in Mexico for the period 2003–2011.Finally, the initial tax on gasoline is negative (i.e., a subsidy of 18 cents/gallon) because this is the average tax for the period 2003–2011. Aspreviously mentioned, such subsidies are not unusual in oil-producingLDCs such as Mexico. These four parameter values yield a labor tax of22% (see Eq. (A8) in Appendix). This number seems reasonable, as thestatutory tax rate on labor income for salaried workers in Mexico isbetween 1.9 and 30%, depending on the level of pre-tax earnings.

4.2. The optimal gas tax

According to the model, the optimal gasoline tax in Mexico is$1.90/gallon. This tax can be decomposed into an adjusted Pigou-vian tax of $1.62 and a Ramsey tax component of $0.28 (seeTable 2).15 The accident cost component is the largest componentof the Pigouvian externalities, explaining approximately 29% of theoptimal tax. Distance-related pollution and congestion costs con-tribute an additional 25 and 22.6% of the optimal tax, respectively.

The effects of the optimal gasoline tax on some variables ofinterest are listed in Table 3. Relative to the benchmark scenario,fuel consumption decreases by nearly 30% and fuel economy rises

Table 1Parameter values.

Parameter values taken from Parry and Small Value

VMT portion of gas price elasticity: β 0.4Uncompensated labor supply elasticity: εLL 0.2Compensated labor supply elasticity: εcLL 0.35VMT expenditure elasticity: ηMI 0.6Gasoline price elasticity: ηFF 0.55

Parameters values obtained from calibration for Mexico Value

Initial fuel efficiency: M0=F0 (miles/gallon) 18.2

Pollution damage, distance-related: EPM (cents/mile) 5.5

Pollution damage, fuel-related: EPF (cents/gallon) 18.2

External congestion costs: EC (cents/mile) 5.0

External accident costs: EA (cents/mile) 6.4

Government expenditure/GDP: αG 0.2Gasoline production share: αF 0.024Producer price of gasoline: qF (cents/gallon) 252Initial tax rate on gasoline: t0F (cents/gallon) (-18)Labor tax: tL (implied by parameter values) 0.22

Note: all elasticities are defined as positive numbers.

Table 2Components of the optimal gasoline tax in Mexicocents per gallon in 2011 US dollars.

Adjusted Pigouvian tax 162Pollution, fuel-related 17Pollution, distance-related 47Congestion 43Accidents 55

Ramsey tax 28Optimal gasoline tax rate ðtnF Þ 190

(footnote continued)according to the research firm Polk (2013). Data for other developed countries can befound in Harrington and McConnell (2003). For these other countries, the averageage of the car fleet is well below the average of 16.5 years reported for Mexico.

10 For their study on Mexico City, Parry and Timilsina (2008) take the estimateof 4 cents per mile for Los Angeles from Small and Kazimi (1995). This estimate wasconstructed as follows. To account for the lower value of statistical life (and thusthe lower value of health risks) in Mexico City, Parry and Timilsina first divided thefigure by 4. Given that per capita income in Mexico City is about one-fourth that inthe United States, this number implies an income elasticity of the value of life closeto 1. This number is slightly higher than the income elasticity value of 0.8 suggestedin a recent OECD (2011) study. On the other hand, Parry and Timilsina argue thatthe younger population of Mexico City (compared to the US population) may bemore willing to pay to avoid mortality risks. To allow for (sensitivity-adjusted)population exposure to pollution, the authors scale up the damage estimate by 4.Thus, they end up with an estimate of 4 cents per mile in 2000 dollars. Thisestimate is equivalent to approximately 5.2 cents per mile in 2011 dollars and isrounded up to 5.5 cents per mile in the paper. The next section provides asensitivity analysis of this and other relevant parameters.

11 As reported by Parry and Timilsina (2008), gasoline combustion producesabout 0.009 t of carbon dioxide per gallon. Accordingly, a pollution damage of $20per ton of carbon dioxide yields a pollution cost of 18 cents/gallon, which is similarto the number reported in Table 1. The estimate of $20 per ton of carbon is wellwithin the values typically reported in the literature (see Parry and Small, 2004 fora detailed discussion on this issue).

12 For example, one-fifth of all the cars in the country circulate in Mexico City'smetropolitan area, where the average speed is 10.6 miles/h. For the estimate ofexternal congestion costs, an average travel speed of 18.6 miles/h is assumed, after

(footnote continued)taking into account the average speed of other major cities using official sources. Inaddition, an average travel speed of 31 miles/h is assumed under no congestion.The value of travel time used is 0.82 dollars/h, after taking into account thesuggestion by Small (1992) of dividing the average wage rate per hour by two.

13 Authors such as Iracheta (2006), Litman (2012) and Medina and Veloz (2013)argue that public transportation in Mexico is anarchically planned, lacks minimumsafety standards, operates in a lawless environment and does not respond to urbanharmony.

14 Perhaps unsurprisingly, the World Health Organization (2013) reports thatMexico is among the 13 countries in the world with the highest number of roadtraffic deaths (approximately 16,700 in 2010). If the estimated road traffic deathrate (per 100,000 people) is used instead, Mexico drops to number 96 (of 181countries) in the World Health Organization data. Nevertheless, the traffic deathrate in Mexico is still twice the average in OECD countries. A better measure ofexternal accident costs is the fatality rate per mile driven. Unfortunately, to the bestof our knowledge, no such comparable data are available at the international level.

15 The congestion feedback effect in Table 2 is ignored, given that it is verysmall in quantitative terms (1 cent/gallon).

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571 567

by 23%. Because government spending G is held constant, the extraresources from gasoline revenue finance a labor tax reduction of9%, from 22 to 20%. Finally, labor (and thus output) remainspractically unchanged. This result is explained by the low shareof gasoline production in national output (which is proportional tothe ratio F=L in Eq. (2)).

Table 4 presents the welfare gain for different levels of theoptimal tax rate tnF . The optimal tax of $1.90/gallon would producea welfare gain of 15.1% of the initial pre-tax fuel expenditure.Interestingly, eliminating the current subsidy on gasoline explainsslightly more than one-fifth of the total welfare gain. Likewise, atax of only 0.25 times the optimal gasoline tax (48 cents/gallon)would cover 63% of the total welfare gain.

4.3. Is a gasoline tax regressive?

A typical argument against the implementation of a gasoline tax isthat such a tax is regressive, although it is less regressive if the analysisis expenditure-based rather than income-based (see, for example,KPMG Peat, 1990; Poterba, 1991; Walls and Hanson, 1999).16 Based onthis concern, Fig. 3 presents the average (“current”) fuel subsidy of 18cents/gallon across deciles in Mexico (numbers are converted toMexican pesos). The figure also includes the “implicit” subsidy arisingfrom the absence of the optimal gasoline tax estimated above.

Interestingly, both the current and implicit subsidies are muchhigher for the higher-income deciles. The explanation for thisfinding is relatively simple: only 9% of the lowest decile demandgasoline, whereas 85% of the highest decile do so. Thus, thegasoline tax is in fact progressive in absolute terms.

Fig. 4 presents the relative incidence across deciles of theoptimal gasoline tax, where gasoline expenditure is measuredrelative to total household expenditures, as suggested by Poterba(1991). The light bars indicate the incidence of such a tax on carowners. Clearly, the tax is regressive in relative terms. However,these estimates do not take into account that only 9% of thepoorest households demand gasoline because the majority of

these households do not own a car. Conversely, 85% of thewealthiest households demand gasoline. When such an adjust-ment is made, the optimal tax is in fact progressive, except for theupper decile (the dark bars). Thus, contrary to conventionalwisdom and the findings reported for industrial economies, thegasoline tax does not necessarily affect the poor the most inMexico, even in relative terms.

5. Explaining the differences in estimates and sensitivityanalysis

It is also useful to compare our findings with those obtained byPS (2005). For this purpose, we initially set all parameters in the

Table 3Impact of optimal fuel tax on variables of interest.

Variable Percentage change(%, relative to the benchmark)

Fuel consumption �29.6Fuel economy 23.4Labor tax �9.1Labor supply 0.0

Table 4Welfare effects of gasoline tax rates (relative to current rate, expressed as percent ofinitial pre-tax fuel expenditures).

Gasoline tax rate Rate(cents/gallon)

Welfare change(% of pre-tax expenditure)

0 0 3.40:25tnF 48 9.50:50tnF 95 13.00:75tnF 143 14.6Optimal rate ðtnF Þ 190 15.11:25tnF 239 14.71:50tnF 286 13.7

Fig. 3. Gasoline subsidy across income deciles in Mexico (quarterly Mexican pesosper household).Source: Author estimate based on the Household Income and Expenditure Survey,INEGI, Mexico (2010).

Fig. 4. Relative gasoline tax incidence (% with respect to household total expenditure).Source: Author estimate based on Household Income and Expenditure Survey, INEGI,Mexico (2010).

Table 5Differences between the Parry and Small (2005) estimate and the optimal gasolinetax in Mexico.

Parameter Change in the optimal gasolinetax rate relative to the Parry andSmall estimate (%, ceteris paribus)

Higher pollution damages, distance-related 38.6Higher external accident costs 37.6Higher producer price of gasoline 20.8Lower initial tax rate on gasoline 18.8Higher external congestion costs 15.8Higher pollution damages, fuel related 13.9Higher gasoline production share �0.0Lower fuel efficiency �0.1Lower government spending/GDP �0.1

16 West (2004) finds that a greater price responsiveness among low-incomehouseholds makes a gas tax progressive across lower incomes and mitigatesprogressivity across upper incomes.

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571568

model at the values reported by PS for the US economy.17 For eachexercise, we then change a single parameter according to thevalues listed in Table 1. The results for relevant parameters arepresented in Table 5.

The parameter that contributes the most to explaining thedifferences between our estimates and those from PS is thedistance-related pollution damage. A higher value in our caseincreases the original PS estimate by 38.6%. Higher externalaccident costs in Mexico also contribute, increasing the originalestimate by 37.6%. Also noteworthy is the lower initial tax rate ongasoline (a subsidy, in our case), which explains 20% of thedifference with respect to PS. Interestingly, lower fuel efficiencyis not a major factor in the differences.

2 4 6 8 10

150

200

250

300

pollution (distance), cents/mile

optim

al ta

x, c

ents

/gal

5 10 15 20 25 30

150

200

250

300

pollution (fuel), cents/gal

optim

al ta

x, c

ents

/gal

2 4 6 8 10 12 14

150

200

250

300

congestion costs, cents/mile

optim

al ta

x, c

ents

/gal

2 4 6 8 10 12 14

150

200

250

300

accident costs, cents/mile

optim

al ta

x, c

ents

/gal

0.3 0.4 0.5 0.6 0.7 0.8 0.9120

140

160

180

200

220

240

gasoline price elasticity

optim

al ta

x, c

ents

/gal

0.2 0.3 0.4 0.5 0.6120

140

160

180

200

220

240

VMT portion of gas elasticity

optim

al ta

x, c

ents

/gal

0.25 0.3 0.35 0.4 0.45 0.5120

140

160

180

200

220

240

compensated labor supply elasticity

optim

al ta

x, c

ents

/gal

0.3 0.4 0.5 0.6 0.7 0.8 0.9120

140

160

180

200

220

240

VMT expenditure elasticity

optim

al ta

x, c

ents

/gal

Fig. 5. Sensitivity of optimal gasoline tax to parameter variation.

17 It should be kept in mind that PS parameters are expressed in 2000 prices.We chose not to update parameter values to 2011 prices to facilitate comparisonwith the original PS estimates.

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571 569

We also conduct a sensitive analysis, given the uncertaintysurrounding the central parameter values. For that purpose, wevary eight of the most relevant parameters, one at a time. The intervalsfor each case approximately follow the intervals considered by PS(2005) for the US. The results are presented in Fig. 5.

The optimal tax is particularly sensitive to distance-relatedpollution, congestion costs, accident costs, and the VMT portion ofgas elasticity. These parameters are related to the marginalexternal cost of fuel use, MECF (see Eq. (1d)). Notice that the firstthree parameters also explain the majority of the differencesbetween the estimates of PS (2005) and our estimates. In all theremaining cases, the optimal tax is predominantly between $1.70and $2.10 per gallon.

6. Final discussion and implications for public policy

Fuel subsidies in Mexico have cost 1.2% of GDP on averageduring the period 2007–2011. It should be noted that this figuredoes not include tax expenditures (foregone public revenues).In fact, after taking into account the estimated 29.6% decrease infuel consumption (see Table 3), foregone public revenues under agasoline tax of $1.90 per gallon would reach 1.42% of GDP(equivalent to 13.5% of non-oil tax revenues in Mexico). Addingdirect subsidies to tax breaks, the cost is approximately 2.62% ofGDP (approximately one-quarter of non-oil tax revenues). Thisfigure is large when one takes into account that value-added tax(VAT) collection in Mexico is approximately 3.8% of GDP and thatcorporate and personal income taxes together account for 5.1% ofGDP. To put this number into perspective, gasoline tax revenue(tax breaks plus direct subsidies) would nearly be able to financethe entire general government expenditure on health (2.8% ofGDP) in Mexico (see the World Health Organization database atapps.who.int/nha/database).

As previously mentioned, this large fiscal subsidy is not atypical inoil-exporting countries. In fact, since 2008, pre-tax energy subsidieshave been steadily increasing over time around the world, especiallyin developing oil-exporting countries (IMF, 2013). Given the negativeeffects that these subsidies may have on human health, the environ-ment, and traffic congestion, there is a strong case against the use ofthese subsidies. In fact, the major issue is not just eliminating fuelsubsidies but also finding ways to ameliorate such externalities. Asfrequently argued in the literature (see, for example, Sterner, 2007;Edlin and Karaca-Mandic, 2006), a gasoline tax may at least partiallyaccomplish this goal.

Our optimal tax estimate of $1.90 per gallon is well above thePS (2005) calculation for the US, even after updating their resultsat 2011 prices. This result is primarily explained by differences inthe estimated distance-related pollution damage, accident costs,and congestion costs. The presence of a fuel subsidy also con-tributes to such differences. Based on the parameters examined,we find that the existence of an older motor vehicle fleet in a LDCdoes not help to explain the difference in optimal tax estimatesbetween Mexico and the US.

We have also demonstrated that fuel subsidies (current andimplicit) primarily benefit the wealthiest deciles of the Mexicanpopulation. Contrary to the findings reported for developed countries,a fuel tax is progressive in an LDC such as Mexico (see also Sterner,2011). The explanation for this result is simple: the majority of peopleat the lowest deciles do not own a car, and thus, their spending on fuelis relatively low; as income increases, the share of the population thatdemands gasoline for vehicles increases (see footnote 4).

This result is striking because fuel taxes are typically avoidedbased on equity grounds. However, it should be stressed that thisfinding does not take into account the negative effects that agasoline tax may have on the poor through fuel-intensive sectors

such as public transport and the once-and-for-all effect on con-sumer prices. This issue merits further study that extends beyondthe scope of this paper. In particular, a general equilibrium modelwith heterogeneous households may provide a useful frameworkfor analyzing such effects. On the other hand, a scheme by which ashare of fuel revenue is used to compensate poor householdsthrough the subsidization of public transport might yield a betteroutcome in terms of lower external damage, including damagerelated to the environment and human health. In this regard, itwould be also interesting to estimate the welfare gains fromcutting subsidies (and implementing optimal taxes) on other fuelproducts (e.g., diesel), and compare such gains with thoseobtained from an optimal gasoline tax. These possible extensionsmay be a fruitful subject of future research.

Appendix

This appendix provides details on the main equations in theParry and Small (2005) model. Consider a static, closed economy inwhich the representative agent has the following utility function:

U ¼ u ψ ðC;M; T ;GÞ;N� �� φðPÞ�δðAÞ ðA1Þ

where variables are expressed in per capita terms. Here, C is theconsumption of the numeraire good, M is the vehicle-miles oftravel, T is the time spent driving, G is the government spending,N is the leisure, P is the amount of pollution, and A is the severity-adjusted traffic accidents. Functions uðU Þ and ψ ðU Þ are quasi-concave, whereas φðU Þ and δðU Þ are weakly convex functionsdenoting disutility from pollution and (external) accident risk.

Vehicle-miles traveled (VMT) are “produced” according to thefollowing homogeneous function:

M¼MðF;HÞ ðA2Þwhere F is fuel consumption and H is the monetary expenditureassociated with driving costs related to vehicle price and attributes.

Time spent driving is a function of both VMT and the inverse ofthe average travel speed π:

T ¼ πM ¼ πðMÞMU ðA3ÞHere, M is the aggregate miles driven per capita, and the functionπðU Þ satisfies π′40, reflecting the concept that an increase inaggregate VMT leads to more congested roads. From the agent'spoint of view, M is treated as exogenous so that the agent does nottake into account her own impact on congestion.

There are two types of pollutants. The first type is carbon dioxide,PF , which is proportional to aggregate fuel consumption per capita F .This pollutant causes the fuel-related damage arising from climatechange. The second type includes local air pollutants, PM , whichdepend positively on aggregate VMT.18 These pollutants can beexpressed in terms of the same units, and therefore, total pollution

18 Of course, local air pollutants such as carbon monoxide, nitrogen oxides andhydrocarbons are part of the gasoline combustion process and thus are related to fuelconsumption. However, the motivation for the specification in Eq. (A4) responds to theway pollution damage is estimated in practice. In particular, (local) environmental costsfrom motor vehicles are mostly explained by health costs (Delucchi, 2000). These costsarise from the emission of pollutants such as carbon monoxide, nitrogen oxides andhydrocarbons but not carbon dioxide, which does not directly impair human health. Onthe other hand, pollution costs from carbon dioxide arise from many sources ofcombustion, such as liquid fuels (e.g., gasoline), solid fuels and gaseous fuels. In fact,the literature estimates (global) environmental costs not only from carbon dioxide butalso from other “greenhouse gases” (for a review of estimates of such costs, see Tol et al.,2000). In the first case, pollution costs may be directly related to VMT. In the secondcase, environmental costs are related to fuel consumption in general and not specificallyto fuel consumption due to VMT. These estimates are incorporated in the equation forthe optimal gasoline tax in (1) through the terms EPM and EPF.

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571570

P may be written as follows:

P ¼ PF ðF ÞþPMðMÞ; ðA4Þwith properties P′

F ; P′M40. Because pollution depends on variables

at the aggregate level, the agent does not internalize the costs ofpollution from her own driving.

Traffic accidents A vary with aggregate VMT according to thefollowing equation:

A¼ AðMÞ ¼ aðMÞM U ðA5Þwhere aðMÞ is the average external cost per mile. This function(A5) assumes that traffic accidents are exogenous to the agent. Forthis reason, the function δðUÞ in (A1) is the expected disutility fromthe external cost of traffic accidents.19 The sign of a′ is ambiguous,reflecting the idea that heavier traffic causes more frequent butless severe accidents.

The representative agent derives income from labor L at thewage rate w. The agent pays a tax rate tL on labor income and a taxtF on gasoline consumption. If qF represents the price of gasoline,the agent's budget constraint is given by the following:

Cþ qFþtF� �

FþH¼ 1�tLð ÞwLU ðA6ÞLet L denote the agent's time endowment, which she maydistribute among labor, leisure, and driving activities. The timeconstraint may be written as follows:

LþNþT ¼ L ðA7ÞGoods are produced by perfectly competitive firms. The single

input is labor, and the production function is linear with respect toL, which implies a constant marginal product of labor and thus aconstant wage rate at the optimum. For convenience, w is normal-ized to one so that net labor income in (A6) may be written simplyas ð1�tLÞL.

Finally, the government finances exogenous spending Gthrough taxes on gasoline consumption and labor income. Thusthe government's budget constraint is as follows:

tLLþtFF ¼ GU ðA8ÞBy totally differentiating the labor supply function and the

government's budget constraint, the following may be shown (seeEquation C9 in Parry and Small, 2004):

tLdLdtF

¼MEBLtFdFdtF

�MEBL

εLLεcLLFðηMI�1ÞþEC

dMdtF

εLL�ð1�ηMIÞεcLL� ��

U

ðA9ÞAfter manipulating this expression and using (1), the effect of the

optimal gasoline tax on labor is given by the following equation:

tLdLdtF

¼ FηFFpF

� �MEBLðtnF�tF ÞþtnF�MECF� � ðA10Þ

Eq. (2) in the main text is obtained after dividing expression(A10) by tLL.

References

Anton-Sarabia, A., Hernandez-Trillo, F., 2013. El impuesto sobre gasolinas: una aplicaciónpara Ecuador, El Salvador y México. Working paper LC/W, Project documents,Economic Commission for Latin America and the Caribbean, September.

Arceo, E., Campos-Vazquez, R., 2010. Labor Supply of Married Women in Mexico:1990–2000. Working Paper XVI – 2010, El Colegio de Mexico.

Blundell, R.W., MacCurdy, T., 1999. Labor supply: a review of alternative approaches.In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics 3 Parte A,pp. 1559–1695.

Bovenberg, A.L., de Mooij, R.A., 1994. Environmental levies and distortionarytaxation. Am. Eco. Rev. 84 (4), 1085–1089.

Bovenberg, A.L., Goulder, L.H., 1996. Optimal environmental taxation in the presenceof other taxes: general-equilibrium analyses. Am. Eco. Rev. 86 (4), 985–1000.

Bovenberg, A.L., van der Ploeg, F., 1994. Environmental policy, public finance, andthe labor market in a second-best world. J. Public Eco. 55, 349–390.

Camaal, C., 2007. Oferta laboral en México: un enfoque de variables instrumentales.Rev. Ensayos 36 (1), 115–154.

Campos-Vazquez, R., 2010. The effects of macroeconomic shocks on employment:the case of Mexico. Estud. Eco. 25 (1), 177–246.

Cragg, M., Epelbaum, M., 1996. Why has wage dispersion grown in Mexico? Is it theincidence of reforms or the growing demand for skills?. J. Dev. Eco. 51 (1), 99–116.

Delucchi, M.A., 2000. Environmental externalities of motor-vehicle use in the US.J. Transp. Eco. Policy 34 (2), 135–168.

Edlin, A.S., Karaca-Mandic, P., 2006. The accident externality from driving.J. Political Eco. 114 (5), 931–955.

Gakenheimer, R., 1999. Urban mobility in the developing world. Transp. Res. A:Policy Pract. 33, 671–689.

Gong, X., Soest, A.V., 2002. Family structure and female labor supply in Mexico City.J. Hum. Resour. 37 (1), 163–191.

Harrington, W., McConnell, V., 2003. Motor Vehicles and the Environment.Resources for the Future Report, Washington, DC.

International Monetary Fund, 2013. Energy Subsidy Reform: Lessons and Implica-tions, Washington, DC.

Iracheta, A.X., 2006. La necesidad de una política pública para el desarrollo desistemas integrados de transporte en grandes ciudades mexicanas. El ColegioMexiquense, Mexico.

INEGI, 2010. Household Income and Expenditure Survey, INEGI, Mexico.Johnson, T., Alatorre, C., Romo, Z., Liu, F., 2009. Low-Carbon Development for

Mexico. The World Bank, Washington, DC.KPMG Peat Marwick, 1990. Changes in the Progressivity of the Federal Tax System, 1980

to 1990. Prepared for the Coallition Against Regressive Taxation, Washington, DC.Levy, S., 2008. Good Intentions, Bad Outcomes: Social Policy, Informality and

Economic Growth in Mexico. Brookings Institution Press, Washington, DC.Lin, C., Prince, L., 2009. The optimal gas tax for California. Energy Policy 37, 5173–5183.Litman, T., 2012. Gestión de la movilidad para México. Beneficios para su desarrollo

económico. ITDP Mexico, British Embassy in Mexico, and Victoria TransportPolicy Institute.

Medina, S., 2012. La importancia de la reducción del uso de los automóviles enMéxico. ITDP Mexico and British Embassy in Mexico.

Medina, S., Veloz, J., 2013. Desarrollo orientado al transporte. ITDP Mexico andBritish Embassy in Mexico.

Melgar Consultants of Mexico, 2011. Estadística de la Población de Vehículos enMéxico, 1972–2011. Report to the Mexican Auto Industry Association, Mexico.

OECD, 2011. Valuing Mortality Risk Reductions in Regulatory Analysis of Environ-mental, Health and Transport Policies: Policy Implications. OECD, Paris.

OECD-IEA, 2012. Energy Price and Taxes, Quarterly StatisticsInternational EnergyAgency, Paris, France.

Parry, I.W.H., Small, K.A., 2005. Does Britain or the United States have the rightgasoline tax? Am. Eco. Rev. 95 (4), 1276–1289.

Parry, I.W.H., Small, K.A., 2004. Does Britain or the United States have the RightGasoline tax? Discussion Paper 02-12, Resources for the Future.

Parry, I.W.H., Strand, J., 2012. International fuel tax assessment: an application toChile. Env. Dev. Eco. 17 (2), 127–144.

Parry, I.W.H., Timilsina, G.R., 2008. How Should Passenger Travel in Mexico City bePriced? Discussion Paper 08-17, Resources for the Future.

Polk Research Firm, 2013. Quarterly Commercial Vehicle Report, June.Poterba, J.M., 1991. Is the gasoline tax regressive?. In: Bradford, D. (Ed.), Tax Policy

and the Economy, vol. 5. MIT Press, Cambridge, pp. 145–164.Sandmo, A., 1975. Optimal taxation in the presence of externalities. Swed. J. Eco. 77 (1),

86–98.Small, K., 1992. Urban Transportation Economics: Fundamentals of Pure and Applied

Economics Series, vol. 51. Harwood Academic Publishers, Chur, Switzerland.Small, K.A., Kazimi, C., 1995. On the costs of air pollution from motor vehicles.

J. Transp. Eco. Policy 29, 7–32.Sterner, T., 2007. Fuel taxes: an important instrument for climate policy. Energy

Policy 35, 3194–3202.Sterner, T. (Ed.), 2011. Fuel Taxes and the Poor: The Distributional Effects of Gasoline

Taxation and Their Implications for Climate Policy. RFF Press, Routledge.Tol, R.S.J., Fankhauser, S., Richels, R., Smith, J., 2000. How much damage will climate

change do? Recent estimates. World Eco. 1, 179–206.Walls, M., Hanson, J., 1999. Distributional aspects of an environmental tax shift: the

case of motor vehicle emissions taxes. Natl. Tax J. 52 (1), 53–65.West, S.E., 2004. Distributional effects of alternative vehicle pollution control

policies. J. Public Eco. 88, 735–757.West, S.E., Williams III, R.C., 2007. Optimal taxation and cross-price effects on labor

supply: estimates of the optimal gas tax. J. Public Eco. 91, 593–617.World Health Organization, 2013. Global Status Report on Road Safety. Geneva,

Switzerland.

19 The internal costs of accidents are implicitly included in H.

A. Antón-Sarabia, F. Hernández-Trillo / Energy Policy 67 (2014) 564–571 571