1 s2.0 S0928765522000094 Main
1 s2.0 S0928765522000094 Main
1 s2.0 S0928765522000094 Main
a r t i c l e i n f o a b s t r a c t
Article history: The demand for motor fuel should decline when its price rises, but how exactly does that
Received 26 May 2021 happen? Do people drive less, do they drive more carefully to conserve fuel, or do they do
Received in revised form 19 October 2021 both? To answer these questions, we use data from the German Mobility Panel from 2004
Accepted 14 January 2022
to 2019, taking advantage of the fluctuations in motor fuel prices over time and across
Available online 17 January 2022
locales to see how they affect Vehicle Kilometers Traveled (VKT) and on-road fuel economy
(expressed in kilometers per liter) for gasoline and diesel cars. Our reduced-form regres-
JEL classification:
sions show that while the VKTs driven by gasoline cars decrease when the price of gasoline
Q41
rises, there is virtually no response among diesel cars. Likewise, the on-road fuel economy
Q53
Q54 is largely unresponsive to fuel price changes, irrespective of the fuel type. Since the price
R41 elasticity of fuel consumption is the difference between the price elasticity of VKT and the
price elasticity of the fuel economy, our results suggest that the fuel economy might be the
Keywords: “weakest link” of price-based policies that seek to address environmental externalities.
On-road fuel economy
© 2022 Elsevier B.V. All rights reserved.
Price elasticity
Vehicle kilometers traveled
Motor fuel prices
1. Introduction
The demand for motor fuel should decline when its price rises, but what is the mechanism for this? Do people simply
drive less, or do they change their driving style so that they also improve their fuel economy while driving less? Or are they
by now, at a time when fuel prices are relatively low, virtually insensitive to modest fluctuations in the price of motor fuels?
Two streams of literature are particularly relevant to these questions. One estimates the percent change in either the
consumption of fuel or in the distance driven given a percent change in the fuel price. The magnitude of such estimates
varies widely across studies, typically ranging from below -0.1 to -0.3 in North American-based studies (Goetzke and Vance,
2020; Hughes et al., 2008; Linn, 2016; Liu, 2014) to between -0.3 and -0.6 in studies from Europe (Frondel and Vance,
2018; Gillingham and Munk-Nielsen, 2019). By measuring the responsiveness to fuel price fluctuations, these estimates
serve to quantify the effect of price-based instruments like fuel taxes in reducing global pollutants, such as carbon dioxide
(CO2 ). They also provide key inputs into benefit-cost analyses of fuel economy regulations (National Highway Traffic Safety
Administration (NHTSA, 2011, 2020) and shed light about the likely magnitude of rebound effects (Sorrell and Dimitropoulos,
2008; Dimitropoulos et al., 2018).
∗ Corresponding author.
E-mail addresses: [email protected] (A. Alberini), [email protected] (M. Horvath), [email protected] (C. Vance).
https://doi.org/10.1016/j.reseneeco.2022.101292
0928-7655/© 2022 Elsevier B.V. All rights reserved.
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
The second stream of literature addresses the related question of how the fuel price influences the fuel economy, i.e., the
kilometers driven per unit of fuel. Much of this literature focuses on new car purchases, recognizing that higher gasoline
and diesel prices should encourage motorists to buy more fuel-efficient vehicles (Kilian, 2008). But what happens to the
fuel economy of existing cars—cars that are already part of the fleet at a specified locale—when the price of fuel changes? In
this instance, driving decisions—such as where, when, and how fast to drive—enter as crucial determinants of fuel economy
alongside the car’s technical features. Several studies have documented a growing divergence between official and real-
world fuel economy values (Fontaras et al., 2017; Pavlovic et al., 2018), reaching a mean of about 36 % for the German car
fleet (Tietge et al., 2017). Driving style—along with weather, road conditions, use of air conditioning, cargo and passenger
load, and vehicle maintenance status—is regularly identified as a contributing factor to this divergence. However, the relation
between driving style (which encompasses speed as well as variability in speed, among other things), fuel costs, and fuel
economy has not received much attention in the economics literature.
We use micro-level data from Germany to estimate the determinants of both distance driven and on-road fuel economy.
Germany is an ideal setting for this research due to the vast potential for fuel economy improvements in its vehicle fleet.
Germany has historically had a fleet with larger and/or faster cars, and lower fuel economy than other European countries.
Moreover, the absence of speed limits on its famed Autobahn system leads many drivers to drive faster than at the speed
recommended for optimal engine efficiency.
That behaviors can influence the fuel economy, and hence emissions, has been recently recognized in the Netherlands,
where in 2020 the highway speed limit was lowered from 130 km/hour to 100 km/hour to help reduce NOx and CO2 emissions
(DW, 2021), and even by the German environmental protection agency, which has recommended the introduction of speed
limits on the Autobahn for the same reasons (UBA, 2020). The European Environmental Agency likewise establishes a link
between speed and emissions, estimating that a reduction in motorway speeds from 120 to 110 km/h could deliver fuel
savings between 12 and 18 % (EEA, 2020). Given that German drivers can reach speeds of up to 200 km/h and beyond, there
would seem to be great potential for fuel savings through speed adjustments.
We use data from the German Mobility Panel from 2004 to 2019. In each of three years, panel participants are asked
to record the odometer reading at the beginning of the observation period in the spring—along with liters of fuel bought,
price per liter, and total expenditure at each refueling during the observation period. These records are summarized into a
monthly distance driven (in kilometers) and an average fuel economy (kilometers per liter of fuel) for that survey wave.
We use this information to estimate the price elasticity of fuel consumption. As this elasticity is comprised of two
parts—the fuel price elasticity of kilometers traveled (VKT), minus the fuel price elasticity of the fuel economy, we fit two
reduced-form models where log VKTs and the log fuel economy, respectively, are regressed on fuel prices, car attributes,
controls for the stock of cars owned by the household, and household sociodemographics. We include household and
location-by-year fixed effects to control for unobserved heterogeneity. Since we control for the stock of cars of the household
(which changes little from one wave of the survey to the next), we interpret our results as short-run elasticities.
We find that the kilometers logged on gasoline cars decline with the price of gasoline. The price elasticity is around -0.37,
and this finding is relatively stable across broad income groups and rural v. urban location. By contrast, diesel cars are driven
more, and have better fuel economy, but their VKTs are insensitive to changes in diesel prices. This finding is consistent
with an extensive v. intensive margin argument: People who must drive many miles for work- or family-related reasons
buy diesel vehicles as their cost per kilometer is lower than for gasoline cars, but then don’t adjust their driving as the price
of diesel changes.
Another key finding is that while we might expect motorists to adjust driving style to conserve fuel in the face of price
shocks, our model of fuel economy indicates that motorists are largely unresponsive in this respect. In fact, we even find some
evidence that higher gas prices are associated with worse fuel economy, though most of the estimates are not statistically
different from zero. The fuel economy would thus appear to be the “weakest link” in the relationship between price and
motor fuel demand, and to have the potential to limit the effectiveness of a carbon tax in reducing emissions.
Our calculation of the elasticity of fuel consumption with respect to the fuel price is the net result from the economet-
ric estimates of VKT and fuel economy. We use our derived elasticity estimate to predict the CO2 impacts of future fuel
price increases stipulated under Germany’s Climate Action Programme 2030. We find the predicted CO2 reductions to be
small—just 1–3 % of the total CO2 of passenger cars in Germany, in part because the modest carbon tax (25D /ton to 55D /ton)
implies minor price hikes, on the order of about ten eurocents by 2025, and in part because the overall price elasticity of
fuel consumption is moderate.
The remainder of the paper is organized as follows. Section 2 provides background information. Section 3 presents the
data. Section 4 describes the model, and section 5 the results. Section 6 concludes.
2. Background
Research addressing the effects of fuel prices on fuel consumption as well as VKTs has been summarized in reviews by
Graham and Glaister (2002); Goodwin et al. (2004), and Dimitropoulos et al. (2018). Aside from identifying relatively low
fuel price elasticities, one of the main findings of this literature is that elasticities derived from VKT are generally lower than
those derived from fuel consumption. This may follow from the fact that fuel consumption elasticities allow for additional
adjustments, including in the sales of new vehicles, and within-household switching between cars of different fuel economies
(Archsmith et al., 2020; De Borger et al., 2016). Other work has focused on improving the quality of the price data and on
2
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
distinguishing the effect of pre-tax fuel prices from that of taxes, with some studies finding the latter to be 3–4 times larger
than the former (Li et al., 2014; Rivers and Schaufele, 2015).
Studies that estimate the effects of fuel prices on the fuel economy have primarily been concerned with new car sales
(Kilian, 2008). The empirical evidence in this regard is ambiguous. For example, Hastings and Shapiro (2013) find that when
the price of gasoline rises, people don’t buy cars with better fuel economy: They simply substitute towards lower-grade
gasoline. Langer and Miller (2013) find that automakers offer offsetting discounts and rebates on fuel-inefficient models
when the price of gasoline rises. On examining new car sales or the registration records of the entire fleet, Li et al. (2009);
Klier and Linn (2010), and Rivers and Schaufele (2017) all find fuel price elasticities of the fuel economy below 0.2.
Relatively fewer studies have investigated what affects the fuel economy of an existing car, and of these, we are aware
of only two, Rouwendal (1996) and Knittel and Tanaka (2019), that have investigated the role of fuel prices. Based on self-
reported cross-sectional data from a sample of motorists in the Netherlands, Rouwendal (1996) estimates that a 1% increase
in the fuel price is associated with 0.15 % increase in on-road fuel economy. Using high-frequency data from a mobile
phone application in Japan, Knittel and Tanaka (2019) estimate a price elasticity of on-road fuel economy of about half this
magnitude, at 0.07 %.
Beyond fuel prices, the literature has identified several other factors that affect fuel economy. As documented by Fontaras
et al. (2017) for the European market, these include the mass of the vehicle, aerodynamic resistance, tires, and auxiliary
systems (e.g. heating, air conditioning, and steering assist systems). Greene et al.’s (2017) examination of over 600,000 self-
reported entries by US motorists to the myMPG.gov portal reveals evidence of considerable variation in fuel economy within
the same vehicle as well as across vehicles. The explanatory variable that accounts for most of the variation in on-road miles
per gallon (MPG) is the test-cycle MPG.1 Temperature also has some effect on the fuel economy—both direct (by affecting
the efficiency of combustion) and indirect (air conditioning hampers the fuel economy), as do powertrain characteristics and
self-reported driving style, while geography has very little explanatory power. Regarding driving style, Barkenbus (2010)
and Jeffreys et al. (2018) find that “eco-driving” in the US and Australia, respectively, can reduce fuel consumption by 5–25
% depending on traffic conditions and length of the trip.
Germany does not have specific fuel economy policies in place, and indeed has one of the least fuel-efficient, and highest-
emitting, fleets in the EU, despite regular improvement over the last decade as a result of the EU CO2 emissions regulations.2
In internal combustion engines, the CO2 emissions rate is inversely proportional to the fuel economy, implying that the latter
should have improved over time—for new cars because it is required by the regulations, and in the fleet as a whole as older
and more inefficient vehicles are retired and replaced with new ones. Fig. A1 in the Appendix shows that the fuel economy
of new cars sold in Germany improved steadily from 2001 to 2016, in anticipation of, and following, the CO2 emissions
regulations, but then plateaued and even worsened in 2017−18.3
Some incentive to purchase new cars with lower emissions rates, and hence better fuel economy, has arisen since 2009
as a result of annual circulation taxes that are a linear function of the CO2 emissions rate (for a given engine size and type of
fuel). The registration taxes are, however, modest, and so are their effects on new vehicle sales (Alberini and Horvath, 2021;
Klier and Linn, 2015).
With respect to existing cars, Tietge et al. (2017) examine on-road fuel economy figures reported by German drivers to the
online portal spritmonitor.de. The dataset covers cars manufactured from 2001 to 2014, and the fuel economy observations
appear to have been collected in 2014. Tietge et al. regress the on-road fuel consumption rate on test-cycle fuel consumption
rate,4 weight, engine size, and dummies for model-year, finding that the coefficients on the latter are positive and statistically
1
It is noteworthy that even US federal government agencies accept that the on-road fuel economy can be different from the test-cycle fuel economy
measured by the automakers or according to the official test procedures. Government assessments of proposed tightening of the Corporate Average Fuel
Economy (CAFE) standards, for example, assume a 20% gap between test-cycle fuel economy and actual, on-road fuel economy (National Highway Traffic
Safety Administration (NHTSA, 2011, 2020).
2
As part of the EU, Germany is subject to EU CO2 emissions regulations, which became effective in 2011 and have since required automakers to produce
and sell vehicles with emissions rates below specified targets. The targets are tightened over time, from 130 g CO2 /km in 2012 to 95 g CO2 /km in 2020
(https://ec.europa.eu/clima/policies/transport/vehicles/cars en).
3
The CO2 emissions rates of new cars followed a similar pattern. Fig. A2 shows that the emissions rates of new cars sold in Germany were consistently
higher (and the fuel economy lower) than those for the EU as a whole, as well as those of a neighboring country, France.
4
In Europe, until recently passenger vehicles were subject to the New European Driving Cycle (NEDC), which simulates a sequence of driving cycles
including accelerations, decelerations, periods of constant speed, and is thus supposed to mimic city and highway driving. The tailpipe gases are collected
and their CO2 content determined to arrive at the car’s CO2 emissions rate in grams per kilometer. The fuel consumption rate is calculated from the CO2
emissions rate by multiplying the latter by a fuel-specific constant.
3
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
significant, and increase monotonically with model-year.5 In other words, the more recently a car was manufactured, the
larger the difference between the certified and actual fuel consumption rates.
One possible explanation is that the test procedure used in the European Union (the NEDC) is based on an outdated
sequence of driving modes that no longer mirrors current driving conditions. Alternatively, automakers have become
increasingly sophisticated at optimizing performance during the test (by using slightly overinflated tires, no cargo, ideal
test environment temperatures, etc.). In the meantime, in September 2017 the EU adopted the Worldwide Harmonised
Light Duty Test Procedure (WLTP) which is expected to result in closer on-road and test-cycle performance.
With existing cars, one would expect fuel prices to be a determinant of kilometers driven as well as on-road fuel economy.
In Germany, taxes account for a considerable portion of the price of motor fuels: Since 2002, German motorists have paid
a tax of 65.45 eurocents per liter of gasoline and 47.04 eurocents per liter of diesel, plus a VAT of 19 %, which means that
taxes account for about 60 % of the fuel price. The excise tax amount has remained unchanged during our study period
(2004–2019), while the VAT was only increased once from 16 % to 19 % at the beginning of 2007, implying that any changes
in fuel prices during our study period were due solely to fluctuations in world oil prices and/or local demand and supply
conditions.6
In the remainder of this paper, we estimate the fuel price elasticities of kilometers traveled and fuel economy and assess
their roles in the determination of the overall fuel price elasticity of fuel consumption. Distinguishing from work to date,
both estimates are based on actual, on-road micro-level data drawn from a panel of households. While earlier research
investigates the “extensive margin,” i.e., the decision to purchase a car, in this paper we analyze the intensive margin—the
usage of cars already owned by a household.
3. The data
We use data from the German Mobility Panel (GMOP), which commenced in 1994. Our sample covers data from 2004—the
first year when the questionnaire included a question on household income—until 2019. Participating households remain
in the panel for 3 years, then are rotated out and replaced by a new cohort.
During a total of six weeks in the spring, participating households record the odometer reading of each of their vehicles at
the beginning of the observation period, and every time they refuel their vehicle. At the pump, they also record the price per
liter of gasoline or diesel, the number of liters they bought, and the total amount paid. A final odometer reading is done at
the end of the observation period. The kilometers driven during the observation period are then computed as the difference
between the final and initial odometer reading, and are converted to a monthly equivalent.7 We compute the average fuel
economy during the observation period as the total number of kilometers driven divided by the total number of liters of
gasoline purchased.
Attention in this paper is restricted to households with at least one car.8 Table A1 in the Appendix provides information
about our sample, which follows a total of 5858 households. About 40 % stayed in the GMOP for three years; about 36 % of
the sample participated for only one year, and 24 % participated for two years.
We checked whether attrition is somehow associated with the dependent variables we study in this paper, namely
kilometers driven and the on-road fuel economy, but found no evidence that this is the case. Rather, “early dropouts” were
more likely among families with small children and less among the elderly, and bore no significant association with the VKT
driven on family cars and their fuel economy.
Our sample contains a total of 8358 cars. About 71 % of the households own only one car, 26 % own two, 3% three, and
less than 1% own four or more cars. About 71 % of all cars are gasoline cars and 26 % are diesel cars. The remaining 3.56 %
run on alternative fuels, such as natural gas or LPG. Company cars account for 6% of the observations.
The turnover of vehicles is similar to what can be expected in the population: Every year about 4.39 % of the cars are new
additions to the family’s stock of cars (compared to the previous year), 2.25 % exit and are not replaced, and 6.50 % of the
households replace one or more cars, keeping the total number of cars owned by the household unchanged.
The average car is driven about 1129 km per month, or 13,548 km on an annual basis. Figs. 1 and 2 display histograms
of the annual kilometers driven, comparing gasoline v. diesel, and gasoline with alternative fuel cars. The distributions of
annual kilometers driven are positively skewed, and diesel and alternative fuel cars are generally driven more than gasoline
cars. That’s unsurprising: People who must—for whatever reason—drive a lot often buy diesel or natural gas cars because
5
The regression explains over 88% of the variation in on road fuel economy for gasoline cars and over 89% for diesel cars. Adding the year dummies
however improved the R square by a mere 1 percentage point compared to simpler models that include only the NEDC fuel consumption rate, weight and
engine capacity.
6
Frondel et al. (2020) and Horvath (2019) examine retail motor fuel prices in Germany, concluding that—especially after the adoption of a federal
requirement that each gas station post its prices daily on an online portal accessible to regulators and members of the general public—competition became
more intense, reducing profit margins and bringing quicker pass-through of world oil prices.
7
The difference between subsequent odometer readings is the number of kilometers driven between refueling operations, which is typically recorded
over a six-week period, give or take a few days. We convert this to a monthly value by multiplying the total kilometers by 30 divided by the number of
days of the survey. The total kilometers driven can also be obtained as the sum of all kilometers driven in between gas station visits.
8
Households with at least one car account for 86% of the households in the Germany Mobility Panel.
4
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Fig. 1. Distribution of kilometers driven (scaled to annual basis), GMOP 2004-2019: Gasoline v. diesel cars.
Fig. 2. Distribution of kilometers driven (scaled to annual basis), GMOP 2004-2019: Gasoline v. alternative-fuel cars.
Fig. 3. Distribution of the fuel economy (km. per liter), GMOP 2004-2019: Gasoline v. diesel cars.
they are more fuel-efficient and, taking fuel prices into account, the cost of driving them is lower than the cost of driving
gasoline cars.
Fig. 3 shows that the actual, on-road fuel economy of the cars in our sample—namely the actual kilometers driven on a
liter of motor fuel—displays considerable variation, and is generally more favorable among diesel cars. The fuel economy is
on average 13 km/liter among gasoline cars and 15 km/liter among diesel cars.
5
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Fig. 4. On-road fuel economy in the Germany Mobility Panel and new-car fuel economy as per Manufacturer certification in Germany, 2011-2019.
Note: For each year from 2011 to 2019, this figure displays i) the average fuel economy of all new cars sold in Germany (source: IHS-Markit), ii) the average
fuel economy of the cars in the German Mobility Panel, and iii) the average fuel economy of cars in the German Mobility Panel that were manufactured in
that year, the year before, or two years before.
Table 1
Fuel economy for cars in the GMOP belonging to different market segments.
mean sd cov
Fig. 4 compares the average, on-road fuel economy in the GMOP with that of new cars sold in Germany between 2011 and
2019. The evidence from this figure is striking. The fuel economy as certified by the manufacturers grew at a steep rate among
the new cars sold in Germany until about 2016, but this trend flattened and even worsened thereafter. By contrast, the on-
road fuel economy has grown slowly but steadily (from about 13.5 km/liter to about 14.5 km/liter) in the cars documented
in the GMOP. When attention is restricted to relatively new cars within the GMOP, which we define as those manufactured
up to two years before each wave of the GMOP, the fuel economy tends to be better than that of all cars in the GMOP, but
follows a similar growth pattern. Fig. 4 shows that the on-road fuel economy of these relatively new cars in the GMOP is
well below the manufacturer-certified fuel economy of the new cars sold in Germany, and that the two had been diverging
over time, confirming evidence from other studies (e.g., Tietge et al., 2017).
In the GMOP, the fuel economy varies across vehicles as well as within a vehicle over time.9 It also varies between and
within households.10 Table 1 shows how the on-road fuel economy varies across different types of cars. As expected, the
smallest cars—namely, those classified as “mini,” “small” or “compact” (e.g., Smart ForTwo, Fiat 500, and Honda Civic)—have
better fuel economy than large cars, sports utility vehicles, and sports cars. The difference can be as large as 6 km/liter.
Table 2 summarizes the results of regressions where the dependent variable is the log fuel economy of a vehicle in
the GMOP and the independent variables are car characteristics and/or various fixed effects. Market segment fixed effects
alone explain 17 % of the variation in the dependent variable, but when all available car characteristics (such as engine
displacement, horsepower per ccm of engine size, car age and make dummies) are added, they explain almost 46 % of the
variation in log fuel economy. Adding household fixed effects brings the R square of the regression to 77 %. If the regression
only includes car-specific fixed effects, the R square of the regression is 88 %, which means that each car does exhibit some
variation in its on-road fuel economy from one year to the next. One of the goals of this paper is to explore how much of this
variation over time is linked to fluctuations in fuel price, presumably reflecting drivers’ behavior adjustments in response
to them.
9
The overall standard deviation of the log fuel economy is 0.2375. The “between” standard deviation is 0.2299, and the “within” standard deviation is
0.0816.
10
The between household standard deviation is 0.2072, and that within households is 0.1373.
6
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 2
Summary of the regressions of ln fuel economy of the cars in the GMOP on car characteristics and various fixed effects.
Regressors R square
Fig. 5 displays the average gasoline and diesel prices (expressed in real 2010 euro/liter) during our study period, showing
that there were considerable fluctuations over time. By 2012, for example, the prices were 60–70 % higher than they were
at the end of the 1990s. Tables A2 and A3 in the Appendix display the mean prices of gasoline and diesel each year along
with their standard deviations, demonstrating that there was quite a bit of variation even within the same year.
4. Methods
The fuel efficiency or economy of a vehicle is defined as useful work (for example, driving ten kilometers) divided by the
fuel input required by that useful work (for example, one liter of gasoline) (Sorrell, 2009). The reciprocal of the fuel economy
is the fuel consumption rate. Better fuel economy figures imply lower fuel consumption rates.
The central aim of this paper is to estimate how the consumption of fuel (C) changes in response to changes in the fuel
price (p), expressed in elasticity form as ∂C · Cp . Since fuel consumption is equal to vehicle kilometers traveled (VKT) divided
∂p
VKT
by fuel economy (FE), C = FE
, where FE equals the kilometers driven per liter of fuel, this elasticity can be decomposed as:
∂C p 1 ∂VKT ∂FE p ∂VKT 1 p ∂FE VKT 1 p
· = · FE − · VKT · = · · − · · · =
∂p C FE 2 ∂p ∂p C ∂p FE VKT ∂p FE FE VKT
FE FE (1)
∂VKT p ∂FE p
= · − · .
∂p VKT ∂p FE
Expression (1) shows that the elasticity of fuel consumption with respect to fuel price is the difference between the fuel
price elasticity of kilometers traveled, ∂VKT · VKT
p
, and the fuel price elasticity of the fuel economy, ∂FE · FE
p
. Our econometric
∂p ∂p
approach estimates each of these elasticities.
Our unit of observation is an individual car, which is followed for up to three years. We fit two regression equations. The
first relates the average fuel economy during the reference period (expressed in km per liter of fuel) to vehicle characteristics,
including its age, and a host of personal, household and location characteristics. Formally,
7
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
where i denotes the car, t the year, z a vector of vehicle characteristics, x denotes individual and household characteristics
(including the household’s stock of vehicles), and w is a vector describing the local geography and the roads. Parameter ˛ is
the price elasticity of the fuel economy, which, as shown in Eq. (1), enters in the fuel price elasticity of total fuel consumption.
Vector z includes make dummies, engine size and horsepower per unit of engine size (a proxy for acceleration), model-
year dummies, and a diesel engine dummy. It does not include the official, manufacturer-certified fuel economy of the
vehicle, which is not included in the GMOP. However, based the relationship between fuel economy and car characteristics
and performance documented in Knittel (2011) and additional evidence from a proprietary dataset with official and on-road
test results, we judge vector z to be sufficient to capture the official fuel economy. 11
We do not observe driving style directly, but it is reasonable to think that car characteristics mirror the driving style of their
owner. Cars with modest acceleration and power presumably do not lend themselves to a sporty, aggressive type of driving,
and, conversely, a conservative driver is unlikely to seek high-performing models. Household or driver characteristics, as
well as location, may also influence (and hence partially capture in our regressions) driving style. Household income and
rural v. urban location may be especially important in this regard. In addition to including these as control variables, we
implement robustness checks that split the sample by income level and location to test for differential fuel price elasticities.
Eq. (2) allows for the possibility that drivers are aware of the relationship between driving behavior and fuel economy,
and attempt to modify that behavior—perhaps by driving at constant speed, refraining from abrupt acceleration or braking,
or seeking to travel at times of the day or types of road where fuel-conserving behavior is possible—when fuel is more
expensive.
Eq. (2) also includes fixed effects to account for unobserved heterogeneity. We considered several candidate fixed effects.
We ruled out vehicle-level fixed effects because including them would reduce the usable sample size by one-third (the share
of vehicles that appear in the sample in only one year) and absorb the effect of many of the covariates we are interested
in. Household fixed effects allow us to keep car characteristics in the model. They are identified either by households that
appear more than one year in the data or that own more than one car.
Finally, it is important to include location-by-time fixed effects to capture demand shocks, the state of the local economy,
and the type and quality of the roads (which can vary over time depending on road work and construction projects), weather,
etc. For this purpose, we include a set of county-by-year interactions. Germany has 410 counties (or Kreise), with an average
size of about 830 square kilometers. The interaction of these dummies with year dummies thus serves to guard against
biases that could otherwise emerge from time-varying, regional circumstances (e.g., unemployment, local demand shocks)
that are correlated with fuel prices and that directly bear on driving behavior.12
Drivers may respond to motor fuel price changes by driving less (when the price increases) or more (when the price
decreases). We specify a demand function for kilometers driven, which are assumed to depend on the same set of factors
that determine fuel economy, including fuel cost per kilometer driven, income, and other factors, such as work and family
needs, etc:
In Eq. (3) the dependent variable is thus the log of the monthly kilometers driven. The regressors, all assumed exogenous,
and the type of fixed effects, are the same as in the log fuel economy regression (Eq. (2).
In sum, we have specified two reduced-form regression equations where the dependent variables are the logs of the
on-road fuel economy and of the kilometers per month, respectively, and the key independent variable is log fuel price.13
We take log transformations for two reasons: First, the distributions of fuel economy and distance driven are positively
skewed (see Figs. 1–3), and, second, the coefficients on log fuel price are immediately interpreted as the price elasticities of
11
Greene et al. (2017) find that the most important predictor of the on-road fuel economy is the test-cycle fuel economy, i.e., that from official
manufacturer- or government-run tests. We acquired a dataset from Emissions Analytics that documents the official and actual, on-road fuel economy of
vehicles tested in Europe between 2011 and 2020. A total 1185 gasoline and diesel cars aged between one and 24 years underwent on-road emissions
testing using approved portable equipment. Regression analyses using this sample show that the official fuel economy of a car is well predicted by the car’s
make, age, engine displacement in cubic centimeters, horsepower, and fuel type (the same variables as those in vector z in equation (2)). These variables
explain 80% of the variation in the log of the technical fuel economy. When we regressed the official fuel consumption rate (expressed in liters/100 km),
which is the reciprocal of the fuel economy, on make dummies, model-year dummies, test-year dummies, engine ccm, and horsepower divided by engine
ccm, the R square of the regression was 85%.
12
Only a handful of people move from one county to another during the study period, and for this reason county-specific fixed effects alone (not interacted
with time) would be absorbed into the household fixed effects.
13
We consider our regressions as “reduced-form” because all right-hand side terms are assumed to be exogenous over the three-year observation period
of the data. We fit least squares separately to each regression equation. The two equations may also be regarded as a system of seemingly unrelated
regressions (SUR) if the error term in (4) is correlated with that in (2) for the same car and GMOP survey wave, due to unobserved factors that influence
both the distance driven as well as the fuel economy. SURs are efficiently estimated using generalized least squares (GLS). In this case, however, since the
regressors are the same in both equations, GLS estimation boils down to fitting OLS equation by equation. One would thus obtain the same point estimates
of the coefficients with SUR GLS estimation or least squares on each equation separately.
8
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 3
Main regressions.
fuel economy and distance driven, respectively. These are the two elasticities needed to predict how fuel consumption will
change as motor fuel prices change.
5. Results
Table 3 presents the estimates from the models of fuel economy and vehicle kilometers traveled (VKT). The mod-
els in columns (3) and (4) are distinguished from those in columns (1) and (2) by the inclusion of household fixed
effects.
9
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
We begin our discussion of the results with the fuel economy regressions, shown in columns (1) and (3). Including the
household fixed effects (col. (3)) has no bearing on our main effect of interest: the fuel price has a negative but small and
statistically insignificant impact in both models. Moreover, as evidenced from the statistically insignificant interaction with
the diesel car dummy, this null effect applies to both gasoline and diesel cars. Unsurprisingly, diesel cars are generally
more fuel efficient in the first place, as shown by the positive and significant coefficients on the diesel car dummy in
both models, while cars with larger engines and with a higher horsepower-to-engine size ratio tend to have a lower fuel
economy.
We entered in the regression several controls to capture the possibility of substitution among the household’s vehicles,
finding some evidence that the number of cars owned and the existence of another gasoline car is positively associated
with fuel economy. The remaining control variables capture sundry household socioeconomic characteristics like income,
location, and sociodemographic composition. The dummy indicating a low-income household, the number of full-time
workers and the number of middle and high school-aged children in the household are significantly associated with the fuel
economy in col. (1), but in practice they suggest small effects (e.g. adding a worker raises the fuel economy by a mere 1.4
%). Moreover, these variables are all uniformly statistically insignificant in the household fixed effects in column (3), which
likely reflects their limited variation over the three years covered by the survey.
In sum, the regressions reported in cols. (1) and (3) of Table 3 suggest that car characteristics (including the make, fuel
type, vintage, engine size and horsepower per unit of engine size), location-by-time fixed effects and household fixed effects
explain up to 80 % of the variation in fuel economy. Adding fuel price in the model raises the R square of the regression by
only one point, up to 81 %.
By contrast, the log VKT regressions in columns (2) and (4) show a stronger responsiveness of kilometers driven to fuel
price —at least for gasoline cars. Starting with the model that omits household fixed effects in column (2), for gasoline cars
the price elasticity of kilometers driven is approximately -0.39. Although this figure is situated toward the higher end of
estimates from the literature (e.g. Gillingham and Munk-Nielsen, 2019; Goetzke and Vance, 2020), it is within the range
of -0.35 to -0.41 reported by Frondel and Vance (2009), who employ a two-part model to model VKT conditional on car
ownership. The results in Table 3 further suggest that the magnitude of the effect of the fuel price may be somewhat weaker
for diesel cars, although the estimate of the interaction term is statistically insignificant.
VKT is also related to the number and types of the other cars owned by the household. Specifically, the signs and magnitude
of the coefficients on the dummies denoting that the household has other diesel cars and other gasoline cars, and the number
of cars owned by the household, are all consistent with the notion that households tend to spread the miles over their different
vehicles, and tend to log more miles on their diesel cars. Mileage is also affected by income, with low-income households
driving about 4% less than middle- and high-income households, and by household composition. Each additional full-time
worker in the household increases the distance driven by about 17 %, while additional babies and children increase it by
about 7%.
The effect of gasoline price on the VKT of gasoline cars remains just as strong in column (4) of Table 3, which adds
household-specific fixed effects. This specification, however, finds no effect of the diesel price on the VKT of diesel cars, as
evidenced by the statistically significant and offsetting magnitude of the interaction effect. Moreover, the coefficient on the
dummy indicating a diesel car itself is small and statistically insignificant. Most of the remaining household-level control
variables are likewise statistically insignificant.
Table A4 in the Appendix displays the results from the main regressions using the full sample but clustering the standard
errors at the vehicle level (cols. (1) and (2)), household level (cols. (3) and (4)), and county-by-year level (cols. (5) and (6)).
The standard errors change very little across the different types of clustering, and we arrive at the same conclusions as in
Table 3 in terms of significance of the coefficients.
Tables 4–6 present models that split the sample along various dimensions to allow for differential responses according
to socioeconomic circumstances. All of these regressions include household fixed effects and county-by-year fixed effects.
Households with working members appear capable of adjusting their VKTs more to increases in gasoline prices than the
other households, at least when they drive gasoline cars. Their fuel economy appears to worsen as the price of gasoline rises
(Table 4).
It might seem surprising that households comprised exclusively of retired persons are relatively insensitive to fuel prices.
We believe this is because they have already chosen to drive fewer kilometers, opting perhaps more often in favor of public
transit or no travel at all. Some support for this is seen in the fact that their vehicles are driven on average 842 km/month,
whereas those owned by households with working members 1255 km/month.
Likewise, higher fuel prices reduce the fuel economy among urban households (Table 5), while the VKT response to
fuel prices is roughly on par with rural households. Cars owned by mid- and high-income households are not statistically
different in their responsiveness to gasoline prices from cars in low-income households (Table 6).
Regional or local shocks are a potentially important factor affecting the VKT and the fuel economy, and are captured in
our models by the county-by-year interactions. Counties, however, are relatively large areas and as such may be a coarse
indicator of household location. To explore the implications of a more spatially precise situating of households, we replace
these interactions with a smaller geographical unit given by the overlap of counties and three-digit zip codes, which produces
10
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 4
Regressions for working/non-working Households.
polygons having an average size of 282 km2 (compared with 830 km2 for counties). As shown in Table 7, the inclusion of
the resulting postal code area-by-year fixed effects results in a smaller estimation sample but does not markedly change our
main results. In fact, the price elasticity of VKT is a bit stronger in these specifications.
Finally, although we regard fuel prices as free from measurement error, because they are recorded directly during refuel-
ing, they may be endogenous with the dependent variable, should cost-conscious drivers “shop around” for lower fuel prices.
We consider two possible approaches for addressing this. The first replaces the fuel price reported by each study participant
with the average of the prices reported by all other study participants in that county and year. This removes endogeneity,
if it is present, but introduces measurement error, and indeed results in similar, but statistically insignificant, coefficients
on log fuel price in the fuel economy and VKT equations.14 The second approach uses the average of all others’ prices as a
14
To illustrate, the coefficient on log fuel price for gasoline cars is 0.0341 (s.e. 0.0750) in the fuel economy equation and -0.1498 (s.e., 0.2574) in the VKT
equation.
11
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 5
Regressions for Households in rural or urban areas.
cross-validation instrument for the respondent-reported fuel price. Again, the coefficients on (instrumented) log fuel price
are qualitatively similar, but statistically insignificant.15
Finally, to address possible adverse effects of multiple drivers using the same car, in Table A5 of the Appendix we show
estimation results based on the subsample where the number of cars is equal to the number of household members with
driver’s licenses, assuming that in this case each household member has their own assigned car. The results are similar in
direction and magnitude to our overall results.
The implications of our findings can be illustrated by evaluating the effect of the carbon tax that was recently implemented
in Germany. Starting in January 2021, the German government introduced a carbon tax of D 25/ton of CO2 , which will
15
The coefficient on log fuel price for gasoline cars is -0.2308 (s.e. 0.4594) in the fuel economy equation and -0.3265 (s.e. 1.5584) in the VKT equation.
12
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 6
Regressions by income groups.
gradually increase to D 55 per ton by 2025.16 Our calculations focus on gasoline cars, which make up 65 % of the 47.7 million
passenger car stock in Germany. We use the CO2 content of gasoline (2.3 kg CO2 per liter) to calculate the increase in the
gasoline price due to the tax, which is about 5.75 euro-cents for the D 25 tax and 12.65 euro-cents for the D 55 tax,17 assuming
that 100 % of the tax is passed on to the consumers (Kopczuk et al., 2013).
Eq. (1) shows that the elasticity of fuel consumption with respect to fuel price is the difference between the fuel price
elasticity of kilometers traveled and the fuel price elasticity of the fuel economy. Although our estimates indicate that the
latter elasticity is statistically insignificant, we include it in the calculation since it is unbiased, expecting its negative sign
to yield a conservative estimate of the effects of a carbon tax. We use the results in cols. (3) and (4) in Table 3, which
puts the elasticity of fuel economy with respect to the price at -0.062 and the elasticity of kilometers driven at -0.371,
for a net price elasticity of fuel consumption of -0.309. We repeat the calculation using the slightly stronger elasticities
from Table 7 (-0.047 for the fuel economy and -0.47 for kilometers driven, for a net price elasticity of fuel consumption of
-0.421).
16
See https://www.bundesregierung.de/breg-de/themen/klimaschutz/co2-bepreisung-1673008
17
See https://ecoscore.be/en/info/ecoscore/co2 for the CO2 content of gasoline. For our calculations we use the average liters consumed per car in our
GMOP sample, which is at 81 liters per month.
13
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 7
Regressions with County-postalcode-by-year fixed effects: Selected Coefficients.
A. Main regressions.
14
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table 7 (Continued)
A. Main regressions.
Based on these two alternate overall price elasticities, we predict a reduction of between 982,911 and 1,339,117 tons of
CO2 , which is roughly 1.02%–1.39% of the total CO2 emissions from passenger cars in Germany, for a CO2 tax of D 25/ton.
A D 55/ton tax would deliver reductions between 2,162,405 and 2,946,189 tons of CO2 (2.25%–3.06% of total passenger car
CO2 emissions). To put these numbers in perspective, consider that the average gasoline car in our GMOP sample emits a
total of 2.067 tons of CO2 a year.18
Using this information, we calculate that the CO2 reduction brought by the tax is equivalent to retiring between 475,525
and 647,884 cars from the fleet for the D 25 tax, and 1,046,156 to 1,425,345 cars for the D 55 tax. These figures represent
1%–1.39%, and 2.19%–2.99%, of the current fleet, respectively. While these numbers may seem small compared to the fleet,
they are actually equal to 13–25% of the new cars sold in Germany every year.19
Our prediction of the CO2 emissions reductions from gasoline cars changes only very marginally if we make assumptions
about a further shift away from diesel cars towards gasoline cars, or replacing a part of the fleet with electric cars. This is in
part because of the modest combined elasticity of -0.309, but also because we only find a reaction for gasoline cars. With
currently roughly 65 % of the stock of cars being gasoline, there would need to be a massive change in the stock of gasoline
cars to change the results meaningfully.
The factor that has the strongest impact is the CO2 tax rate itself. A CO2 tax of D 200—similar to proposals by some
environmental organizations but most likely politically infeasible in the short run—would increase the price of gasoline by
46 euro-cents and reduce the CO2 emissions of passenger cars by about 6.56%–8.94%. These calculations are likely to be
lower bound estimates as they assume that the responsiveness to a carbon tax would be the same as the responsiveness
to a change in the fuel price. As Rivers and Schaufele (2015) document in their study of tax policy in British Columbia,
the elasticity associated with a fuel or carbon tax is upwards of four times larger, which would imply a more substantial
reduction in the CO2 emissions of passenger cars.
6. Conclusions
Using data from the German Mobility Panel from 2004 to 2019, we have estimated the determinants of both distance
driven and fuel economy. This dual focus is motivated by our aim to estimate the elasticity of fuel consumption with respect
to the fuel price, which equals the difference between the fuel price elasticity of VKT and the fuel price elasticity of fuel
18
This is based on the average annual VKT in our sample (13,060 km) and the average fuel economy (14.52 km per liter) in 2019.
19
During the last decade, about 3 to 3.5 million new cars are sold in Germany every year.
15
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
economy. We have fit reduced-form regressions that related log VKTs and log on-road fuel economy to log fuel price, car-,
household- and location characteristics. We interpret our estimates as short-run fuel price elasticities.
Overall, our models do a reasonable job explaining VKTs, but the determinants of on-road fuel economy remain elusive. In
our base models, car characteristics and fixed effects accounting for unobservables capture much of the variation in on-road
fuel economy, with little contribution from sociodemographics and fuel price.
Notwithstanding the challenges in modeling fuel economy, three key findings emerge: First, for diesel cars, neither the
distance driven nor the fuel economy seems to respond to fuel price changes. This could be the result of a selection- or an
extensive v. intensive margin argument: People who need to drive a lot for work or family reasons buy diesel cars in the
first place, because their cost per kilometer is lower, but do not further adjust their driving to fuel price changes. Whatever
its source, this unresponsiveness of diesel motorists to the fuel price bodes well for tax revenue under the Climate Action
Programme. The Programme’s tax on the CO2 content of fuel, which is about 13 % higher for diesel than for petrol, will
address what critics have long contended to be excessively low taxes for diesel, costing the government D 7.1 billion in lost
revenue in 2019 according to Germany’s Green Party (Wehrmann, 2020).
Second, for gasoline cars, kilometers driven decline with a higher gasoline price, with a price elasticity of about -0.37.
Third, as with diesel cars, the fuel economy appears relatively unresponsive to price increases for gasoline. The reason
for this result is not immediately obvious, as we would have expected drivers to adopt a more “efficient” driving style when
the price increases. Evidently, the behaviors that drivers undertake to reduce driving—forgoing long trips, perhaps driving
more in the city, combining trips with those of other family members, carpooling—end up having little bearing on the car’s
fuel economy and, if anything, seem to make it worse.
As a result, our calculation of the mean elasticity of fuel consumption with respect to price, -0.309, is lower than that
implied by the elasticity of kilometers driven, -0.371. Whether such a downward adjustment is generally required in assess-
ments of fuel consumption is an open question, and even in this case it should be regarded as conservative given the
statistically insignificant estimate of the fuel economy elasticity. Moreover, Germany may be unique in this regard. The
country’s deep-seated cultural identification with cars, which is reflected partly in popular resistance to speed limits on the
Autobahn, may figure into the explanation of why motorists don’t adjust driving style in the expected manner to fuel prices.
We speculate that in many other countries, higher fuel prices are associated with better on-road fuel economy, but this is
ultimately an empirical question that awaits future research.
Data availability
The survey data of the German Mobility Panel is hosted by the Clearing House of Transport Data at the German Aerospace
Center (DLR). Data access can be requested there.
Acknowledgement
Alberini gratefully acknowledges support from the Czech Science Foundation under grant no. 19-26812X. Horvath and
Vance have received support by the Federal Ministry of Education and Research under grant 01LA1809C (Project DIPOL)
and by the Collaborative Research Center “Statistical Modeling of Nonlinear Dynamic Processes” (SFB 823) of the German
Research Foundation (DFG), within Project A3, “Dynamic Technology Modeling.”
16
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Appendix A
Table A1
Descriptive Statistics of the Sample.
Table A2
Gasoline price by year (2010 real euro/liter).
Table A3
Diesel prices by year (2010 euro).
17
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table A4
Main regressions with clustered Standard Errors.
enginesize (in 1000 cm3) −0.224*** 0.098*** −0.224*** 0.098*** −0.224*** 0.098***
(0.011) (0.031) (0.013) (0.033) (0.012) (0.032)
Walking minutes to public transit −0.000 0.002 −0.000 0.002 −0.000 0.002
(0.001) (0.004) (0.001) (0.003) (0.001) (0.003)
Number of babies (age <2) in HH −0.016 0.034 −0.016 0.034 −0.016 0.034
(0.026) (0.086) (0.018) (0.064) (0.017) (0.059)
Number of children between 10–17 in HH 0.025* −0.002 0.025** −0.002 0.025*** −0.002
(0.013) (0.045) (0.010) (0.035) (0.009) (0.033)
18
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Table A5
Regression for the subsample with number of cars equal to number of driver’s licenses in the household.
19
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Fig. A1. Fuel economy and CO2 emissions rate of new cars in Germany.
Source: https://theicct.org/sites/default/files/publications/European vehicle market statistics 20192020 20191216.pdf.
Fig. A2. Average CO2 Emissions Rates for New Cars in different countries.
https://www.eea.europa.eu/publications/co2-emissions-from-cars-and-vans-2018.
Source: Monitoring CO2 emissions from passenger cars and vans in 2018, EEA Report 02/2020
References
Alberini, Anna, Horvath, Marco, 2021. All car taxes are not created equal: evidence from Germany. Energy Econ., 105329.
Archsmith, James, Gillingham, Kenneth, Knittel, Christopher R., Rapson, David S., 2020. Attribute substitution in household vehicle portfolios. RAND J.
Econ. 51 (4), 1162–1196.
Barkenbus, Jack N., 2010. Eco-driving: an overlooked climate change initiative. Energy Policy 38 (2), 762–769.
De Borger, Bruno, Mulalic, Ismir, Rouwendal, Jan, 2016. Substitution between cars within the household. Transp. Res. Part A 85, 135–156.
Dimitropoulos, Alexandros, Oueslati, Walid, Sintek, Christina, 2018. The rebound effect in road transport: a meta-analysis of empirical studies. Energy
Econ. 75, 163–179.
DW – Deutsche Welle, 2021. Netherlands Reluctantly Introduces Reduced Speed Limit.
https://www.dw.com/en/netherlands-reluctantly-introduces-reduced-speed-limit/a-52796095.
European Environmental Agency, 2020. Do Lower Speed Limits on Motorways Reduce Fuel Consumption and Pollutant Emissions?
https://www.eea.europa.eu/themes/transport/speed-limits-fuel-consumption-and.
Fontaras, Georgios, Zacharof, Nikiforos-Georgios, Ciuffo, Biagio, 2017. Fuel consumption and CO2 emissions from passenger cars in Europe – laboratory
versus real-world emissions. Prog. Energy Combust. Sci. 60, 97–131.
Frondel, Manuel, Colin, Vance, 2009. Do high oil prices matter? Evidence on the mobility behavior of German households. Environ. Resour. Econ. 43, 81–94.
Frondel, Manuel, Vance, Colin, 2018. Drivers’ response to fuel taxes and efficiency standards: evidence from Germany. Transportation 45, 989–1001.
Frondel, Manuel, Horvath, Marco, Vance, Colin, Kihm, Alexander, 2020. Increased Market transparency in Germany’s gasoline Market what about rockets
and feathers? J. Transp. Econ. Policy 54 (2), 102–120.
Gillingham, Kenneth, Munk-Nielsen, Anders, 2019. A tale of two tails: commuting and the fuel price response in driving. J. Urban Econ. 109, 27–40.
Goetzke, Frank, Vance, Colin, 2020. An increasing gasoline price elasticity in the United States? Energy Econ., 104982.
Goodwin, Phil, Dargay, Joyce, Hanly, Mark, 2004. Elasticities of Road traffic and fuel consumption with respect to price and income: a review. Transport
Rev. 24 (3), 275–292.
Graham, Daniel J., Glaister, Stephen, 2002. The demand for automobile fuel: a survey of elasticities. J. Transport Econ. Policy 36 (1), 1–25.
Greene, David L., Liu, Jun, Khattak, Asad J., Wali, Behram, Hopson, Janet L., Goeltz, Richard, 2017. How does on-road fuel economy vary with vehicle
cumulative mileage and daily use? Transp. Res. Part D 55, 142–161.
Hastings, Justine S., Shapiro, Jesse M., 2013. Fungibility and consumer choice: evidence from commodity price shocks. Q. J. Econ. 128 (4), 1449–1498.
Horvath, Marco, 2019. Germany’s Market Transparency Unit for Fuels: Fostering Collusion or Competition?,” Ruhr Economic Papers #836. RGS, RWI.
Hughes, Jonathan E., Knittel, Christopher R., Sperling, Daniel, 2008. Evidence of a shift in the short-run price elasticity of gasoline demand. Energy J. 29 (1),
93–114.
Jeffreys, Ian, Graves, Genevieve, Roth, Michael, 2018. Evaluation of eco-driving training for vehicle fuel use and emission reduction: a case study in
Australia. Transp. Res. D. Transp. Environ. 60, 85–91.
Kilian, Lutz, 2008. The economic effects of energy price shocks. J. Econ. Lit. 46 (4), 871–909.
Klier, Thomas, Linn, Joshua, 2010. The price of gasoline and New vehicle fuel economy: evidence from monthly sales data. Am. Econ. J. Econ. Policy. 2 (3),
134–153.
Klier, Thomas, Linn, Joshua, 2015. Using taxes to reduce carbon dioxide emissions rates of New passenger vehicles: evidence from France, Germany, and
Sweden. Am. Econ. J. Econ. Policy. 7 (1), 212–242.
Knittel, Christopher R., 2011. Automobiles on steroids: product attribute trade-offs and technological progress in the automobile Sector. Am. Econ. Rev.
101 (7), 3368–3399.
Knittel, Christopher R., Tanaka, Shinsuke, NBER working paper 26488 2019. “Driving Behavior and the Price of Gasoline: Evidence from Fueling-Level
Microdata. November., Cambridge, MA.
20
A. Alberini, M. Horvath and C. Vance Resource and Energy Economics 68 (2022) 101292
Kopczuk, Wojciech, Marion, Justin, Muehlegger, Erich, Slemrod, Joel, NBER Working Paper 19410, Cambridge, MA, September 2013. Do the Laws of Tax
Incidence Hold? Point of Collection and the Pass-Through of State Diesel Taxes.
Langer, Ashley, Miller, Nathan H., 2013. Automakers’ short-run responses to changing gasoline prices. Rev. Econ. Stat. 95, 1198–1211.
Li, Shanjun, Timmins, Christopher, von Haefen, Roger H., 2009. How Do gasoline prices affect Fleet fuel economy? Am. Econ. J. Econ. Policy. 1 (2), 113–137.
Li, Shanjun, Linn, Joshua, Muehlegger, Erich, 2014. Gasoline taxes and consumer behavior. Am. Econ. J. Econ. Policy. 6 (4), 302–342.
Linn, Joshua, 2016. The rebound effect for passenger vehicles. Energy J. 37 (2), 257–288.
Liu, Weiwei, 2014. Modeling gasoline demand in the United States: a flexible semiparametric approach. Energy Econ. 45, 244–253.
National Highway Traffic Safety Administration (NHTSA), 2011. 2017-2025 Corporate Average Fuel Economy Compliance and Effects Modeling System
Documentation. US Department of Transportation, John A. Volpe National Transportation Systems Center, Cambridge, MA, December, available at
https://www.nhtsa.gov/sites/nhtsa.dot.gov/files/cafe model documentation 2011.12.03.pdf.
National Highway Traffic Safety Administration (NHTSA), 2020. CAFE Model Documentation. US Department of Transportation, John A. Volpe National
Transportation Systems Center, Cambridge, MA, March, DOT HS 812 934.
Pavlovic, Jelica, Cuiffo, Biagio, Fontaras, Georgios, Valverde, Víctor, Marotta, Alessandro, 2018. How much difference in type-approval CO2 emissions from
passenger cars in Europe can be expected from changing to the new test procedure (NEDC vs. WLTP)? Transp. Res. Part. A. Policy Pract. 111, 136–147.
Rivers, Nicholas, Schaufele, Brandon, 2015. Salience of carbon taxes in the gasoline market. J. Environ. Econ. Manage. 74, 23–36.
Rivers, Nicholas, Schaufele, Brandon, 2017. Gasoline price and new vehicle fuel efficiency: evidence from Canada. Energy Econ. 68, 454–465.
Rouwendal, Jan, 1996. An economic analysis of fuel use per kilometre by private cars. J. Transp. Econ. Policy., 3–14.
Sorrell, Steve, 2009. Jevons’ paradox revisited: the evidence for backfire from improved energy efficiency. Energy Policy 37 (4), 1456–1469.
Sorrell, Steve, Dimitropoulos, John, 2008. The rebound effect: microeconomic definitions, limitations and extensions. Ecol. Econ. 65, 636–649.
Tietge, Uwe, Mock, Peter, Franco, Vicente, Zacharof, Nikiforos, 2017. From laboratory to Road: modelling the divergence between official and Real-world
fuel consumption and CO2 emissions values in the German passenger car Market for the years 2001-2014. Energy Policy 100, 212–222.
UBA, Umweltbundesamt, No. 09/2020,
https://www.umweltbundesamt.de/en/press/pressinformation/speed-limits-on-motorways-clearly-cut-down-co2 2020. Speed Limits on Motorways
Clearly Cut Down CO2 Emissions.
Wehrmann, Benjamin, available at https://www.cleanenergywire.org/news/lower-taxes-diesel-cost-germany-seven-billion-euros-2019 2020. Lower
Taxes for Diesel Cost Germany up to Seven Billion Euros in 2019. Clean Energy Wire, Berlin, Germany.
21