Carbon Leakage from Fuel Taxes: Evidence from a Natural Experiment

Abstract

We exploit a fuel tax increase in Portugal to identify its effect on cross-border fuel sales and associated carbon leakage in the Spanish border regions. Using a difference-in-difference strategy, we find that while gasoline sales remained unaffected, diesel sales in Spanish border regions increased by 6–9%. Synthetic control methods confirm these estimates and attribute this differential effect by fuel type to routes frequented by heavy-duty vehicles, with large diesel tanks. We estimate a carbon leakage equivalent to 14–20% of Portugal’s annual mitigation commitment for road transport emissions. Our findings imply that heavy goods vehicles’ strategic behavior undermines the potential mitigation effects and revenue gains of transport climate policy, underscoring the need for coordinated policies in similar federal or quasi-federal contexts.

Full text

Vol.:(0123456789)
Environmental and Resource Economics (2024) 87:3235–3270
https://doi.org/10.1007/s10640-024-00914-6
Carbon Leakage fromFuel Taxes: Evidence fromaNatural
Experiment
JordiJ.Teixidó1 · F.JavierPalencia‑González2· JoséM.Labeaga2·
XavierLabandeira3
Accepted: 15 August 2024 / Published online: 29 August 2024
© The Author(s) 2024
Abstract
We exploit a fuel tax increase in Portugal to identify its effect on cross-border fuel sales and
associated carbon leakage in the Spanish border regions. Using a difference-in-difference
strategy, we find that while gasoline sales remained unaffected, diesel sales in Spanish
border regions increased by 6–9%. Synthetic control methods confirm these estimates and
attribute this differential effect by fuel type to routes frequented by heavy-duty vehicles,
with large diesel tanks. We estimate a carbon leakage equivalent to 14–20% of Portugal’s
annual mitigation commitment for road transport emissions. Our findings imply that heavy
goods vehicles’ strategic behavior undermines the potential mitigation effects and revenue
gains of transport climate policy, underscoring the need for coordinated policies in similar
federal or quasi-federal contexts.
Keywords Carbon leakage· Fuel tax· Cross-border fuel sales· Carbon price· Road
transportation· Climate policy
JEL Classification Q58· R48· H23· H26
1 Introduction
As climate-related crises worsen, policymakers are increasingly turning their attention to
the mitigation of greenhouse gas (GHG) emissions from the transport sector, especially
those generated by road transport. Transport is the only sector in which current GHG emis-
sions are still above 1990 levels—33% higher in the EU (EEA, 2022)– and it has become
the largest GHG contributor in many countries, including the US (EPA 2023). Moreover,
population and income growth project further increases in miles traveled, car ownership
* Jordi J. Teixidó
j.teixido@ub.edu
1 Universitat de Barcelona, Barcelona, Spain
2 UNED, Madrid, Spain
3 Universidade de Vigo andEcobas, Vigo, Spain
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3236
J.J.Teixidó et al.
rates and demand for freight transport globally, which, given current policies and technolo-
gies, will result in higher GHG emissions (IEA 2022).
Against this backdrop, many countries are ramping up their climate policies on road
transportation. This includes the EU, which in recent years has adopted more ambitious
climate targets1 by proposing, inter alia, a revision of the Energy Taxation Directive (ETF)
and the coverage of road transport emissions by a new emissions trading system that
will become operative as of 2027. In this regard, pricing instruments –including carbon
pricing and energy taxes—are considered a cost-effective approach to reducing emissions
(Gago etal. 2014). Yet, despite the efforts of EU legislation to harmonize energy taxation
across the Union, differences in fuel prices between neighboring countries can jeopardize
potential gains from these policies by becoming a source of carbon leakage and revenue
loss. Here, we examine how cross-border fuel purchases, so-called ‘fuel tourism’ –that is,
the optimizing behavior of drivers who cross a border to fill up their vehicles at a lower
price– is interacting with climate policies in the road transportation sector.
Cross-border fuel purchase substitution has been well documented in many territories,
including Europe (Banfi etal. 2005; Jansen and Jonker 2018; Leal etal. 2009; Morton
etal. 2018) and the US (Manuszak and Moul 2009). In this paper, we analyze the role that
this strategic behavior plays in the current context of climate policies, especially, that of
increasing fuel prices via taxes or carbon pricing. Significantly, drivers have been found
to react more to changes in fuel prices resulting from taxes or carbon pricing than to the
same price change derived from market forces (Antweiler and Gulati 2016; Li etal. 2014;
Scott 2012; Tiezzi and Verde 2016). This reaction is explained in terms of the salience or
persistence of the tax versus market price oscillations. Hence, based on the assumption
that drivers fill up their tanks in the low-price country when the latter is considered close
enough, this paper explores whether a tax-motivated price change further increases cross-
border fuel purchases. In short, we seek to determine the elasticity of cross-border fuel
purchases to cross-border changes in energy taxes or similar climate policies.
To do so, we exploit the plausibly exogenous change in the fuel tax in Portugal, the
Imposto sobre os Produtos Petrolíferos (ISP), to analyze fuel consumption—both of
gasoline and diesel—in Spain at the province level (NUTS 3). Spain and Portugal share
the longest uninterrupted border in the EU (1214km), characterized by numerous crossing
points, while gasoline and diesel prices have traditionally been much lower in Spain, an
ideal mix to ensure cross-border fuel purchases are an everyday reality. In February 2016,
Portugal raised its fuel tax by six cents of a euro, making cross-border fuel substitution, in
theory, even more appealing. Here we identify, and quantify, the effect that this tax increase
had on fuel consumption and emission rates and discuss its implications in terms of climate
mitigation policies.
In our identification strategy we use Spain’s non-border provinces and, as such, those
not exposed to the tax change in Portugal, as a control group for the seven treated prov-
inces that do share a border with Portugal. We employ two seminal quasi-experimental
methods: a two-way fixed effects difference-in-difference estimator for the average effect
and a synthetic control approach (Abadie 2021) for heterogeneous effects analysis. In so
doing, we use monthly data, spanning January 2011–December 2019, of both gasoline and
diesel consumption at the province level, controlling for a range of potential confounders,
including fuel prices and the number of filling stations as well as income and demographic
1 In July 2021, the EU Commission published its “fit-for-55” package, committing itself to reduce GHG
emissions by 55% (compared to 1990 levels) as a step to achieving climate neutrality by 2050.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3237
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
characteristics. We show that our main key assumptions are met for both identification
strategies, hence yielding credible causal results.
The main result to emerge from the difference-in-difference estimation is the
significantly different outcomes presented by gasoline, on the one hand, and diesel, on the
other. While consumption of the former shows no significant response to the cross-border
tax increase—indicating that cross-border fuel substitution follows a ‘business-as-usual’
pattern, diesel sales increase by around 6–9% in the border provinces. These results are
consistent across different specifications and matching procedures, including propensity
score and entropy balance matchings (Abadie and Imbens 2011; Hainmueller 2012). They
are also consistent with alternative difference-in-differences estimators like the doubly-
robust estimator proposed by Sant’Anna and Zhao (2020) or the synthetic difference-
in-differences proposed by Arkhangelsky etal. (2021). Moreover, this result cannot be
attributed to a potentially endogenous distribution of filling stations. We estimate a cross
elasticity of fuel sales in Spain with respect to Portuguese tax changes of 1.8 for diesel and
0.1 for gasoline. This difference can be explained by the fact that heavy goods vehicles—
run almost exclusively on diesel and with huge fuel tanks—constitute the main source of
the cross-border response to the change in tax.
To analyze heterogeneity across the border (treated) provinces, we construct synthetic
provinces for each of the seven border provinces, which provides additional confirmation
of our outcomes –namely, a marked impact on diesel consumption but no effect on that
of gasoline. The synthetic control procedure provides additional insights into the local
distribution of this particular effect. Although a positive effect on diesel consumption
remains for most border provinces, only three of them –Badajoz up 7%, Huelva up
17%, Zamora up 20%—show a statistically significant increase at the standard levels of
confidence. These three provinces lie on routes carrying the highest volumes of heavy-duty
vehicles between Portugal and Spain (OTEP 2020).
The main implication of our findings is that heavy goods vehicles are channeling the
carbon leakage attributable to pricing instruments in the road transportation sector. This
result is relevant not only for cross-country trade but also for trade in federal or quasi-
federal countries where taxation policies might differ. Emission reduction is likely to be
confounded by emission leakage to neighboring countries in conjunction with a loss in
revenue. Here, the tax change introduced in Portugal results in an annual carbon leakage
of 55,000–80,000 tCO2, equivalent to 14–20% of the country’s annual CO2 mitigation
commitment for road transport for 2030 (NECP-Portugal 2019). These emissions, however,
far from being mitigated, are added to Spain’s annual emissions, while Portugal must face
the corresponding foregone revenue from its diesel tax.
We contribute to the broader literature on fuel taxation, border differences in taxes on
purchase decisions and, more generally, horizontal tax externalities by factoring in the
issue of carbon leakage in the current context of mitigation policies. Hence, this paper can
be related to several strands of this literature. First, several papers show that tax-driven
changes in fuel prices have higher elasticities than market-driven changes –the case, for
example, of changes in fuel tax in the US (Tiezzi and Verde 2016; Li etal. 2014; Scott
2012; Davis and Kilian 2011) and carbon taxes in Sweden (Andersson 2019) and British
Columbia, Canada (Antweiler & Gulati 2016). This outcome, however, has not previously
been analyzed from a cross-border and carbon leakage perspective, which is of obvious
relevance in the current context of the ramping up of climate policies. Second, the literature
analyzing the influence on domestic fuel demand from cross-border price differences—
also called fuel tourism—has primarily delivered information about cross-border (final)
price elasticities but with no clear focus on tax-motivated price changes (Banfi etal. 2005;
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3238
J.J.Teixidó et al.
Coglianese etal. 2017; Coyne 2017; Ghoddusi etal. 2022; Jansen and Jonker 2018; Leal
etal. 2009; Manuszak and Moul 2009; Morton etal. 2018). Here, we show that tax-driven
fuel tourism can be fuel-specific.
A number of papers have studied horizontal tax externalities in multi-jurisdictional
taxation for different goods. In the most similar study to the current one, Marion and
Muehlegger (2018) analyze the case of the diesel taxes owed by interstate truck drivers
in the US and show how they evade taxes by underreporting the amount of fuel consumed
and their mileage in high-tax states. A part of this literature has focused on cross-border
cigarette taxes (Agaku etal. 2016; DeCicca etal. 2013; Harding etal. 2012; Lovenheim
2008). The main lesson to be drawn from these papers is that the health benefits from
a higher tax on cigarettes are not fully captured because of smuggling and other cross-
border tax avoidance strategies. Here, we assert the same rationale for transport fuel, only
that besides any potential health benefits (also present in the transportation sector), any
climate policy gains are foregone due to carbon leakage, to which we must add a notable
tax revenue loss.
In the section that follows, we describe the setting of this natural experiment, i.e. the
fuel tax increase in Portugal, and report transport fuel demand data for both Spain and
Portugal. In Sect.3, we describe the data used and the identification strategies we employ.
Section4 presents our main results and Sect.5 discusses the main policy implications to be
derived from them. Section6 concludes.
2 Fuel Prices inPortugal andSpain
In February 2016, the Portuguese government raised excise taxes on transportation fuels:
the ISP saw an increase of 0,06 €/l. Figure1 shows the evolution of all fuel taxes—includ-
ing those on both diesel and gasoline– in Spain and Portugal between 2011 and 2019.
While taxes on diesel are lower in both countries, the difference in the case of gasoline is
more marked, although after February 2016, the gap between the two countries widened
for both fuel types.
However, fuel prices tend to be somewhat volatile and these tax differences do not
necessarily translate proportionally to final fuel prices; indeed, the tax increase can
be offset by the fuel price variation. This being the case, the relative price differences
between Spain and Portugal would not have been as dramatic as the tax increase itself
might suggest. Similarly, the timing of the tax implementation could influence its
salience, either amplifying or diminishing its effects. In this context, despite the tax
hike occurring during a period of low fuel prices in both countries, the price dispar-
ity between them exhibits a relatively stable trend above its minimum level, fluctuat-
ing by approximately 20 cents for gasoline and around 10 cents for diesel. Figure2
shows final fuel prices in the two countries. Spain has traditionally charged lower prices
than Portugal, especially as regards gasoline. One month before the introduction of the
Portuguese tax increase, the pump price of gasoline in Portugal stood at an average of
€1.31 per liter compared to €1.11 in Spain. This 20-cent difference climbed to 25 cents
after the rise in tax. Diesel prices, in contrast, were more similar before the new tax: on
average, diesel in Spain was about 9 cents cheaper before the rise in ISP and 15 cents
cheaper after. Hence, the price differential of Spanish gasoline continued to be greater
than that of diesel prices when compared to the respective price at the pumps in Portu-
gal: Spanish gasoline being about 25 cents cheaper and Spanish diesel 15 cents cheaper.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3239
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Fig. 1 All fuel taxes in Portugal and Spain (€/l). Notes: This figure plots the evolution in all diesel and gaso-
line taxes in Spain and Portugal. The vertical line signals the six-cent increase in ISP in Portugal. Overall,
following the tax hike, the ISP on diesel and gasoline increased to €0.34 and €0.58 per liter, respectively.
This represents 52 and 62%, respectively, of all fuel taxes. Source: Weekly Oil Bulletin prices History, pro-
vided by Directorate-General Energy (DG-ENER)
Fig. 2 After-tax fuel prices in Portugal and Spain (€ /l). Notes: The upper panel plots the evolution in after-
tax fuel prices—diesel and gasoline—in Spain and Portugal. The vertical line signals the six-cent increase
in ISP in Portugal. The graph in the lower panel shows price differences between the two countries (where
zero represents no difference and negative values represent cheaper prices in Spain). Source: Weekly Oil
Bulletin prices History, provided by Directorate-General Energy (DG-ENER)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3240
J.J.Teixidó et al.
Importantly, Fig.2 shows national averages under imperfect competition. Petrol stations
on both sides of the border may behave strategically in price setting, potentially playing
an important role in regional demand. Unfortunately, we do not observe these prices at
the petrol station level.
However, these differences in fuel price are only of any relevance to those regions
located near the border between Spain and Portugal. Figure3 identifies Spain’s provinces,
our observation unit, and differentiates between the seven border provinces that serve as
our treated group (in orange) and the remaining control provinces (in green). As such, we
assume that the strategic tax avoidance behavior we seek to identify manifests itself solely
in these seven provinces while all the other provinces will be totally unaffected.
Our natural experiment relies on the similarity between our treated and control prov-
inces in all aspects regarding their transport fuel demand. Importantly, because transporta-
tion fuel can be considered a homogenous product, price is expected to be a highly rel-
evant demand factor, ceteris paribus. Figure 4 compares the evolution in the fuel prices
of the treated and control provinces between 2011 and 2019. On average, fuel prices have
remained largely parallel, with those in the border provinces 2 cents per liter higher than
those in the other provinces. Thus, on average, the drivers in our control group have no
incentive to fill their tanks in the provinces of the treated group, the incentive existing
solely for drivers from/to Portugal. This small, yet parallel, difference is further confirmed
when we examine the evolution in fuel prices in each border province compared to the
price evolution in that of its immediate neighbor (Fig.8). Only Zamora and Salamanca
are capable of attracting drivers from Ourense and Cáceres (also treated), respectively,
but not to any greater degree after the rise in the Portuguese fuel tax. In other words, we
detect a parallel trend in prices. In short, our strategy is designed to identify changes in fuel
Fig. 3 Spanish provinces (NUTS 3). Notes: Spanish provinces are NUTS 3 regions. In orange, provinces
bordering Portugal (in light gray); in green, the remaining provinces serving as controls
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3241
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
consumption as a response to the tax change in Portugal, above and beyond the prevailing
regional pattern, such as, any existing cross-border fuel substitution. In the following sec-
tion we outline our research design in detail.
3 Methodology
We exploit the exogeneity of the tax increase in Portugal to analyze to what extent
domestic fuel consumption is explained by the cross-border tax. In this section, we
describe the two empirical strategies employed—difference-in-difference and a synthetic
control procedure—addressing the key assumptions that each method requires for a net
causal identification. Methodologically, each method provides different levels of results:
while the difference-in-differences strategy is intended to identify the average effect of the
cross-border tax on the entire border provinces and thus determine the resulting carbon
leakage, the synthetic control focuses at the case study level, examining the effect specific
effect to particular provinces and providing a more nuanced analysis of the heterogeneity
of the cross-border tax effect. Finally, we also describe the data used.
3.1 Difference‑in‑Difference Estimation
The validity of our difference-in-difference approach rests on the fact that Portuguese
fiscal policy can be considered exogenous from Spanish fuel consumption and, related
to this, that the parallel trends assumption holds, i.e., had the tax not been increased,
fuel consumption in the treated and control groups would have followed the same par-
allel trends as before the intervention. The plausibility of this assumption can only be
assessed by examining pre-trends: that is, if fuel consumption in the border provinces
Fig. 4 Gasoline and diesel price evolution in treated and control Spanish provinces
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3242
J.J.Teixidó et al.
followed the same evolution as that in the other (comparable) Spanish provinces before
the tax hike in Portugal. Then, we could reasonably assume that had there been no
change in the tax rate, these trends would have continued to follow the same parallel
course, indicating that any significant differences can be attributed to the change in tax
policy. Figure5 shows this assumption to be plausible in our context: before the inter-
vention in February 2016, treated and control provinces followed a largely parallel trend
in terms of both their diesel and gasoline consumption.
We estimate the following two-way fixed effects difference-in-difference model:
where FC represents the fuel consumption (either diesel or gasoline) in province i in
month t. The main variable of interest is
Tit =Borderi×TaxChanget
, where
Borderi=1
if the province is a border province and
TaxChanget=1
after February 2016. Therefore,
𝛽
is our difference-in-difference coefficient, identifying the change in fuel consumption
in Spanish border regions as a response to a tax change in Portugal. Xit is the set of rel-
evant control covariates, including the logarithm of after-tax fuel prices, regional GPD per
capita (in logs), share of population in the province’s capital as a measure of the level of
urbanization, the province’s population (in logs) to account for the province scale effect
and (monthly) average daily traffic as measured by permanent control stations in main
(1)
log(FC)it =𝛼+𝛽Tit +𝜆Xit +𝛾i+𝜂t+𝜀it
Fig. 5 Fuel consumption in treated (border) and control Spanish provinces. Notes: This figure shows the
average consumption of gasoline and diesel (in liters) at the provincial level for seven provinces sharing a
border with Portugal (Border provinces) and for the remaining forty-one Spanish provinces. Our identifica-
tion strategy relies on the fact that, before the tax change in Portugal –our treatment (dashed vertical line)–
both groups evolved in parallel
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3243
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
highways. Fixed effects include those of the province (NUTS 3), NUTS 2 region-year
(Comunidad Autónoma), month and year and, finally, month-year effects.2
Because certain differences in the covariates might confound our effect of interest,
Table1 shows the different strategies adopted in making the control and treatment groups
as comparable as possible in terms of these very covariates. The first group of rows show
the differences between the treated and control provinces. The treated provinces are sig-
nificantly poorer, less populated, with higher average prices and lower average daily traffic
than the controls. Although these differences can be controlled for by using them as con-
trol variables in the main regressions, we seek to improve comparability using matching
procedures. This serves to make our estimators doubly robust (Bang and Robins 2005).
In the second group of rows, we use a propensity score matching (nearest neighbor) pro-
cedure. This improves comparability between the treated and control groups in terms of
the relevant covariates. However, one disadvantage of using propensity score matching is
that the sample is reduced to the provinces that have common support in the covariates.
In the third group of rows, we use the entropy balancing procedure (Hainmueller 2012),
which involves a generalization of the propensity score matching: instead of using only
Table 1 Treated and control
Spanish provinces by observable
characteristics
Mean values and differences (t-test) between treated (border with
Portugal) and control provinces (no border) for the main control
covariates for the year before the tax change in Portugal (i.e. 2015).
***p<0.01, **p<0.05, *p<0.1
Variables T = 0 T = 1 Diff
Full sample
ln (income) 6.731 6.664 0.067***
Share inhab. in Prov. Capital 0.323 0.276 0.046***
ln(population) 13.282 12.972 0.310***
ln(before-tax price of diesel, 2016) 0.088 0.121 − 0.033***
ln(before-tax price of gasoline, 2016) 0.182 0.212 − 0.030***
ln(avg. daily traffic) 9.602 9.030 0.572***
PSM matched sample
ln (income) 6.663 6.669 − 0.006
Share inhab. in Prov. Capital 0.266 0.278 − 0.012
ln(population) 12.921 12.975 − 0.054
ln(before-tax price of diesel, 2016) 0.105 0.122 − 0.018*
ln(before-tax price of gasoline, 2016) 0.195 0.214 − 0.019**
ln(avg. daily traffic) 9.117 9.057 0.060
Entropy balanced sample
ln (income) 6.654 6.669 − 0.015
Share inhab. in Prov. Capital 0.272 0.278 − 0.006
ln(population) 12.997 12.975 0.022
ln(before-tax price of diesel, 2016) 0.113 0.122 − 0.009
ln(before-tax price of gasoline, 2016) 0.200 0.214 − 0.014
ln(avg. daily traffic) 9.083 9.057 0.026
2 It is important to note that the isolated variables composing the treatment variable (Tit), are part of the
province (Borderi) and monthly (TaxChanget) effects.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3244
J.J.Teixidó et al.
provinces with common support, the entropy balance reweights observations in the control
group so that the mean and variance of the covariates resemble the mean and the vari-
ance of the treated group. Hence, in contrast to propensity score matching, where some
units are discarded, entropy balancing uses all the observations in the control group, prop-
erly reweighted. This means that entropy balancing minimizes information loss from the
pre-processed data. Again, differences between the treated and control groups are further
reduced for most of the covariates.
The estimates can be considered as being unbiased as long as the parallel trends, the
no anticipation and stable unit treatment value assumptions (SUTVA) hold. While parallel
trends and no anticipation seem plausible here (see Fig. 5), event estimates are also
provided to further assess their plausibility (by assessing pretreatment differences in trends,
i.e., pre-trends).
A SUTVA violation might originate from control provinces being affected by the cross-
border tax hike. For instance, this would be the case if border provinces reacted to the tax
by lowering or raising their prices. This could affect neighboring provinces and, hence,
lead to a SUTVA violation. However, this does not seem to be the case, as the provinces
follow parallel trends as regards their average pricing (see Figs.4 and 8 for further details).
Other spillovers could be attributable to drivers (with origin or destination in Portugal)
filling their tanks before/after the border provinces, which would impact fuel sales in the
control provinces. This would be rational if, for instance, filling stations near the border
increased their prices in response to the tax hike. Unfortunately, we are unable to observe
actual filling behavior. We do, however, control for some of the other potential confounders
by means of observables; yet, we recognize that this remains vulnerable to unobservable
factors for which we cannot control.
In this regard, unobserved time-varying factors are not dealt with in a difference-in-
difference strategy, only time-invariant factors are controlled for.3 These unobserved factors
could affect differently border provinces even in the absence of the tax change in Portugal,
casting doubts on the key assumption of parallel trends. Recent research has shown that
although pre-trends testing may be intuitive to assess the plausibility of post-treatment
parallel trends, conditioning the analysis on passing pre-trends tests can introduce several
statistical issues (Roth 2022). In what follows, the synthetic control methodology serves
as a generalization of the difference-in-difference framework, accounting for these time-
varying unobserved factors (Abadie and Gardeazabal 2003; Abadie etal. 2010, 2015).
Moreover, by focusing on the case-study level (heterogeneity in the treatment effects by
province), the synthetic control methodology enables a more nuanced understanding of the
geographic determinants driving the average effect.
3.2 Synthetic Control Method
The synthetic control method estimates a counterfactual case scenario for each of the
treated provinces by using control units as a donor pool. Control provinces are properly
weighted by optimally chosen weights that minimize pre-treatment characteristics with the
treated unit so as to resemble a synthetic treated unit. Thus, for example, we can compare
observed Ourense with synthetic Ourense, the difference between the two being that the
3 Note, however, that our difference-in-difference specification does capture some time-varying unobserved
factors by including fixed effect interactions for region-year and month-year.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3245
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
latter did not experience the increase in the Portuguese fuel tax. As discussed earlier, this
method controls for unobserved time-varying heterogeneity.
More formally, the synthetic province serving as the counterfactual is represented by a
vector of optimal weights w = (w2,..., wJ+1)’, where 0 ≤ wj ≤ 1 for j ∈ {2,..., J + 1}
and
∑J+1
j=2
wj=
1
. The value of w in the synthetic unit is selected to resemble the pre-treat-
ment characteristics of the unit of interest (a specific border province). The optimal weights
w are chosen by minimizing the difference between the pre-intervention predictors for the
treated units and each control unit, so that
w=argmin
w
[X
1
−X
0
w]�V[X
1
−X
0
w]
, where X1
and X0 are the pre-treatment characteristics of the treated and control units, respectively,
and V is a diagonal matrix that weights pre-intervention predictors in accordance with their
power to predict the outcome (i.e., amount of fuel consumption).
The impact of the tax on fuel consumption can then be evaluated simply in terms of the
difference between the actual outcome of the treated province and that of the optimally
weighted control provinces (which resemble the treated unit). Thus,
𝛽it
in Eq. (2) is the
impact of the cross-border tax increase on domestic fuel consumption:
Table9 in the appendix shows the mean values of the predictor variables used for both
the observed border provinces and their synthetic controls. Here, instead of population
and the average daily traffic, we use the number of filling stations per capita. Note that we
did not use this variable before because of potential issues of endogeneity (not an issue
for synthetic methods) and because it is only available from 2014 onwards, which would
reduce the time span of the sample in the panel estimator. Here, however, this does not
constitute a problem and, moreover, it reduces the mean squared prediction error (MSPE),
which captures the difference between the observed unit and the estimated counterfactual
and, hence, the match between the treated and the synthetic unit. This, together with using
pre-treatment fuel consumption as a covariate, helps “soak up” the heterogeneity (Abadie
etal. 2010). Tables10 and 11 show the synthetic control weights used.
The synthetic control method provides specific estimates for each border province. This
allows us to focus more closely on the effects of each particular border province and its
related geographical characteristics, altogether resulting in a more informative tool for
policy making.
3.3 Data
Our main variable of interest is transportation fuel sales at Spanish filling stations, aggre-
gated at the monthly provincial level (NUTS 3). Our data sample spans January 2011 to
December 2019 and includes 48 peninsular provinces (the Canary Islands and the autono-
mous cities of Ceuta and Melilla having been excluded). We obtain these from the Span-
ish National Markets and Competition Commission (CNMC Data 2021). We also record
average fuel prices, the number of filling stations, and how many of these are located near
the Portuguese border, taken from CNMC (2021) and Geoportal Gasolineras (Ministerio
Transición Ecol. 2021). These covariates, together with other relevant socio-demographic
characteristics obtained from the Spanish Institute of Statistics (INE 2021) and DGT
(2021)—namely, population of the provinces, share of that population in the provincial
capital, income and average traffic intensity in main highways—are used to balance treated
and control provinces. Table2 shows descriptive statistics and dataset sources.
(2)
𝛽
it =log(FC1t)−∑
J+1
j=2
wjlog(FCjt
)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3246
J.J.Teixidó et al.
4 Results
4.1 Average Treatment Effects ontheTreated: Difference‑in‑Difference Results
Table3 shows the main results from the difference-in-difference strategy for the various
samples. We report our results for both diesel (top panel) and gasoline sales (bottom
panel). We provide treatment effects for the full sample (column 1), for the corresponding
matched sample (column 2) and for the entropy balanced sample (column 3). According to
our model, diesel sales in border provinces increased by about 8.6% compared to sales in
the control group (column 1). This is 8.3% in the matched sample, with fewer observations,
and 8.9% in the entropy balanced sample.4 The latter shows the best balance and is, hence,
our preferred specification. The implication is, therefore, that diesel sales at filling stations
in border provinces increased by about 8–9% in response to the cross-border fuel tax
increase.
In the case of gasoline, our results differ strikingly. Here, the cross-border tax does not
appear to affect gasoline sales at all.5 This is counterintuitive also because the price of
gasoline is about 20 cents cheaper in Spain than in Portugal, while diesel is only 10 cents
cheaper. Note that this does not mean that there is no cross-border fuel substitution for gas-
oline; rather, for gasoline drivers this does not increase in response to the rise in fuel tax in
Portugal. Our empirical strategy is designed to identify the response to a cross-border tax
increase and not the response to price differentials. Hence, what our results show is that,
unlike diesel sales, the sales of gasoline do not increase in the border provinces in response
to the cross-border tax increase.
In-time placebo tests (Table13 in the Appendix), i.e. moving the treatment date to Feb-
ruary of the four previous years, while dropping the observations from the period actually
Table 2 Descriptive statistics
This table shows descriptive statistics and sources of the datasets used. These are monthly averages
expanding from January 2011 to December 2019
Variable Mean Min Max Source
Income 872.30 630.71 1 249.04 INE (2021)
Share inhab. in Prov. Capital 0.32 0.09 0.77 INE (2021)
Population 896 919 78 863 6 686 513 INE (2021)
Diesel sales (liters) 41 700 000 613 089 242 000 000 CNMC Data (2024)
Gasoline sales (liters) 9 606 808 360 411 76 100 000 CNMC Data (2024)
Before-tax price of Diesel 1.19 0.66 1.47 Min. Tr. Ecol. (2024)
Before-tax price of Gasoline 1.27 0.80 1.56 Min. Tr. Ecol. (2024)
Avg. daily traffic 18 322.47 2 636.50 91 024.60 DGT (2021)
Border prov. (Treated = 1) 0.13 0 1 –
4 Table4 in the Annex show estimations without covariates. Results remain highly consistent with those in
Table3.
5 In Table4 in the annex, we show DiD estimates with no control covariates. For gasoline, this results in a
small significant negative effect that vanishes when potential confounders are controlled for in Table3.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3247
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Table 3 Difference-in-difference
estimates
This table shows the two-way fixed effects difference-in-difference
estimator for the different specifications and samples. Coefficients
can be interpreted as the change in the fuel consumption of the border
provinces as a result of the tax change in Portugal (ISP). This is shown
for all border provinces. Robust standard errors, clustered at the
province level, in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1
(1) (2) (3)
ln(Diesel sales)
Border ×ISP
0.086*** 0.083*** 0.089***
(0.030) (0.024) (0.026)
Constant 14.831 40.333** 36.603**
(15.890) (17.530) (16.577)
R-squared 0.768 0.841 0.985
R2 adj 0.757 0.816 0.985
ln(Gasoline sales)
Border ×ISP
− 0.011 0.011 0.007
(0.013) (0.016) (0.015)
Constant 7.676 6.201 16.259
(10.751) (13.876) (13.341)
R-squared 0.818 0.868 0.986
R2 adj 0.810 0.847 0.986
Observations 4644 1432 4644
Number of id_province 43 36 43
Sample All PS match Entropy B
Control vars YES YES YES
Province FE YES YES YES
Year FE YES YES YES
Month FE YES YES YES
Year-month FE YES YES YES
Year x Region FE YES YES YES
Cluster s.e Province Province Province
Table 4 Synthetic control
estimates
This table shows the average difference between observed diesel
consumption and the estimated counterfactual scenario after February
2016, when the fuel tax was increased in Portugal. We use the
placebo-based inference by which we rank just how extreme the result
of the actual treated unit is by means of the ratio between the pre- and
post-treatment MSPE. ***p<0.01, **p<0.05, *p<0.1
Province Effect p value
Badajoz 0.07* 0.08
Cáceres 0.01 0.56
Huelva 0.17* 0.08
Ourense − 0.06 0.51
Pontevedra − 0.03 0.70
Salamanca 0.07 0.26
Zamora 0.20** 0.02
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3248
J.J.Teixidó et al.
treated, further confirm these results. Thus, in-time placebos for diesel or gasoline show no
significant effect.
Figure6 shows the event coefficients according to the different sampling specifications
when considering all the borders. In all cases, the coefficients add further plausibility to
our parallel trends assumption: i.e., before the Portuguese tax hike, any differences in fuel
sales between the treated and control provinces were not significantly different from zero.6
In the case of gasoline, pre-treatment trends are better dealt with in the entropy-balanced
sample, our preferred specification, which confirms this differential effect by fuel type.
One potential explanation for this differential effect might be the higher share of die-
sel vehicles in both Spain and Portugal—in 2020, diesel cars represented 59.9 and 57.9%
of the total in Portugal and Spain, versus 37 and 39.5% of gasoline-fueled cars, respec-
tively (ACEA 2022). This, however, cannot account for the full story. A more plausible
Fig. 6 Event study for diesel (top) and gasoline(bottom) sales. Notes:This figure plots results from an event
study of the difference in fuel consumption –both diesel and gasoline– between border and non-border
Spanish provinces by sample and matching strategy
6 Although we cannot reject zero pre-trends from event estimates (hence providing plausibility of paral-
lel trends), we also cannot reject pre-trends that, under smooth extrapolations to the post-treatment period,
would bias treatment estimates. For example, one could extrapolate an upward-sloping linear trend within
the 95% CI from the last placebo in January 2011 to the last treatment effect in December 2019. This would
suggest that diesel sales in border provinces might have been on an upward trend different from that in the
control provinces, violating our parallel trends assumption. Based on observed pre-trend patterns, Ram-
bachan and Roth (2023) propose a sensitivity test to assess a bound (M) on how much the counterfactual
difference in trends can change to not reject the null effect. Applying the proposed test to collapsed annual
data, we find M < 0.04 in 2017 and M < 0.06 in 2018 and 2019 (breakdown values). This means our esti-
mates remain significant, provided that the parallel trends violation comes from a linear difference in trends
(M = 0) or with deviations no greater than the breakdown values M for each annual estimate. Essentially,
the slope of that trend cannot change by more than M across consecutive periods to rule out the zero effect.
See Appendix Figure2.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3249
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
explanation is that heavy goods vehicles, run on diesel and fitted with large tanks with a
capacity for up to 1500 L of fuel, drive the cross-border tax response.
Overall, these estimates translate into 30 million additional liters of diesel consumed per
year in the border provinces because of the cross-border tax, representing an annual carbon
leakage of 80,000 tCO2. In the following section, we analyze each border province using
the synthetic control method to further disentangle this effect.
4.2 Heterogeneity intheTreatment Effects: Synthetic Control Results
Figure7 shows the fuel consumption trajectories for both diesel and gasoline in the syn-
thetic border provinces (grey plots) and in the observed border provinces. Despite some
small differences, the trajectories of the synthetic provinces provide a close match with
those of the treated units (border provinces). In the case of diesel sales, some provinces
present a marked increase in their consumption over that of their counterfactual scenario: a
visual inspection shows that Zamora, Badajoz, Huelva and Salamanca all present a greater
divergence after treatment. This indicates that the cross-border tax change increased Zamo-
ra’s diesel consumption by an average of 20%, Huelva’s by 17% and Badajoz and Salaman-
ca’s by 7% each between February 2016 and December 2019 (Table4). In contrast, in the
case of gasoline sales, no differences are detected between the observed and the synthetic
consumption series, thus confirming our main findings from the difference-in-difference
analysis, and providing further robustness to our findings regarding diesel consumption.
To assess the statistical significance of the impact on diesel sales, we construct p-values
using the placebo-based inferential technique (Abadie etal. 2010). This involves applying
the synthetic control method to each province in the sample as if it were a treated unit
and then computing their respective synthetic controls to see if there is any post-policy
treatment effect. If the estimated effect for the actual treated units—the border provinces—
is relatively larger than that found for the control provinces, then we can assert the
significance of the effect. Figure10 in the appendixes shows the post- and pre-treatment
MSPE ratios for the treated and placebo units: a relatively high ratio is indicative of a
unit presenting a larger gap post-policy than pre-policy. We then calculate p-values as the
ranking for this ratio over total units. Table4 shows the estimated effects for each province
and their statistical significance according to this method.
Only the effects for Badajoz, Huelva and Zamora are statistically significant at the
standard levels. Zamora increases its diesel sales by 20% (but not its gasoline sales), while
Badajoz’s and Huelva’s diesel consumption is up by 7% and 17% respectively.7 These
provinces lie on the main freight transport routes, further suggesting that commercial trucks
are the main channel by which both leakages, from carbon and from revenue, operate.
7 Badajoz’s and Huelva’s lower statistical significance –compared to that of Zamora– is attributable to
the fact that Girona, Gipuzkoa and La Rioja have a higher post- to pre-treatment MSPE ratio. Girona and
Gipuzkoa both border France, where fuel prices are higher. La Rioja is not a border province but it shares
a border with the Basque Country, which had higher fuel prices after the reform and enjoys high mobility
with La Rioja. Likewise, Navarra, also a neighbor of La Rioja saw its regional fuel tax (known as “centimo
sanitario”) increased in January 2019, increasing its own fuel consumption at the expense of La Rioja. All
these circumstances explain why Badajoz and Huelva does not have the highest ranking in its post to pre-
treatment MSPE ratio.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3250
J.J.Teixidó et al.
5 Robustness Checks
In this section, we examine our core estimates through several robustness checks. Firstly,
we investigate whether the reported effects are influenced by the number of filling stations
in the border provinces. Secondly, given the continuous and highly fluctuating nature of
Fig. 7 Synthetic control estimates of fuel sales. Notes: This figure shows the (ln) consumption of diesel
(top panel) and gasoline (bottom panel) in the seven border provinces (solid lines) compared to that of their
counterfactual or synthetic control unit (dashed line), where that province is not impacted by the Portuguese
tax increase of February 2016 (vertical dashed line). The synthetic province is an optimally weighted aver-
age of the other Spanish non-border provinces. The credibility of the causal impact lies in how closely the
synthetic unit resembles the (observed) border province, the effect being the difference between the latter
and the synthetic unit after the tax has been raised
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3251
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
fuel prices, and considering that our identification strategy relies on a binary variable indi-
cating the change in tax regime in Portugal, we assess the robustness of the effect when
focusing on a shorter period, allowing better control over time trends. Lastly, in light of
recent developments in the differences-in-differences literature, we employ two additional
estimators to further validate the robustness of our core results. The first is the doubly
robust estimator, developed by Sant’Anna and Zhao (2020), specifically tailored for differ-
ence-in-differences models with time-varying covariates, as is the case here. Additionally,
we utilize the Synthetic Difference-in-differences estimator, developed by Arkhangelsky
etal. (2021), which combines the features of both difference-in-differences and synthetic
control methods.
5.1 Filling Stations
Having a higher share of filling stations might imply greater exposure to the treatment
and, therefore, account for the bulk of response. To control for this, we replicate the above
difference-in-differences specification by limiting the treatment group to border provinces
with different ranges of filling station shares located close to the Portuguese border. Dif-
ferences with regard to the baseline estimates should inform about the role that this factor
potentially plays.
Table5 shows the number of filling stations in each border province and their distribu-
tion according to the share of filling stations at different distances from the border. Com-
pared to the other Spanish provinces, border provinces do not have a higher number of
filling stations while in per capita terms they present similar magnitudes: border provinces
have 0.27 filling stations per capita on average; the controls, 0.29. In per capita terms, the
Spanish provinces with the most filling stations are Cuenca (0.51), Huesca (0.52), Lleida
(0.43) and Teruel (0.44). Zamora, one of our border provinces, also has 0.43 filling stations
per capita, placing it in the 90th percentile of the distribution. The distribution of these sta-
tions does not reveal a marked concentration near the border with Portugal, which might
indicate that the higher relative number is not driven by its being a border province.
To factor the distribution of filling stations in our empirical strategy, we analyze two
additional samples according to three different exposures to the treatment (i.e., the border).
Table14 in the appendixes shows the observables of these treated and control provinces
Table 5 Number of filling stations (FS) in border Spanish provinces and percentage of stations close to the
Portuguese border
This table shows the number of filling stations (FS) in each province sharing a border with Portugal and the
percentage of FS within a specific distance of the Portuguese border
Province # Filling
stations
#FS per
capita
FS within
5km of
border
FS within
15km of
border
FS within
25km of
border
Badajoz 249 0.36 12 (5%) 41 (16%) 54 (22%)
Cáceres 123 0.30 1 (1%) 4 (3%) 11 (9%)
Huelva 122 0.23 9 (7%) 15 (12%) 26 (21%)
Ourense 90 0.28 2 (2%) 17 (19%) 35 (39%)
Pontevedra 171 0.18 12 (7%) 36 (21%) 71 (42%)
Salamanca 96 0.28 4 (4%) 9 (9%) 12 (13%)
Zamora 78 0.43 1 (1%) 7 (9%) 11 (14%)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3252
J.J.Teixidó et al.
for the three additional samples after the propensity score matching and entropy balancing
have been applied. Conditional on data availability, we define the new treatment as condi-
tional on having more than 20 and 5% of filling stations within the first 25 and 5km of the
border, respectively. As a result, we distinguish three different treatment levels according to
different intensities: the first treatment is the baseline and it considers all (7) border prov-
inces (previous Table3); the second restricts the treatment to provinces with at least 20% of
their filling stations within 25km of the frontier (that is, Pontevedra, Badajoz, Ourense and
Huelva) and the third restricts the treatment to provinces with more than 5% of their filling
stations within 5km of the border (that is, Pontevedra, Badajoz and Huelva). In these last
two treatments, we exclude all the other border provinces. In all cases, entropy balancing
achieves a better balance of the covariates between the treated and control groups.
Table6 shows this effect does not seem to change greatly when we limit the treatment
group in terms of the percentage of filling stations near the border—remaining similar in
magnitude and significance—suggesting that the effect is not driven by the latter. The same
is true for the different treatment specifications estimating responses in terms of the share
of filling stations located at various distances from the border.
5.2 Shorter Time Period
Our core results consider a sampled period that takes from January 2011 to December
2019, hence covering the long-run effects of the February 2016 tax change in Portugal.
However, this is only accurate if we assume that our specification is able to capture
underlying time trends during this period. To verify this assumption, we replicate the
analysis with a shorter period, from January 2015 to December 2017, where the time
trends better account for price variations.
Table7 shows estimates for diesel are only slightly lower than those from longer periods
but within the same range of one or two standard deviations, depending on the specifica-
tion. For gasoline, results become statistically significant when focused on the shorter time
period, indicating that potential short-run effects can also be relevant for gasoline. How-
ever, these are still half of those found for diesel.
5.3 Alternative Identification Methods
Recent developments in econometrics have revealed challenges associated with employing
two-way fixed effects in difference-in-differences (DiD) models. These challenges include
dealing with multiple time periods (Goodman-Bacon 2021; Callaway and Sant’Anna, 2021)
and incorporating time-varying covariates (Sant’Anna and Zhao 2020). Although our two-
way fixed effects model does not involve multiple periods (as there is only one treatment
period after February 2016), it does incorporate time-varying covariates. In doing so, we
are assuming homogenous treatment effects across covariates (i.e., no bad controls) and
that time-varying covariates do not exhibit specific trends in both treated and control
groups. Sant’Anna and Zhao (2020) propose a doubly robust estimator (DR-SZ) based on
pre-treatment characteristics that combine the inverse probability weighting estimator by
Abadie (2005) with the outcome regression approach proposed by Heckman etal. (1997),
both of which address time-varying covariates in a DiD context.
Similarly, synthetic control methods have also evolved to accommodate multiple treated
units, as seen in estimators proposed by Arkhangelsky etal. (2021), Ben-Michael et al.
(2021), or Abadie and L’Hour (2021). Among these, the synthetic difference-in-differences
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3253
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
method (SDID) developed by Arkhangelsky etal. (2021) aims to reconcile both the DiD
and synthetic control methods. Unlike the traditional synthetic control method, the SDID
uses time weights in addition to unit-specific weights. As a result, SDID does not rely on
an exact pre-treatment match between the observed treated unit and the estimated coun-
terfactual but on parallel trends between these two, as in the traditional DiD estimators.
However, unlike traditional difference-in-differences, SDID controls for time-varying
unobserved heterogeneity.
Table8 shows the main parameter estimates using these two alternative methods for
both diesel and gasoline sales. We also show results considering the shorter period as in the
Table 6 Difference-in-difference estimates conditional on filling stations
This table shows the two-way fixed effects difference-in-difference estimator for the different specifications
and samples. Coefficients can be interpreted as the change in the fuel sales of the border provinces as a result
of the tax change in Portugal (ISP). This is shown for provinces with a higher share of filling stations within
the first km after the border (and removing the other border provinces). These coefficients show how the
treatment effect varies in response to an increase in the share of filling stations located close to the border.
Robust standard errors, clustered at the province level, in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1
(1) (2) (3) (4) (5) (6)
ln(Diesel sales)
Border ×ISP
(25km from border) 0.063* 0.079*** 0.071***
(0.032) (0.021) (0.022)
Border ×ISP
(5km from border) 0.080** 0.085*** 0.096***
(0.033) (0.013) (0.014)
Constant 9.164 9.910 25.793* 31.804** 37.620*** 34.413***
(14.824) (14.795) (13.597) (14.740) (12.794) (11.739)
R-squared 0.756 0.759 0.829 0.855 0.985 0.985
R2 adj 0.743 0.746 0.785 0.810 0.984 0.984
ln(Gasoline sales)
Border ×ISP
(25km from border) − 0.016* 0.013 − 0.012
(0.008) (0.016) (0.007)
Border ×ISP
(5km from border) − 0.010 − 0.036** − 0.013
(0.009) (0.013) (0.012)
Constant 7.763 9.568 − 9.930 17.064 15.311 13.417**
(11.386) (11.412) (13.179) (12.706) (11.222) (6.372)
R-squared 0.810 0.807 0.856 0.827 0.983 0.986
R2 adj 0.800 0.797 0.820 0.772 0.981 0.985
Observations 4,320 4,212 864 648 4,320 4,212
Number of id_province 40 39 26 17 40 39
Sample All All PS match PS match Entropy B Entropy B
Control vars YES YES YES YES YES YES
Province FE YES YES YES YES YES YES
Year FE YES YES YES YES YES YES
Month FE YES YES YES YES YES YES
Year-month FE YES YES YES YES YES YES
Year x Region FE YES YES YES YES YES YES
Cluster s.e Province Province Province Province Province Province
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3254
J.J.Teixidó et al.
previous robustness check. Results remain consistent in all cases: diesel sales increase as a
result of the tax increase in Portugal, while gasoline sales do not. According to both DR-SZ
and SDID, gasoline sales do not significantly increase at any standard level of significance,
even when restricting the sample to the shorter period. This is particularly relevant for the
Table 7 Difference-in-difference estimates (2015–2017)
This table shows the two-way fixed effects difference-in-difference estimator for the different specifications
and samples. Coefficients can be interpreted as the change in the fuel consumption of the border provinces
as a result of the tax change in Portugal (ISP). Robust standard errors in parentheses. ***p < 0.01,
**p < 0.05, *p < 0.1
(1) (2) (3) (4) (5) (6)
VARIABLES ln(Diesel) ln(Diesel) ln(Diesel) ln(Gasoline) ln(Gasoline) ln(Gasoline)
Border ×ISP
0.068*** 0.058*** 0.069*** 0.033*** 0.035*** 0.035***
(0.013) (0.015) (0.012) (0.012) (0.012) (0.009)
Constant 37.325** 36.232 47.207** 29.671* 49.599 39.076**
(15.874) (27.424) (19.200) (15.770) (34.043) (18.583)
R-squared 0.801 0.861 0.992 0.832 0.854 0.988
R2 adj 0.793 0.841 0.992 0.825 0.832 0.987
Observations 1,548 486 1,548 1,548 486 1,548
Number of provincies 43 34 43 43 34 43
Sample All PS match Entropy B All PS match Entropy B
Control vars YES YES YES YES YES YES
Province FE YES YES YES YES YES YES
Year FE YES YES YES YES YES YES
Month FE YES YES YES YES YES YES
cluster s.e Province Province Province Province Province Province
Table 8 Difference-in-differences
according to alternative
estimators
This Table shows estimates for alternative relevant dif-in-dif
estimators. First column shows doubly robust (DR-SZ) estimator
proposed by Sant’Anna and Zhao (2020). The second column
shows synthetic difference-in-difference estimator, proposed by
Arkhangelsky et al. (2021). Robust standard errors in parentheses.
***p < 0.01, **p < 0.05, *p < 0.1
DR-SZ SDID
ln(diesel consumption)
Border ×ISP
(2011–2019) 0.061* 0.064***
(0.035) (0.025)
Border ×ISP
(2015–2017) 0.049** 0.061***
(0.019) (0.015)
ln(gasoline consumption)
Border ×ISP
(2011–2019) 0.036 − 0.01
(0.067) (0.01)
Border ×ISP
(2015–2017) − 0.001 0.005
(0.039) (0.017)
Obs 4644 4644
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3255
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
SDID estimator, which, as mentioned, can also control for time-varying unobserved fac-
tors. Figure11 in the appendixes illustrates the observed border provinces as compared to
the estimated counterfactual.
6 Discussion andPolicy Implications
Our results highlight a novel differential effect by fuel type. Only diesel consumption
appears, to react (and then very robustly) to the cross-border tax change, in marked contrast
that is with gasoline consumption. Specifically, our estimates show a 6–9% rise in diesel
sales in those Spanish provinces that share a border with Portugal in response to the tax
hike in that country. This represents additional annual consumption of 21–30 million liters
of diesel and implies a cross elasticity of demand (for Spanish fuel consumption) with
respect to the Portuguese tax change that is roughly nine times greater for diesel than for
gasoline: 1.2–1.8 for diesel vs. 0.1–0.3 for gasoline.8
We attribute this differential effect to drivers of heavy goods vehicles reacting to tax
changes—given their large capacity diesel fuel tanks– while the drivers of passenger cars
use gasoline and diesel-fueled cars in similar proportions in the two countries (40 and 60%,
respectively). As mentioned, this does not necessarily imply that the drivers of passenger
cars do not take advantage of Spain’s lower prices. Simply, our empirical models are una-
ble to identify this. However, it does imply that such behavior does not result in increased
fuel consumption because of the cross-border tax change.
The absence of a reaction from passenger car drivers can, potentially, be explained by
the fact that cross-border fuel substitution may well have reached satiation, i.e., no increase
in consumption results from the tax change because all drivers that engage in cross-border
fuel substitution are already engaging in it. Additionally, or alternatively, car drivers are
only sensitive to changes in the price at the pump, which are certainly less dramatic in our
experimental setting than changes in the tax rate. Whatever the case, these drivers appear
to be inelastic to cross-border tax changes, unlike truck drivers. A potential explanation for
this is that the latter probably equip themselves better to track price changes using differ-
ent navigation tools and applications, given that the potential savings are huge when filling
their massive tanks. This makes these drivers more elastic to cross-border tax changes. Yet,
this does not fully align with a greater response to tax changes because of the higher sali-
ence or persistence of the tax. If this were the case, gasoline sales should have reacted just
as strongly.
The Portuguese tax on petroleum and energy products (ISP) was environmentally
motivated, insofar as it sought to “promote low-carbon economy and fight climate change”.
Yet, despite these intentions, our results indicate a carbon leakage of 55,000 to 80,000 tCO2
per year,9 attributable exclusively to the consumption of diesel. While this is only 0.5% of
8 In the case of diesel, the cross-price elasticity is derived from the 6–9% increase in demand for diesel in
the Spanish border provinces divided by the 5% overall increase in diesel taxes in Portugal (that is, €0.06
of the tax increase over €0.95, the average full tax levied on diesel). In the case of gasoline, this is a non-
statistically significant demand increase of 1% over the 10% increase in the tax (that is, €0.06 of the average
full tax levied on gasoline €0.60). The highest value of the estimate yields a short-term cross-price elasticity
of 0.3.
9 The total of million liters of diesel is derived from the difference between observed diesel consumption
and the counterfactual liters consumed (without the cross-border tax according to our estimates). We cal-
culate these figures taking our lower estimate (6%) and our higher estimate (9%). In the case of CO2 emis-
sions, we apply a conversion factor of 2.68 kg of CO2 for each liter of diesel consumed.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3256
J.J.Teixidó et al.
Portugal’s total annual transport emissions, it represents 14–20% of the country’s annual
mitigation commitment by 2030 (NECP-Portugal 2019). According to the National Energy
and Climate Plan 2021–2030, projected emissions for transport by 2030 are 11.7M tCO2,
while in 2019 they were registered at 17M tCO2. This means 4.3 M tCO2 mitigated in
11years, hence 397,367 tCO2 per annum: 80,000 tCO2/397,367 tCO2 (i.e. 20%). However,
far from being mitigated, these GHG emissions are simply being transferred to Spain,
with obvious consequences for this country’s mitigation objectives and strategies. The
transport sector is Spain’s main CO2 emitter and freight transport is responsible for 25%
of these emissions. Moreover, Spain faces an above-average fuel consumption compared
to the EU due, among other reasons, to the fact that it opted to develop its road freight
transport to the detriment of rail alternatives (NECP-Spain 2020). In this context, fuel
tax harmonization would mitigate emission leakage from Portugal and also help Spain to
reduce its overabundant fuel consumption.
In the context of carbon pricing, policies aimed at mitigating leakage theoreti-
cally encompass a range of measures, from carbon border adjustments to various forms
of subsidy and exemption, such as free allowances and export rebates (see, for example,
Böhringer et al. 2017; Kortum and Weisbach 2017). Fowlie and Reguant (2021) advo-
cate output-based subsidies for sectors deemed highly vulnerable to carbon leakage, even
though such subsidies might attenuate incentives to abate domestic emissions. Neverthe-
less, the reduction in emission leakage significantly outweighs the reduction in domestic
abatement incentives. In the particular context of the EU, harmonizing fuel taxes alone
could reduce within-EU leakage attributable to freight transportation; at the same time, for
freight transportation to and from non-EU countries, additional leakage mitigation policies
would be indispensable, especially with the forthcoming implementation of the EU’s new
road transport emission trading system.
Finally, on the revenue side, if we consider the total tax rate levied on diesel fuel in Por-
tugal (€0.71 per liter being the average during the post-treatment period), the carbon leak-
age documented herein implies an annual foregone revenue of €21 million, that is, 1.3% of
the total revenue generated by diesel fuel taxation in Portugal in 2016 (European Commis-
sion, 2023). The future EU carbon market for transport must take steps to mitigate carbon
leakage, albeit if only within the EU, and provided that it is accompanied by the simultane-
ous harmonization of fuel taxation, especially for diesel fuel and for freight transport.
In order to meet the climate targets set for the next few decades, the reduction in CO2
emissions is becoming more and more pressing and mitigation policies need not only be as
effective but also as efficient as possible. As of today, freight transport—demand for which
is subject to constant increases (IEA 2023)—accounts for 27% of road transport emissions
in the EU (EEA, 2022); however, it accounts for less than 1.7% of the vehicle fleet (ACEA
2022). Hence, while the internal combustion engine continues to make up the lion’s share
of freight transport, targeting this sector appears appropriate from an efficiency perspec-
tive and may justify the adoption of stringent ad hoc approaches, especially given its high
carbon leakage risk.
7 Conclusion
Reducing the GHG emissions of the transportation sector is critical to achieving the
climate targets that have been set by most developed countries, especially the climate
neutrality objectives fixed for the 2050 horizon. Yet, the socio-economic importance of
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3257
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
this sector has precluded progress to date. Indeed, in marked contrast with the significant
advances made in other activities, in the transport sector policies have failed to reduce
GHG emissions below 1990 levels. In many developed countries, transport, today, is the
main GHG emitter and, thus, there is a significant gap between this reality and the urgency
of climate mitigation and the implementation of effective measures. In this sense, carbon
pricing –the favored policy approach– has been environmentally relevant in no more than
a handful of countries and significant progress is still awaited in this area. However, given
the mobility of the transport sector, pricing instruments of this kind are exposed to the
risk of carbon leakage. As is well documented in the empirical literature, cross-border fuel
substitution in countries that share borders but not fuel price levels has become common
practice.
This paper has shown empirically that climate policies based on pricing instruments
implemented in the road transportation sector can result in carbon leakage and foregone
revenue, thereby significantly undermining both the environmental effectiveness and
economic efficiency of such measures. We provide robust causal evidence that a rise in
Portugal’s fuel tax aimed at environmental goals increased diesel sales (and derived
emissions) in neighboring border provinces in Spain, providing evidence of notable
carbon leakage (i.e., emissions shifted from Portugal to Spain, while recorded as emission
reduction in Portugal). Critically, our results are also robust in reporting a non-statistically
significant effect in the case of gasoline sales, revealing a novel differential effect by fuel
type. This differential effect is attributed to different elasticities to cross-border tax changes
between heavy goods vehicles (which predominantly use diesel) and passenger vehicles
(which use both diesel and gasoline). The higher elasticity of truck drivers, equipped with
large-capacity diesel fuel tanks, drives the cross-border tax response.
Previous research has shown the key roles of salience and persistence of fuel taxes in
shaping drivers’ responses. Here we show fuel type might be as relevant. Additionally,
we find that a lack of tax coordination across countries undermines mitigation policies in
the transportation sector. These findings offer valuable insights for future climate policies,
particularly by identifying road freight transport as the primary source of carbon leakage
within the sector. As the implementation of an emission trading system for transport emis-
sions approaches, enhanced tax coordination among Member States is essential to avoid
the negative outcomes identified in this paper.
Appendix
See Tables9, 10, 11, 12, 13, 14 and Figs.8, 9, 10, 11.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3258
J.J.Teixidó et al.
Table 9 Actual and synthetic predictor means for the period prior to the tax change
This table shows the mean values of the predictors used to estimate the counterfactual scenario (i.e. the synthetic unit). Here we show values for diesel consumption. The last
column shows the sample averages of the donor group to facilitate comparison with the optimally weighted averages in the synthetic units. (BADA. Badajoz; CAC: Cáceres,
HUE: Huelva; OUR: Ourense; PONT: Pontevedra; SAL: Salamanca; ZAM: Zamora)
BADA Synt CAC Synt HUE Synt OUR Synt PONT Synt SAL Synt ZAM Synt Avg. control
Group
ln(income) 6.53 6.69 6.54 6.62 6.55 6.71 6.75 6.81 6.75 6.73 6.79 6.83 6.80 6.83 6.74
% inhab. in capital city 0.22 0.28 0.23 0.27 0.28 0.34 0.33 0.31 0.09 0.26 0.43 0.43 0.34 0.40 0.32
ln(Filling S. per capita) − 1.08 − 1.33 − 1.26 − 1.05 − 1.53 − 0.99 − 1.25 − 1.14 − 1.71 − 1.57 − 1.30 − 1.26 − 0.89 − 0.99 − 1.37
ln(price diesel, 2016) 0.24 0.22 0.26 0.25 0.26 0.22 0.26 0.25 0.26 0.26 0.24 0.26 0.25 0.24 0.24
ln(diesel liters, Oct2011) 17.51 17.51 16.89 16.87 17.03 17.06 16.66 16.69 17.68 17.70 17.25 17.24 16.69 16.68 17.35
ln(diesel liters, Aug2012) 17.53 17.52 16.95 16.93 17.17 17.17 16.89 16.89 17.74 17.76 17.13 17.12 16.68 16.66 17.36
ln(diesel liters, Jun2013) 17.26 17.33 16.66 16.67 16.94 16.99 16.66 16.69 17.59 17.60 16.71 16.84 16.36 16.37 17.21
ln(diesel liters, Apr2014) 17.43 17.42 16.76 16.74 17.11 17.07 16.62 16.63 17.55 17.57 16.90 16.90 16.46 16.44 17.25
ln(diesel liters, Jan2016) 17.37 17.36 16.63 16.62 16.93 16.95 16.46 16.48 17.44 17.47 17.00 16.94 16.52 16.49 17.19
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3259
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Table 10 Synthetic control weight per border province (diesel consumption)
Zamora Huelva Badajoz Salamanca Ourense Pontevedra Cáceres
Álava 0 0 0 0 0 0 0
Albacete 0 0 0 0 0 0 0
Alicante 0 0 0 0 0 0 0
Almería 0 0.155 0 0 0 0 0
Ávila 0 0 0 0 0 0 0
Badajoz – – – – – – –
Balears (illes) 0 0 0 0 0.32 0.22 0
Barcelona 0.078 0 0.259 0.246 0 0 0
Burgos 0 0 0.011 0.126 0 0.099 0.092
Cáceres – – – – – –
Cádiz 0 0 0 0 0 0.128 0.235
Castelló 0 0 0 0 0 0 0
Ciudad Real 0 0 0 0 0 0 0
Córdoba 0 0 0 0 0 0 0
Coruña (A) 0 0 0 0 0 0 0
Cuenca 0 0.392 0.271 0 0.164 0 0.47
Girona 0 0 0 0 0 0 0
Granada 0 0 0 0 0 0 0
Guadalajara 0 0 0 0 0 0 0
Gipuzkoa 0 0 0 0 0 0 0
Huelva – – – – – – –
Huesca 0.248 0.19 0 0 0.072 0 0
Jaén 0 0.058 0.207 0 0 0 0.044
León 0 0 0.065 0 0 0 0
Lleida 0 0 0 0 0 0 0
Rioja (La) 0 0 0 0 0 0 0
Lugo 0 0 0 0 0 0 0
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3260
J.J.Teixidó et al.
Note: This table shows the optimal weights for estimating each synthetic control unit for diesel consumption
Table 10 (continued)
Zamora Huelva Badajoz Salamanca Ourense Pontevedra Cáceres
Madrid 0 0 0 0 0 0.096 0
Málaga 0 0 0 0 0 0 0
Murcia 0 0 0 0 0 0 0
Navarra 000 0 000
Ourense – – – – – – –
Asturias 0 0 0 0 0 0.093 0
Palencia 0.622 0 0 0.628 0 0 0
Palmas (Las) 0 0.046 0.121 0 0 0 0
Pontevedra – – – – – – –
Salamanca – – – – – – –
S.C. Tenerife 0 0 0 0 0 0 0
Cantabria 0 0 0 0 0 0 0
Segovia 0 0 0 0 0 0 0
Sevilla 0 0 0 0 0 0 0
Soria 0.051 0.001 0.066 0 0.146 0 0.158
Tarragona 0 0 0 0 0 0 0
Teruel 0 0 0 0 0.299 0 0
Toledo 0 0 0 0 0 0.339 0
València 0 0 0 0 0 0 0
Valladolid 0 0 0 0 0 0 0
Bizkaia 0 0 0 0 0 0.026 0
Zamora – – – – – – –
Zaragoza 0 0.158 0 0 0 0 0
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3261
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Table 11 Synthetic control weight per border province (gasoline consumption)
Zamora Huelva Badajoz Salamanca Ourense Pontevedra Cáceres
Álava 0 0 0 0 0 0 0
Albacete 0 0 0 0 0 0 0
Alicante 0 0 0 0 0 0 0
Almería 0 0 0 0 0 0 0
Ávila 0 0.081 0 0 0 0 0
Badajoz – – – – – – –
Balears (illes) 0 0.059 0 0 0.268 0.463 0
Barcelona 0 0 0 0 0 0.016 0
Burgos 0 0 0 0.084 0 0 0
Cáceres – – – – – – –
Cádiz 0 0.208 0.132 0 0 0 0.246
Castelló 0 0 0 0 0 0 0
Ciudad Real 0 0 0 0 0 0 0
Córdoba 0 0 0 0 0 0 0
Coruña (A) 0 0 0 0 0 0 0
Cuenca 0.47 0.226 0 0 0 0 0.164
Girona 0 0 0 0.029 0.056 0 0
Granada 0 0 0 0 0 0 0
Guadalajara 0 0.143 0 0 0 0 0.211
Gipuzkoa 0 0 0 0 0 0 0
Huelva – – – – – – –
Huesca 0 0 0 0 0 0 0
Jaén 0 0 0.147 0 0 0 0
León 0 0 0 0 0 0 0
Lleida 0.159 0 0.154 0 0 0 0.01
Rioja (La) 0 0.047 0 0 0 0 0
Lugo 0 0 0 0 0 0 0
Madrid 0 0 0 0 0 0.093 0
Málaga 0 0.183 0 0 0 0 0
Murcia 0 0 0 0 0 0 0
Navarra 0 0 0 0 0 0 0
Ourense – – – – – – –
Asturias 0 0 0 0 0 0 0
Palencia 0.213 0 0 0.251 0 0 0
Palmas (Las) 0 0 0.008 0 0 0 0
Pontevedra – – – – – – –
Salamanca – – – – – – –
S.C. Tenerife 0 0 0 0 0 0 0
Cantabria 0 0 0 0.451 0 0 0
Segovia 0.015 0 0.182 0.14 0.132 0.386 0
Sevilla 0 0 0.183 0 0 0 0
Soria 0.142 0 0 0.046 0.544 0 0
Tarragona 0 0 0 0 0 0 0
Teruel 0 0 0 0 0 0 0.17
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3262
J.J.Teixidó et al.
This table shows the optimal weights for estimating each synthetic control unit for gasoline consumption
Table 11 (continued)
Zamora Huelva Badajoz Salamanca Ourense Pontevedra Cáceres
Toledo 0 0.054 0.194 0 0 0.041 0.2
València 0 0 0 0 0 0 0
Valladolid 0 0 0 0 0 0 0
Bizkaia 0 0 0 0 0 0 0
Zamora – – – – – – –
Zaragoza 0 0 0 0 0 0 0
Table 12 Difference-in-
difference estimates with no
control covariates
Robust standard errors in parentheses
*** p < 0.01, **p < 0.05, *p < 0.1
(1) (2)
VARIABLES ln(Diesel) ln(g95)
Border ×ISP
0.079** − 0.029***
(0.032) (0.010)
Constant 17.229*** 15.646***
(0.017) (0.030)
Observations 5184 5184
R-squared 0.683 0.767
R2 adj 0.668 0.756
Number of id_province 48 48
Sample All All
Control vars NO NO
Province FE YES YES
Year FE YES YES
Month FE YES YES
Year-month FE YES YES
Year × Region FE YES YES
cluster s.e Province Province
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3263
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Table 13 In-time placebo for difference-in-difference estimates
Post-treatment data are removed to avoid confounding the placebo treatments. Robust standard errors in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1
VARIABLES (1) (2) (3) (4) (5) (6) (7) (8)
ln(Diesel) ln(Diesel) ln(Diesel) ln(Diesel) ln(g95) ln(g95) ln(g95) ln(g95)
Placebo DiD (Feb 2015) 0.008 − 0.026
(0.020) (0.019)
Placebo DiD (Feb 2014) − 0.000 − 0.025
(0.019) (0.018)
Placebo DiD (Feb 2013) − 0.015 − 0.023
(0.020) (0.018)
Placebo DiD (Feb 2012) − 0.023 − 0.013
(0.019) (0.019)
Constant 33.216* 33.697* 35.795** 35.668** 14.297*** 14.281*** 14.271*** 14.263***
(16.932) (17.183) (16.587) (16.591) (0.553) (0.556) (0.561) (0.564)
Observations 2,666 2,666 2,666 2,666 2,666 2,666 2,666 2,666
R-squared 0.987 0.987 0.987 0.987 0.988 0.988 0.988 0.988
R2 adj 0.986 0.986 0.986 0.986 0.987 0.987 0.987 0.987
Sample Entropy B Entropy B Entropy B Entropy B Entropy B Entropy B Entropy B Entropy B
Control vars YES YES YES YES YES YES YES YES
Province FE YES YES YES YES YES YES YES YES
Year FE YES YES YES YES YES YES YES YES
Month FE YES YES YES YES YES YES YES YES
cluster s.e Province Province Province Province Province Province Province Province
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3264
J.J.Teixidó et al.
Table 14 Treated and control Spanish provinces. Alternative matched sample
Mean values for year 2015, the year prior to the tax change in Portugal
Propensity Score Matching (PSM)
sample
Entropy balanced sample
T = 0 T = 1 Diff T = 0 T = 1 Diff
20% filling stations closer than 25km
(Pontevedra, Badajoz, Ourense and Huelva)
ln (income) 6.606 6.634 − 0.028 6.641 6.634 0.007
Share inhab. in Prov. Capital 0.222 0.230 − 0.008 0.235 0.23 0.005
ln(population) 13.427 13.258 − 0.169* 13.225 13.258 − 0.033
ln(before-tax price of diesel, 2016) 0.126 0.130 − 0.004 0.127 0.13 − 0.003
ln(before-tax price of gasoline, 2016) 0.217 0.224 − 0.006 0.221 0.224 − 0.003
ln(avg. daily traffic) 9.385 9.215 − 0.169** 9.187 9.215 − 0.028
5% filling stations closer than 5km
(Pontevedra, Badajoz and Huelva)
ln (income) 6.535 6.598 − 0.062*** 6.571 6.598 − 0.027
Share inhab. in Prov. Capital 0.248 0.195 0.052*** 0.191 0.195 − 0.004
ln(population) 13.4 13.455 − 0.054 13.508 13.455 0.053
ln(before-tax price of diesel, 2016) 0.101 0.125 − 0.025* 0.117 0.126 − 0.009
ln(before-tax price of gasoline, 2016) 0.194 0.22 − 0.026** 0.209 0.220 − 0.011
ln(avg. daily traffic) 9.511 9.395 0.116 9.436 9.395 0.041
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3265
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Fig. 8 Fuel price evolution in treated provinces vs. immediate neighboring provinces.Notes: Each graph
plots the evolution in diesel (top panel) and gasoline prices (bottom panel) for a specific treated province
(solid line) vs. the other Spanish provinces with which it shares a border (dashed lines). When a dashed
line rises above the main solid line, this means that the treated province shares a border with a province that
charges higher fuel prices. This can confound our identification strategy only when that price difference
coincides with the vertical line (February 2016: tax change in Portugal)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3266
J.J.Teixidó et al.
Fig. 9 Rambachan and Roth (2023) sensitivity tests on parallel trends. Notes: The top panel replicates event
estimates for difference-in-differences on an annual basis for diesel sales. The bottom panel plots sensitiv-
ity tests using smoothness restrictions (Rambachan and Roth 2023). The original estimate for each year is
shown in red (plotted in the top panel). Estimates for different M values are shown in blue
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3267
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Fig. 10 In-space placebo tests. Notes: These graphs show the ratio of post- to pre-treatment MSPE allowing
inferences to be made by comparing each unit with its synthetic control. In this case, only Zamora, and to
a lower extent Badajoz and Huelva, show high RMSPE ratios. Following Abadie etal. (2015), this empiri-
cal distribution of ratios can be used to calculate p-values as the probability of obtaining as large a ratio
if these ratios were randomly assigned. For Zamora the value is 1/50 = 0.02, for Badajoz and Huelva it is
4/46 = 0.08
Fig. 11 Synthetic Difference-in-differences. Notes: This figure shows the average (ln)consumption of die-
sel (top panels) and gasoline (bottom panels) in the seven border provinces (solid lines) compared to that
of their counterfactual (dashed line) with no tax cross border-tax increase, as estimated by the synthetic
Difference-in-differences (Arkhangelsky etal 2021). This is shown for two different sampled periods. The
credibility of the causal impact lies on parallel trends between observed border provinces and the synthetic
border province. The synthetic border provinces are an optimally weighted average of the other Spanish
non-border provinces. Green areas signal the time weights
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3268
J.J.Teixidó et al.
Acknowledgements Financial support from the Ministerio de Ciencia e Innovación/Agencia Estatal de
Investigación TED2021-130638A-I00, PID2022-138866OB-I00 and PID2019-1055517RB-I00 is gratefully
acknowledged. We are also very thankful to participants at the Workshop on “Effectiveness and Distribu-
tional Impacts of Environmental Policy” at the Berlin School of Economics, the Economic Science Associa-
tion (ESA) World Conference 2022 in Boston; the X Conference of the Spanish-Portuguese Association of
Natural and Environmental Resource Economics (AERNA), XI International Academic Symposium (IEB-
Chair of Energy Sustainability) and the III Catalan Economic Society Conference (CESC) for their valuable
questions and comments.
Funding Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature.
Declarations
Conflict of interest No interests to disclose.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Com-
mons licence, and indicate if changes were made. The images or other third party material in this article
are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
Abadie A (2021) Using synthetic controls: feasibility, data requirements, and methodological aspects. J
Econ Lit 59(2):391–425. https:// doi. org/ 10. 1257/ jel. 20191 450
Abadie A, Gardeazabal J (2003) The economic costs of conflict: a case study of the Basque Country. Am
Econ Rev 93(1):113–132. https:// doi. org/ 10. 1257/ 00028 28033 21455 188
Abadie A, Imbens GW (2011) Bias-corrected matching estimators for average treatment effects. J Bus Econ
Stat 29(1):1–11. https:// doi. org/ 10. 1198/ jbes. 2009. 07333
Abadie A, Diamond A, Hainmueller J (2010) Synthetic control methods for comparative case studies:
estimating the effect of California’s Tobacco Control Program. J Am Stat Assoc 105(490):493–505.
https:// doi. org/ 10. 1198/ jasa. 2009. ap087 46
Abadie A, Diamond A, Hainmueller J (2015) Comparative politics and the synthetic control method. Amer-
ican Journal of Political Science 59:495–510. https:// doi. org/ 10. 1111/ ajps. 12116
ACEA (2022) Vehicles in use Europe 2022. https:// www. acea. auto/ publi cation/ report- vehic
les- in- use- europe- 2022/
Agaku IT, Blecher E, Filippidis FT, Omaduvie UT, Vozikis A, Vardavas CI (2016) Impact of cigarette
price differences across the entire European Union on cross-border purchase of tobacco products
among adult cigarette smokers. Tob Control 25(3):333–340. https:// doi. org/ 10. 1136/ tobac cocon
trol- 2014- 052015
Andersson JJ (2019) Carbon taxes and CO2 emissions: Sweden as a case study. Am Econ J Econ Pol
11(4):1–30. https:// doi. org/ 10. 1257/ pol. 20170 144
Antweiler W, Gulati S (2016) Frugal cars or frugal drivers? How Carbon Fuel Taxes Influence Choice Use
Cars. https:// doi. org/ 10. 2139/ ssrn. 27788 68
Arkhangelsky D, Athey S, Hirshberg DA, Imbens GW, Wager S (2021) Synthetic difference-in-differences.
Am Econ Rev 111(12):4088–4118
Banfi S, Filippini M, Hunt LC (2005) Fuel tourism in border regions: the case of Switzerland. Energy Econ
27(5):689–707. https:// doi. org/ 10. 1016/j. eneco. 2005. 04. 006
Bang H, Robins JM (2005) Doubly robust estimation in missing data and causal inference models. Biomet-
rics 61(4):962–972
Ben-Michael E, Feller A, Rothstein J (2021) The augmented synthetic control method. J Am Stat Assoc
116(536):1789–1803. https:// doi. org/ 10. 1080/ 01621 459. 2021. 19292 45
Böhringer C, Rosendahl KE, Storrøsten HB (2017) Robust policies to mitigate carbon leakage. J Public
Econ 149:35–46
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3269
Carbon Leakage fromFuel Taxes: Evidence fromaNatural…
Callaway B, Sant’Anna PHC (2021) Difference-in-differences with multiple time periods. J Econom
225(2):200–230. https:// doi. org/ 10. 1016/j. jecon om. 2020. 12. 001
Coglianese J, Davis LW, Kilian L, Stock JH (2017) Anticipation, tax avoidance, and the price elasticity of
gasoline demand. J Appl Econom 32(1):1–15. https:// doi. org/ 10. 1002/ jae. 2500
European Commission, & Directorate-General Taxaxion and Customs Union. (2023) Excise Duty Tables
(Tax receipts—Energy products and Electricity). Revenues from Taxes on Consumption.
Coyne D (2017) How Political boundaries affect gas price competition and State Motor Fuels Tax. In: Pro-
ceedings of the Annual Conference on Taxation and Minutes of the Annual Meeting of the National
Tax Association, vol 110, pp 1–51. https:// www. jstor. org/ stable/ 26794 453
CNMC Data (2021) Comisión Nacional de Mercados y Competencia. Estadística de Productos Petrolíferos.
Retrieved Ferbuary 2021 from https:// data. cnmc. es/
Davis LW, Kilian L (2011) Estimating the effect of a gasoline tax on carbon emissions. J Appl Economet
26(7):1187–1214. https:// doi. org/ 10. 1002/ jae. 1156
DeCicca P, Kenkel D, Liu F (2013) Who pays cigarette taxes? The impact of consumer price search. Rev
Econ Stat 95(2):516–529
DGT. Direccion General de Tráfico (2023) Datos estaciones. Dirección General de Tráfico. Retrieved Octo-
ber 2023 from https:// nap. dgt. es/ datas et
EEA-European Environment Agency (2022) Decarbonising road transport: the role of vehicles, fuels and
transport demand. Publications Office of the European Union. https:// books. google. es/ books? id=
Hf9Oz wEACA AJ
EPA (2023) Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990–2021. https:// www. epa. gov/
ghgem issio ns/ sourc es- green house- gas- emiss ions
Fowlie ML, Reguant M (2021) Mitigating emissions leakage in incomplete carbon markets. J Assoc Environ
Resour Econ 9(2):307–343. https:// doi. org/ 10. 1086/ 716765
Gago A, Labandeira X, López-Otero X (2014) A panorama on energy taxes and green tax reforms. Rev
Public Econ 208:145–190. https:// doi. org/ 10. 7866/ HPE- RPE. 14.1.5
Ghoddusi H, Morovati M, Rafizadeh N (2022) Dynamics of fuel demand elasticity: evidence from Iranian
subsidy reforms. Energy Econ. https:// doi. org/ 10. 1016/j. eneco. 2022. 106009
Goodman-Bacon A (2021) Difference-in-differences with variation in treatment timing. J Econom
225(2):254–277. https:// doi. org/ 10. 1016/j. jecon om. 2021. 03. 014
Hainmueller J (2012) Entropy balancing for causal effects: a multivariate reweighting method to produce
balanced samples in observational studies. Polit Anal 20(1):25–46. https:// doi. org/ 10. 1093/ pan/ mpr025
Harding M, Leibtag E, Lovenheim MF (2012) The Heterogeneous geographic and socioeconomic incidence
of cigarette taxes: evidence from Nielsen Homescan Data. Am Econ J Econ Pol 4(4):169–198. https://
doi. org/ 10. 1257/ pol.4. 4. 169
IEA (2023) Energy Technology Perspectives 2023. https:// www. iea. org/ repor ts/ energy- techn ology- persp
ectiv es- 2023
INE (2021). INEbase. Atlas Distribucion de renta. Retrieved Ferbuary 2021 from https:// www. ine. es/ dyngs/
INEba se/ lista opera ciones. htm
Jansen D-J, Jonker N (2018) Fuel Tourism in Dutch border regions: Are only salient price differentials rel-
evant? Energy Econ 74:143–153. https:// doi. org/ 10. 1016/j. eneco. 2018. 05. 036
Kortum S, Weisbach D (2017) The design of border adjustments for carbon prices. Natl Tax J 70(2):421–446
Leal A, López-Laborda J, Rodrigo F (2009) Prices, taxes and automotive fuel cross-border shopping.
Energy Econ 31(2):225–234. https:// doi. org/ 10. 1016/j. eneco. 2008. 09. 007
Li S, Linn J, Muehlegger E (2014) Gasoline taxes and consumer behavior. Am Econ J Econ Pol 6(4):302–
342. https:// doi. org/ 10. 1257/ pol.6. 4. 302
Lovenheim MF (2008) How far to the border? The extent and impact of cross-border casual cigarette smug-
gling. Natl Tax J 61(1):7–33. https:// doi. org/ 10. 17310/ ntj. 2008.1. 01
Manuszak MD, Moul CC (2009) How far for a buck? Tax differences and the location of retail gasoline
activity in Southeast Chicago land. Rev Econ Stat 91(4):744–765
Marion J, Muehlegger E (2018) Tax compliance and fiscal externalities: evidence from U. S. Diesel Taxa-
tion. J Public Econ 160(C):1–13. https:// doi. org/ 10. 1016/j. jpube co. 2018. 02. 007
Ministerio para la Transición Ecológica y el RD (2021) Geoportal Gasolineras. Retrieved February, 2021,
from: https:// geopo rtalg asoli neras. es/ geopo rtal- insta lacio nes/ Inicio
Morton C, Lovelace R, Philips I, Anable J (2018) Fuel price differentials and car ownership: a spatial analy-
sis of diesel cars in Northern Ireland. Transp Res D: Transp Environ 63:755–768. https:// doi. org/ 10.
1016/j. trd. 2018. 07. 008
Naegele H, Zaklan A (2019) Does the EU ETS cause carbon leakage in European manufacturing? J Environ
Econ Manag 93:125–147. https:// doi. org/ 10. 1016/j. jeem. 2018. 11. 004
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

3270
J.J.Teixidó et al.
NECP-Portugal (2019) National Energy and Climate Plan 2021–2030. Portugal. https:// energy. ec. europa. eu/
system/ files/ 2020- 06/ pt_ final_ necp_ main_ en_0. pdf
NECP-Spain (2020) Integrated National Energy and Climate Plan 2021–2030. https:// energy. ec. europa. eu/
system/ files/ 2020- 06/ es_ final_ necp_ main_ en_0. pdf
OTEP (2020) Observatorio Transfronterizo España/Portugal. Documento 9. Secretaría General de Trans-
portes y Movilidad Ministerio de Transportes, Movilidad y Agenda Urbana / Ministério da Economia
(Portugal). https:// www. mitma. gob. es/ recur sos_ mfom/ lista do/ recur sos/ obser vator io_ otep_ no_9_ esp5.
pdf
Rambachan A, Roth J (2023) A more credible approach to parallel trends. Rev Econ Stud 90(5):2555–2591.
https:// doi. org/ 10. 1093/ restud/ rdad0 18
Roth J (2022) Pretest with caution: event-study estimates after testing for parallel trends. Am Econ Rev:
Insights 4:305–322
Sant’Anna PHC, Zhao J (2020) Doubly robust difference-in-differences estimators. J Econom 219(1):101–
122. https:// doi. org/ 10. 1016/j. jecon om. 2020. 06. 003
Scott KR (2012) Rational habits in gasoline demand. Energy Econ 34(5):1713–1723. https:// doi. org/ 10.
1016/j. eneco. 2012. 02. 007
Sterner T (2007) Fuel taxes: an Important Instrument for Climate Policy. Energy Policy 35(6):3194–3202.
https:// doi. org/ 10. 1016/j. enpol. 2006. 10. 025
Tiezzi S, Verde SF (2016) Differential demand response to gasoline taxes and gasoline prices in the U.S.
Resour Energy Econ 44:71–91. https:// doi. org/ 10. 1016/j. resen eeco. 2016. 02. 003
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center
GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers
and authorised users (“Users”), for small-scale personal, non-commercial use provided that all
copyright, trade and service marks and other proprietary notices are maintained. By accessing,
sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of
use (“Terms”). For these purposes, Springer Nature considers academic use (by researchers and
students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and
conditions, a relevant site licence or a personal subscription. These Terms will prevail over any
conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription (to
the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of
the Creative Commons license used will apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may
also use these personal data internally within ResearchGate and Springer Nature and as agreed share
it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not otherwise
disclose your personal data outside the ResearchGate or the Springer Nature group of companies
unless we have your permission as detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial
use, it is important to note that Users may not:
use such content for the purpose of providing other users with access on a regular or large scale
basis or as a means to circumvent access control;
use such content where to do so would be considered a criminal or statutory offence in any
jurisdiction, or gives rise to civil liability, or is otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association
unless explicitly agreed to by Springer Nature in writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a
systematic database of Springer Nature journal content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a
product or service that creates revenue, royalties, rent or income from our content or its inclusion as
part of a paid for service or for other commercial gain. Springer Nature journal content cannot be
used for inter-library loans and librarians may not upload Springer Nature journal content on a large
scale into their, or any other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not
obligated to publish any information or content on this website and may remove it or features or
functionality at our sole discretion, at any time with or without notice. Springer Nature may revoke
this licence to you at any time and remove access to any copies of the Springer Nature journal content
which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or
guarantees to Users, either express or implied with respect to the Springer nature journal content and
all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published
by Springer Nature that may be licensed from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a
regular basis or in any other manner not expressly permitted by these Terms, please contact Springer
Nature at
onlineservice@springernature.com