How Cost-effective are Electric Vehicle Subsidies in Reducing Tailpipe-CO2 Emissions? An Analysis of Major Electric Vehicle Markets

We estimate the cost-effectiveness of plug-in electric vehicle (PEV) subsidies in reducing tailpipe-CO2 emissions in China, the U.S., and nine European countries. We find that the per-tonne cost of tailpipe-CO2 avoided increases linearly with the government-subsidized percentage of the PEV price. Costs are relatively higher in the Netherlands and Denmark, which subsidized high-priced PEVs including plug-in hybrids, and lower in the U.S., where PEVs replaced higher-emissions cars. Chinese PEV subsidies have a short-run static cost of up to $1,600 per tonne, far exceeding the social cost of carbon, suggesting that subsidies are more a part of China’s industrial policy than its carbon policy. When subsidy-induced PEV sales and power generation emissions are considered, the ordering of countries based on the cost-effectiveness of subsidies changes. The long-run dynamic subsidy cost is expected to be lower, as current subsidies may drive future innovation and sales, and due to grid decarbonization.


INTRODUCTION
The transportation sector accounts for 24% of the global greenhouse gas (GHG) emissions (IEA 2019).Within the transport sector, road transport is the most utilized mode because of its convenience (Van Essen 2008).However, it is also the most emissions intensive mode, accounting for 75% of global transport GHG emissions, with roughly 44% coming from road passenger vehicles alone (IEA 2019).
One way to lower passenger vehicle CO 2 emissions is through deployment of lower-and zero-tailpipe emission technologies including plug-in electric vehicles (PEVs) (Kalhammer et al. 2007).Demand-side fiscal policies represent one of the most commonly used policy levers for promoting deployment of PEVs (IEA 2019, Langbroek, Franklin, and Susilo 2016, Lévay, Drossi-nos, and Thiel 2017, Lieven 2015, Baldursson, Nils-Henrik, and Lazarczyk 2021).However, early evidence from the U.S., European and Canadian light duty vehicle (LDV) markets suggests that promoting deployment of PEVs through subsidies is expensive (Xing, Leard, and Li 2021, Sheldon and Dua 2018, 2019, Azarafshar and Vermeulen 2020, DeShazo, Sheldon, and Carson 2017, Miess et al. 2022).This paper further explores the evolution of tailpipe-CO 2 emissions avoided as well as subsidy cost per tonne of tailpipe-CO 2 avoided across a range of major PEV markets from 2010 to 2017.In particular, we focus on China, the U.S. and nine major European countries, which are currently the leaders in the PEV market, to determine the spatio-temporal evolution of the subsidy impact and cost-effectiveness.
Given the lack of literature on comparisons between PEV markets in different countries, we identified the following research questions for detailed investigation: • How much tailpipe-CO 2 emission has been avoided through PEV adoption in these different countries, both in absolute terms (million tonnes) and relative terms (percentage)?• What is the subsidy cost per tonne of tailpipe-CO 2 avoided and how does it vary across the different countries?• How do the cost numbers change when the extent of PEV sales induced by the subsidy are taken into account?• How do the cost numbers change when the emissions from electricity generation used for powering PEVs are taken into account?• How can we improve the cost-effectiveness of the subsidy policy, especially for the subsidies that are a part of the COVID-19 stimulus packages?• What are the policy lessons that different countries can learn from one another?
Existing literature focuses on a few individual markets while considering only a few years of data, including our own previous work on U.S. and Chinese markets (Sheldon andDua 2020, 2019).In particular, in our previous work we were only able to analyze one year of data, 2015 for the U.S. (Sheldon and Dua 2019) and 2017 for China (Sheldon and Dua 2020).Thus, the range of PEV market share penetration that our previous work explored was rather limited, ranging from 0.81% in U.S. to 2.47% in China.Furthermore, the temporal evolution of markets was not explored.This paper draws insights from our and other prior work 1 and contributes to the existing PEV subsidy literature by making cross-country comparisons using detailed annual micro-level data from 11 countries, and looking at the temporal evolution over 8 years from 2010-2017.It compares countries with very different PEV market shares ranging from 1 to 40 percent and subsidy percentages (percentage of PEV price subsidized by the government) ranging from 0 to 55 percent.Through these comparisons, it provides a broader view of how trends and various metrics such as-the extent of CO 2 avoided and subsidy cost-effectiveness-may evolve with changes in subsidy amounts, PEV prices, and market shares.Overall, this study offers the most comprehensive set of cost-effectiveness estimates for multiple countries in the literature.It also provides insights on some of the factors that contribute to differences in cost-effectiveness figures across countries, as well as lessons that can be drawn from them.
Finally, in our previous work we estimated the subsidy cost-effectiveness in terms of subsidy dollars per gallon of gasoline saved, which can be translated to subsidy dollars per tonne of tailpipe-CO 2 avoided.Here, unlike our previous work, we also take into account the emissions from the combustion of fuels associated with the generation and distribution of electricity used to power these vehicles.
The rest of this paper is organized as follows: Section 2 provides a brief background on the CO 2 emissions associated with the LDV sector.Section 3 showcases the rich dataset utilized in this study and provides summary statistics for the various vehicle attributes broken down by country and year.Section 4 presents details on the methodology applied in this study together with highlighting all the previous literature where the approach has been utilized.It also lays out the mathematical formulation for the impact and cost-effectiveness calculations.Section 5 provides a detailed discussion on the results obtained from the analysis.Section 6 highlights the policy implications of our findings.Section 7 specifies the caveats of this work.Section 8 highlights the conclusions drawn from the findings of this paper.

BACKGROUND
Internal combustion engine vehicles emit both smog-forming pollutants and greenhouse gases (GHGs) from their tailpipes.CO 2 makes up roughly 99% of total tailpipe-GHG emissions (U.S. EPA 2019).Battery electric vehicles, on the other hand, do not produce any tailpipe emissions.Emissions are produced though from the combustion of fuels associated with the generation and distribution of electricity used to power these vehicles.Beyond the emissions related to fuel combustion, there are emissions related to extraction, refinery, transport of fuel and vehicle manufacturing.These emissions have not been included in this analysis because of lack of such data for all the countries and for the time period considered in this analysis.Accounting for these emissions is likely to lower the subsidy cost effectiveness.Moreover, the additional benefits of supporting PEVs, such as non-linearities in development of the technologies have not been accounted for in this analysis, thereby highlighting that the cost assessment in this study represents short-run static costs.In other words, short-run static costs only account for the impact of current subsidies on current PEV sales.They do not take into account the current subsidies' impact on future PEV sales.Increased current PEV sales as a result of current subsidies encourages future PEV sales as well.They achieve this by influencing both future PEV prices and consumer perception.Higher current sales from PEV subsidies can contribute to lower future PEV prices through learning-by-doing, economies of scale, and/ or induced innovation effects (Gillingham and Stock 2018).Furthermore, increased current PEV sales can positively influence future consumer perceptions of PEVs in terms of popularity, quality, and reliability, potentially encouraging future PEV adoption.Increased current PEV sales can influence future PEV adoption via the neighborhood effect (Chakraborty et al. 2021).To summarize, failing to account for the long-run effects of current subsidies in promoting future PEV sales renders short-run static cost assessments inherently limiting.Finally, here we focus on reduction in GHG emissions only and do not consider the reduction in local pollution from particulate matter in urban settings through increased PEV adoption.

DATA
Our main dataset is a rich panel of passenger car sales and characteristics for the U.S., European and Chinese car market, obtained from JATO Dynamic Limited.The dataset includes the sales, prices, and product characteristics for every new passenger car sold during 2010-2017 in the U.S., China and nine European countries including Norway, the Netherlands, Sweden, Italy, Spain, Great Britain, Denmark, France, and Germany.The datasets for U.S. and China also include regional information.
Each car is defined at the make-model-powertrain-body type level, e.g., Honda Accord hybrid sedan.Sales are defined as new car registrations.Prices are manufacturer suggested retail prices (MSRPs), excluding any taxes.2Car characteristics include measures of vehicle size (wheel base, width, length, height and kerb weight), horsepower and tailpipe-CO 2 emissions.Among the different countries, the U.S. new vehicle fleet is the largest, tallest, heaviest, strongest (in terms of horsepower) and dirtiest (in terms of tailpipe-CO 2 emissions).On the other hand, the fleets in Denmark and the Netherlands tend to be the smallest, shortest, lightest, weakest (in terms of horsepower) and cleanest (in terms of tailpipe-CO 2 emissions) for most years.

Summary statistics
For most years, the fleets in Norway, Sweden and Germany tend to be the priciest (excluding taxes).The U.S. fleet became more expensive over time (excluding taxes) and overtook its European counterparts in 2015 and 2016.While all the fleets are becoming greener over time (in terms of PEV market share), the Norway fleet is the greenest followed by the Netherlands and Sweden.While the market shares for both BEVs and PHEVs are rising over time, there is no clear trend of one of these powertrain types dominating the other.
The respective plots summarizing the statistics for the PEV fleets in the different countries are given in the online appendix.Based on figures A1-A2, in recent years, the Chinese PEV fleet is the lightest, cheapest, cleanest and smallest amongst all.On the other hand, the PEV fleets in the U.S. and the Netherlands are relatively heavier, costlier, dirtier and bigger.Looking at the ratios of the PEV fleet to the overall fleet (figures A3-A4) suggests that the PEV fleet is costlier and cleaner than the overall fleet in all countries across all years.
For European countries, we obtain the average subsidy amounts per PEV from Münzel et al. (2019).For the U.S., both the federal and state level subsidy amounts per PEV were obtained from the U.S. department of energy websites (U.S. DOE-EPA , AFDC).For China, the federal subsidy amounts were obtained from Qian (2018) and ICCT (2017) and the 2015 province level subsidy amounts including registration incentive were obtained from ICCT (2018) and were assumed to be the same from 2013 to 2017.A Chinese vehicle tax exemption of 10% for PEVs was also included (Argus 2020).The average subsidy numbers for all the countries from 2010-2017 are plotted in figure 3d.Denmark, Norway and China lead in the total amount of subsidies offered for PEVs.While the generous waivers of point of sales tax dominates in Norway and Denmark, for China it is the combination of federal, provincial and registration incentives in vehicle ownership-restricted cities.

METHODOLOGY
To estimate the tailpipe-CO 2 emissions of the counterfactual vehicle fleet in the absence of the subsidy, we assume that a PEV buyer would have bought the average ICEV of the same body type as the PEV.While simplistic versions of this counterfactual vehicle assumption have been utilized extensively in the literature (Holland et al. 2016, Archsmith, Kendall, and Rapson 2015, Graff Zivin, Kotchen, and Mansur 2014, Sheldon and Dua 2018), it does limit precise estimation of the tailpipe-CO 2 savings.Alternative approaches to the commonly used conventional counterfactual approach include choice-model based counterfactuals (Xing, Leard, andLi 2021, Sheldon andDua 2018).In fact our own previous research suggests that detailed choice-model based counterfactuals certainly allow for a more precise estimation of the fuel savings (equivalent to CO 2 savings) (Sheldon and Dua 2018).However, given that the focus of this work was to get ballpark trend estimates by analyzing countries with varied market share penetrations, we chose to utilize a detailed version of the conventional counterfactual vehicle assumption.Indeed, as highlighted later on in the results section, the scale of percentage CO 2 savings dwarfs the variations related to the counterfactual vehicle assumption.
In addition, consumer preferences for body-type are likely to be heterogeneous, giving rise to obvious differences in substitutability across vehicles (Bunch and Chen 2007).Thus, we chose PEV substitution with average ICEV within the same body type, instead of the simple substitution with average ICEV.Finally, it is worth pointing out that the cost per additional tonne of tailpipe-CO 2 avoided (as discussed later in this methodology section) incorporates information on the elasticity of PEV market share with respect to PEV subsidy from the literature that were estimated using standard econometric approaches.
We calculate the sales-weighted average characteristics for both the actual and the counterfactual fleets as well as their respective components.More specifically, we calculate the fleet tailpipe-CO 2 average emissions for the new vehicle fleet including PEVs. 3 Next, we replace the PEVs in the new vehicle fleet using the conventional counterfactual approach.Then, we calculate the fleet tailpipe-CO 2 average for this counterfactual fleet.
Next, we calculate the percentage of tailpipe-CO 2 emissions reduction in the new vehicle fleet.It is calculated using equation 1 as follows: 4 3. For PHEVs, a utility factor of 0.5 is assumed for calculating the combined tailpipe-CO2 emissions (mean of tailpipe-CO2 emissions in electric mode and internal combustion engine mode).
4. Refer to the online appendix for information on how the equation is derived.Finally, using equation 2, we estimate the cost per tonne of tailpipe-CO 2 avoided by dividing the total subsidy cost by the total tailpipe-CO 2 avoided over the lifetime of the vehicle: 5 5.The annual mileages in Europe, the U.S. and China were assumed to be 9,072 miles (Ricardo-AEA 2014), 11,500 miles (Federal Highway Administration (FHWA) 2019) and 8,787 miles (Sheldon and Dua 2020), respectively.The vehicle lives in Europe, the U.S. and China were assumed to be 15 years (Ricardo-AEA 2014), 16 years (Davis, Diegel, and Boundy  2013) and 14  (2) represents the sales-weighted average fleet tailpipe-CO 2 level (g/mile) for the PEV fleet, Subsidy (%) represents the subsidy percentage on electric vehicles, MSRP represents the manufacturer suggested retail price for the vehicle and Tax represents the overall tax applied on the vehicle.
To calculate the cost per tonne of additional tailpipe-CO 2 avoided, we combine equation 3 with the extent of electric vehicle sales induced by the subsidy, i.e., elasticity of PEV market share with respect to PEV subsidy.This is because an additional tonne of tailpipe-CO 2 avoided as a result of the subsidy is equal to the tonne of tailpipe-CO 2 avoided in the case of no PEVs multiplied by the elasticity value.It is based on the following reasoning.Suppose that the number of tonnes of tailpipe-CO 2 avoided in the case of a 100% drop in PEV sales is denote by t.Then, the additional tonnes of tailpipe-CO 2 avoided in the case of a x% drop in PEV sales because of the removal of subsidy can be denoted by t*x/100, as shown below in equation 5.As shown in the results section, since tailpipe-CO 2 avoided varies linearly with the PEV market share, the assumption is justified.
( ) For Europe, we calculate the elasticity of PEV market share with respect to PEV subsidy using the point estimates for the effect of incentives from (Münzel et al. 2019).Münzel et al. (2019) estimate that an incentive of 1,000 Euros increases PEV sales by about 5-7% on average, all else being equal.A change of 1,000 Euros in subsidy translates to a change of 13.07% based on the sales-weighted average subsidy of 7,650 Euros for the European countries considered in this analysis.Thus, the elasticity of change in PEV market share to change in subsidy is roughly 0.459 (6%/13.07%).For the U.S., we use an average elasticity of PEV market share with respect to PEV subsidy of 0.4 taken from Li et al. (2017).For China, no such estimates for elasticity of PEV market share with respect to the combined PEV subsidy are available in literature and thus China was excluded from this piece of the analysis.It is also worth noting that the elasticity estimates taken from literature account for the fact that the subsidy is not the same for all PEV models.This is because the estimation of these elasticity values in the literature accounted for variation in subsidies for different PEV models.
It is worth highlighting that the cost effectiveness prior to incorporating elasticity values is the cost per tonne of tailpipe-CO 2 avoided.After incorporating the elasticity information, the cost effectiveness represents the cost per additional tonne of tailpipe-CO 2 avoided.The two separate cost effectiveness estimates are provided to separate the elasticity effect from other inputs.

RESULTS AND DISCUSSION
In this section, we analyze how the subsidy cost-effectiveness varies with other observable indicators and across countries.This includes the variation in the subsidy cost per tonne of tailpipe-CO 2 avoided with the percentage of PEV price subsidized by the government.It starts with a detailed examination of the many elements that go into exploring this variation.These include: tailpipe-CO 2 avoided, vehicle price, subsidy amounts, and the validation and impact of the bodytype equivalence assumption on subsidy costs.Finally, in this section, we investigate how much the subsidy costs increase as a result of accounting for both the extent of electric vehicle sales induced by the subsidies and the emissions from electricity generation.
We begin by confirming the appropriate application of the body-type equivalence assumption in our counterfactual approach.We examine the ratio of the average new vehicle fleet characteristics for the PEV fleet and the replaced fleet, both normalized by the non-PEV fleet, as shown in Figure A5.The vehicle dimensions considered include length, width, height and wheelbase.The closeness of the ratios for the PEV fleet and the replaced fleet to the 45-degree line is reflective of the body-type equivalence assumption in our counterfactual approach.In other words, figure A5 highlights that the characteristics of the actual PEV fleet are very similar to the replaced fleet characteristics, as expected based on our conventional counterfactual approach.
Figure 4 shows the same ratio for average (a, c) tailpipe-CO 2 , and (b, d) price.We plot the average tailpipe-CO 2 for the panel to visually confirm how clean the PEV fleet is in comparison to the fleet it replaces, as well as whether the replaced fleet is cleaner or dirtier than the average non-PEV fleet.There are three important things to note from figure 4(a, c 6 ).First, the PEV fleet has significantly lower (near-zero) tailpipe-CO 2 emissions compared to the non-PEV fleet.This is because the PEV fleet in most countries is dominated by BEVs that have zero-tailpipe emissions.Second, with the exception of the Netherlands, the tailpipe-CO 2 fleet average ratio for the replaced fleet is less than or close to one in the majority of countries.This means that the replaced fleet is more efficient than the average non-PEV fleet.This indicates that if we had not used the body-type equivalence assumption, we would have overestimated the tailpipe-CO 2 savings while underestimating the cost per tonne of tailpipe-CO 2 avoided.In other words, the traditional application of the conventional counterfactual approach in the literature, which does not include the body-type equivalence assumption, leads to an overestimation of subsidy cost-effectiveness.Third, the PEV fleet has lower tailpipe-CO 2 emissions than the replaced fleet given that the data lies well below the 45-degree line.The lower it is below the 45-degree line, the greater the tailpipe-CO 2 emissions avoided by PEV substitution.
6. Figure 4c is simply a zoomed-in version of Figure 4a.
Figure 4(b, d 7 ) depicts the average MSRP for the panel to visually confirm how expensive the PEV fleet is in comparison to the fleet it replaces, as well as whether the replaced fleet is more or less expensive than the average non-PEV fleet.The graph also gives a general idea of whether a country's PEV subsidies are aimed at luxury or economical PEVs.The figure highlights three significant points.First, as seen by the y-axis in figure 4b, the PEV fleet is more expensive on average than the non-PEV fleet.Second, the PEV fleets are typically more expensive than non-PEV fleets they replace, as indicated by the data points lying above the 45-degree line.The further the data deviate above the 45-degree line, as in Denmark and the Netherlands, the greater the PEV price premium in these countries, suggesting subsidization of more expensive or luxury PEVs.Subsidizing expensive luxury PEVs may not be the most cost-effective use of government funds.This is due to the fact that buyers of such high-priced PEVs are typically high-income consumers whose PEV purchasing 7. Figure 4d is simply a zoomed-in version of Figure 4b.decisions are unlikely to be influenced by the subsidy (Sheldon and Dua 2019).Third, the replaced fleet has similar or slightly lower MSRP than the average non-PEV fleet for most countries other than the Netherlands.This is highlighted by the fact that the ratio of replaced fleet to non-PEV fleet (x-axis) is lower than one for most countries.

Tailpipe-CO 2 emission avoided through PEV adoption
We next compute the cumulative and percentage of tailpipe-CO 2 emissions avoided by PEV adoption in the new fleet.We perform these calculations to get a rough estimate of how many cumulative tailpipe-CO 2 emissions have been avoided as a result of PEV adoption, because the cli-mate change mitigation impact is proportional to the cumulative CO 2 avoided.For the percentage of tailpipe-CO 2 emissions avoided, we use the broad nature of our panel data to see if there is a simpler way to approximate this percentage based solely on the PEV market share in a country.In other words, we investigate the relationship between the percentage of tailpipe-CO 2 emissions avoided and PEV market share, which could be used to predict the percentage of tailpipe-CO 2 emissions saved in another country based solely on its PEV market share.The results are shown in figure 5.
Figure 5 illustrates five important points.First, in 2017, the total cumulative tailpipe-CO 2 avoided as a result of PEV adoption in these eleven countries was roughly 5.63 million tonnes (0.51% in relative terms 8 ).To put these numbers in context, Bayer and Aklin (2020) estimated that the European Union emissions trading system, the largest carbon market in the world, reduced emissions by a cumulative 1.2 billion metric tons per year between 2008 and 2016, an average of 133 million metric tons per year.Second, the highest tailpipe-CO 2 savings were seen for the U.S. followed by China.This is expected given that China and U.S. represent the top two markets in terms of overall sales.Moreover, since the U.S. fleet tends to be the dirtiest and most driven, the expected tailpipe-CO 2 savings from switching to PEVs are found to be the highest.
8. The cumulative tailpipe-CO2 numbers are based on the assumption that all of the new vehicles sold since 2010 are still running on the road in 2017, in line with the average vehicle life assumption for all these countries.Third, we find that the percentage of tailpipe-CO 2 avoided through PEV substitution varies linearly with PEV market share.This finding implies that for a country where detailed data on new vehicle fleet is unavailable, the PEV market share in that country could be used as a proxy to estimate the percentage of tailpipe-CO 2 avoided through PEV substitution.To be more specific, the calculated percentage of tailpipe-CO 2 avoided closely follows the special case curve, as can be seen in figure 5b. 9  Fourth, we also compute the percentage deviation from the special case curve depicted in figure 5c.The percentage deviation from the special case curve appears to be large, especially 9.The special case represents a hypothetical scenario in which: (i) the PEV fleet consists exclusively of BEVs, so Fleet CO PEV Fleet 2 = 0, and (ii) the replaced fleet has the same characteristics as the average non-PEV fleet, i.e.,

Fleet CO
Fleet CO placed Fleet non PEV Fleet 2 2

Re
. By substituting these values into equation 1, we obtain ' 100 1 x x PEV PEV / ( )). Figure 5b's dashed curve corresponds to this special case.A close examination of the dashed curve reveals that at higher PEV market share values (roughly above 10 percent), the rate of increase in the percentage of tailpipe-CO2 avoided with PEV market share is greater than the rate of increase at lower PEV market share values (below 10 percent ).Consideration of the special case curve is warranted because the data points appear to follow it more closely than at low market shares, reaching -37 percent in the case of the U.S. at a PEV market share of 0.4 percent.However, at such low PEV market shares, the percentage of tailpipe-CO 2 avoided is low.In the above-mentioned U.S. data point, it is 0.24 percent.This means that even a large percentage deviation from the special case curve has little effect on the overall finding that the percentage of tailpipe-CO 2 avoided closely follows the special case curve.
Fifth, two factors contribute to larger negative deviations from the special case curve visible in figure 5c.First, higher proportion of PHEVs in a country's PEV mix lead to non-zero η values.This helps to explain the high percentage deviation of roughly -38 percent for Sweden, where PHEVs dominated the PEV market in Sweden in 2010, as be seen in the accompanying excel file for figure 3. Second, the replaced non-PEV fleet on average being cleaner compared to the entire non-PEV fleet, i.e., countries with a lower ratio of or where PEVs tend to replace relatively cleaner vehicles.This helps to explain the case of the U.S., where this ratio ranged between 0.7 and 0.8, as shown in figure 4c.

Subsidy cost per tonne of tailpipe-CO 2 avoided
Figure 6 shows the variation in the cost per tonne with subsidy percentage for PEVs for the different countries.There are three things worth noting in figure 6.First, the cost per tonne varies almost linearly with subsidy percentage for PEVs.This is in line with expectation, as high subsidy percentage results in high subsidy costs.Ideally, higher subsidies should also induce more PEV purchases and thus a greater CO 2 emissions reduction.The CO 2 emissions reduction varies linearly with PEV market share, as can be seen in figure 6b.However, we find the PEV market share does not increase with increasing subsidy, as can be seen in figure A6 in the online appendix.This intuitively explains why the cost per tonne varies almost linearly with subsidy percentage for PEVs.To get a rough idea of the elasticity of the cost per tonne with respect to subsidy percentage, we run a panel regression on the logarithms of the two variables while controlling for time-and country-fixed ef- fects.It yields an elasticity of 0.90, which is statistically significant at the 1% level.It implies that a 1 percent increase in a country's subsidy percentage for PEVs results in a nearly proportional increase of 0.9 percent in the cost per tonne of tailpipe-CO 2 avoided.A visual examination of figure 6 reveals that some countries appear to be above and below this trend.These countries include the Netherlands and Denmark, which appear to be above the trend.The United States, on the other hand, is below this trend.
The Netherlands' higher cost per tonne values are most likely due to its higher market share of PHEVs (as shown in figure 3c) and its subsidization of higher-priced PEVs (as shown in figure A2) (b).The Netherlands and Sweden have similar subsidy percentage values, but the average amount of subsidy offered in the Netherlands was higher than in Sweden.This is because the subsidies in the Netherlands went to higher-priced PEVs than in Sweden.A higher subsidy amount equates to a higher per-tonne cost.Furthermore, in the Netherlands, subsidies were targeted at PHEVs, which have a lower CO 2 reduction potential than BEVs.Overall, comparing the Netherlands to other countries reveals that subsidizing PHEVs and high-priced PEVs is not a cost-effective way to reduce tailpipe-CO 2 emissions.
A similar conclusion emerges when comparing Denmark with Norway and France.While Denmark's subsidy percentage (as shown in Figure 6) is comparable to that of Norway and France, the average subsidy offered in Denmark is on the higher end (as can be seen in figure 3d).Because the PEVs sold in Denmark were the higher-priced PEVs (as can be seen in figure A2 in the online appendix), the subsidy percentage for PEVs was similar to that of Norway and France, despite the higher subsidy amounts in Denmark.The higher subsidy amounts in Denmark resulted in a higher cost per tonne.
For the case of U.S., both the average subsidy amount and the subsidy percentage values are comparable to Great Britain and Spain, as shown in figures 3 and 6, respectively.As a result, one could expect the cost per tonne in the U.S. to be similar to that of the Great Britain and Spain.For the U.S., however, it is much lower.This can be explained by the fact that the average non-PEV fleet being replaced in the U.S. is dirtier than the non-PEV fleet being replaced in the Great Britain and Spain, as shown in figures 2a and 4a.This suggests that PEV subsidies are likely to be a more cost-effective policy for countries with a dirtier base fleet.
China, on the other hand, is at the extreme end of the spectrum in terms of both subsidy percentage and cost per tonne.The high subsidy percentage is due to the fact that the average subsidy offered in China is among the highest, as shown in figure 3, and PEVs sold there are among the cheapest, as shown in figure A2.Because of the high average subsidy amount, the cost per tonne is also high.When the cost is adjusted for relative purchasing power parity, it increases even more.In other words, compared to both the U.S. and Europe, China spends the most on subsidizing tailpipe-CO 2 emissions reduction through PEV deployment.The high costs suggest that in China, car electrification is not merely a means of reducing tailpipe-CO 2 emissions, but also an industrial policy.It is also designed to help China catch up with other nations in the global car market, which is dominated by Japan, Germany, and the United States, notably in the production of internal combustion engine cars.This may explain China's aggressive subsidy policies to some extent.
Second, the cost per tonne of tailpipe-CO 2 avoided can be as high as $1,600/tonne (on PPP basis in 2010 $), with a weighted 10 average cost of ∼$739/tonne (on PPP basis in 2010 $). 11To provide context, we compare these cost numbers based on the detailed conventional counterfactual approach to the numbers from our previous papers based on a choice model counterfactual approach.
Using a choice model-based counterfactual approach applied to 2015 data, our previous published work for the United States estimated annual gasoline savings from PEVs of 0.04 billion gallons (Sheldon and Dua 2018).This translates to 0.64 billion gallons of gasoline saved over a vehicle's lifetime of 16 years.We combine this number of gallons of gasoline saved with the total subsidy budget calculated in this work.We are unable to use the subsidy budget figures from our previous work because they did not include regional subsidies.Dividing the overall subsidy expenditure by the total gallons of gasoline saved yields a cost per gallon of gasoline saved of $1.41.Using the EPA's greenhouse gas equivalence factors to convert gallons of gasoline saved to tonnes of tailpipe-CO 2 avoided (EPA 2019) and converting to 2010 dollars, we arrive at a cost of $173 per tonne of tailpipe-CO 2 avoided.Based on the detailed conventional counterfactual assumption used in this paper, the corresponding figure for 2015 is $193 per tonne of tailpipe-CO 2 avoided (in 2010 dollars).
To make a similar comparison for China, we divide the total subsidy budget calculated in this work by the amount of gasoline saved, 6.66 billion liters, estimated using the choice model-based approach from our previous work using 2017 data (Sheldon and Dua 2020).We use the subsidy budget numbers from this work because our previous work's subsidy budget numbers did not include provincial subsidies, registration fees, or tax exemptions.We use the EPA's greenhouse gas equivalence factors to convert gallons of gasoline saved to tonnes of tailpipe-CO 2 avoided (EPA 2019).Dividing the total subsidy expenditure by the total tonnes of tailpipe-CO 2 avoided yields a cost of $1229 per tonne (on PPP basis in 2010 dollars).Based on the detailed conventional coun-10.Using PEV sales data as weights 11.A sensitivity analysis was performed to explore the impact of the assumed PHEV utility factor of 0.5 as well as PEV lifetime mileage.Using 10% lower values for PHEV utility factors and PEV lifetime mileages produced elasticity estimates of cost per tonne of tailpipe-CO 2 avoided of -0.05 and -1.11, respectively.
terfactual assumption used in this paper, the corresponding figure for 2017 is $1252 per tonne of tailpipe-CO 2 avoided (on PPP basis in 2010 dollars).
Overall, the cost-effectiveness figures in this analysis appear to be consistent with previous single-year estimates for these two countries from our previous papers.
Third, the cost-effectiveness is roughly more than an order of magnitude higher than the social cost of carbon. 12The high cost could be attributed to both the high subsidy amounts per PEV as well as the low extent of tailpipe-CO 2 avoided.Note that the cost per tonne of CO 2 avoided would be higher than our estimate for two reasons: (1) we are only considering tailpipe-CO 2 emissions, and (2) we assume that the elasticity of PEV market share with respect to PEV subsidy is equal to one.

Subsidy cost per additional tonne of tailpipe-CO 2 avoided (accounting for induced sales)
The cost estimate increases further by accounting for the actual extent of electric vehicle sales induced by the subsidy (versus sales that would have occurred even without the subsidy), i.e. elasticity of PEV market share with respect to PEV subsidy.Using equation 8, we combine the results from figure 6 with the extent of electric vehicle sales induced by the subsidy to obtain the cost per tonne of additional tailpipe-CO 2 avoided. 13Figure A7 in the online appendix shows the variation in the cost per additional tonne of tailpipe-CO 2 avoided.Combining the results for Europe and the U.S., we get a PEV sales-weighted average cost of $701 per additional tonne of tailpipe-CO 2 avoided (on PPP basis in 2010 $). 14,15This is in contrast to the average across Europe and the U.S. of $309 per tonne of tailpipe-CO 2 avoided (on PPP basis in 2010$).Thus, when taking into account the fact that the elasticity of PEV market share with respect to PEV subsidy is less than half (the majority are allocated to consumers who would have purchased the PEV regardless), the cost per tonne of emissions avoided more than doubles.
Furthermore, the results from Xing, Leard, and Li (2021) suggest that the conventional counterfactual approach overestimates the environmental benefits by 27 percent, relative to choicemodel based counterfactual approach.Their numbers translate to an underestimation of the cost of CO 2 emission reduction by the conventional counterfactual approach by roughly 29%.Adjusting for 12.The social cost of carbon (SCC) is an estimate of the net present value of economic damage that would result from emitting one additional tonne of CO 2 into the atmosphere.It puts the effects of climate change into terms that help policymakers and other decision makers understand the economic impacts of decisions that would increase or decrease emissions.Based on the Obama Administration's central case numbers, the 2020 SCC is closer to $54/tonne in today's dollars (United States Government 2016).However, the Obama Administration emphasized the uncertainty in the numbers and discouraged the use of a single figure .13.For China, no such estimates for elasticity of PEV market share with respect to the combined PEV subsidy are available in literature and thus we exclude China from this part of the analysis.
14.The findings from the detailed conventional counterfactual assumption appear to be in line with the findings from the choice-model based counterfactual measured for the case of U.S. for which model-level subsidy information was available unlike the Europe data.For 2017 data, applying a choice model based counterfactual resulted in a federal-subsidy cost of $3.2 per additional gallon of gasoline saved (Tamara L. Sheldon, Omar Al Harbi, and Dua 2020b, a).Using greenhouse gas equivalence factors from the EPA (2019) and converting to 2010$, it translates to a cost of $403 per additional tonne of tailpipe-CO 2 avoided.The corresponding number based on the detailed conventional counterfactual assumption used in this paper comes out to be $438 per additional tonne of tailpipe-CO 2 avoided (in 2010$).While there is ~8.5% difference in the cost estimates, however, given that the cost numbers are an order of magnitude higher than the social cost of carbon, the precision obtained through choice-model based approach is almost immaterial.
15.A sensitivity analysis was performed to explore the impact of the elasticity of PEV market share to PEV subsidy.Using 10% lower values for the elasticity of PEV market share produced an elasticity estimate of cost per additional tonne of tailpipe-CO 2 avoided of -1.11.ages, policymakers could opt for targeted subsidy designs.Targeted design is considered as one of the "three T" principles-"timely, temporary, and targeted"-for designing an effective stimulus package (Elmendorf andFurman 2008, Bordoff 2020).Our previous research suggests that targeted subsidy designs based on consumer income can improve PEV subsidy policy cost-effectiveness twofold (Sheldon andDua 2020, 2019).Real world pilot tests involving targeted subsidy designs for vehicle retirement and replacement have also proven to be more cost-effective in inducing additional PEV sales (Sheldon and Dua 2019a).
Despite potential improvements that can be achieved through targeted designs, PEV subsidies are likely to remain much less cost-effective compared to a gasoline tax and even supply-side policies such as fuel economy standards, which the literature suggests costs between $18-47 and $48-310 per tonne of CO 2 avoided respectively (Gillingham and Stock 2018).

Policy lessons across countries
Another policy implication worth noting is that countries such as Germany that only recently (2016) began offering subsidies for PEV deployment (and lower subsidy per PEV than most other countries, at that), could benefit from the early cycles of subsidization offered by other countries in their local markets.Higher local subsidization results in increased local demand.However, given the global nature of the automotive and battery manufacturing industries, local subsidies can result in global benefits in terms of economies of scale and learning-by-doing related to both battery and PEV manufacturing.Reduced battery and PEV manufacturing costs could also help lower subsidization costs of local PEV deployment for late entrant countries relative to early entrant countries.The solar photovoltaic (PV) industry is a case-in-point in this regard.In the case of solar PVs, research suggests that a subsidy in one country assists in increasing adoption elsewhere, too, because it increases investment in innovation by international firms (Gerarden 2017).In particular, the German solar PV consumer subsidies are argued to have also subsidized lower-cost solar for the rest of the world (Gillingham and Stock 2018).It is worth noting that Germany, together with Japan, was among the first countries to lead the solar PV manufacturing and deployment through its generous local subsidies (Quitzow 2015).However, Germany's early mover advantage in the PV manufacturing space did not pan out as expected once China (a late entrant) flooded the global PV market while banking on the initial technology leaps made by both Japan and Germany and the higher consumer subsidies in Germany (Quitzow 2015).Given Germany's experience in the PV sector and its extensive expertise in automobile manufacturing, it could be argued that German policymakers were well placed to play the waiting game until now in subsidizing local PEV manufacturing and deployment.

CAVEATS
Recall that the elasticity of PEV market share with respect to PEV subsidy for the U.S. and Europe is taken from prior studies that estimated it using standard econometric approaches.It is worth noting that the elasticity estimates 18 for the U.S. (0.4) and Europe (0.459) are in the similar range.Given the lack of literature on the variation of this elasticity with PEV market share, we assume these respective single point estimates for the U.S. and Europe stay constant over the spectrum of PEV market share considered in our analysis.Although the single point estimates for Europe, where the market share has varied the most, are based on an analysis covering the 2010-2017 time period, this is still a limiting assumption.This is because from a policy perspective, the subsidy effectiveness in inducing additional sales is expected to decrease with increasing PEV market share.The policy expectation is that the PEV market will reach an inflection point and become self-sustaining in its growth, beyond which the subsidies could be discontinued without any significant impact on the market (Jenn et al. 2020).
It is worth noting that most of these countries have other supply-side policy measures in place, such as the corporate average fuel economy standards and/or tailpipe-CO 2 emission standards.Since the automakers would have had to comply with the tailpipe-CO 2 emission standards even in the counterfactual case, the estimates on tailpipe-CO 2 avoided from PEV adoption in this paper represent an upper limit.Correspondingly, our subsidy cost per tonne of tailpipe-CO 2 avoided represent a lower estimate.
Moreover, given that today's subsidies could also encourage future technological innovation and sales, the cost assessment in this study represents short-run static costs.For example, as more consumers adopt PEVs, economies of scale could lead to lower prices, accelerating PEV adoption.Long-run dynamic cost could differ due to spillovers (e.g., learning by doing) and peer effects.That being said, the suggested targeted subsidy designs focused on inframarginal adopters would not only reduce the short-run static costs but also allow the subsidy programs to run longer with the same overall subsidy budget.In fact, in targeted designs, the subsidy amount could increase over time so as to encourage uptake among late-adopters with relatively lower willingness-to-pay compared to innovators and early adopters, as suggested by technology adoption theory.Recent research focused on residential solar in California shows that the efficient subsidy increases over time (Langer and Lemoine 2018).Alternatively, as suggested by previous literature, subsidizing charging infrastructure deployment might be more cost-effective than subsidizing electric vehicle deployment (Li et al. 2014, Springel 2017).
Finally, the paper uses average CO 2 emissions factors for electricity consumed instead of the marginal emissions factor to calculate the subsidy cost per additional net tonne of tailpipe-CO 2 avoided.The average electricity emissions factor may be higher or lower than the marginal electricity factor, and thus the subsidy cost per additional net tonne of tailpipe-CO 2 avoided will vary accordingly.Furthermore, as low-carbon electricity generation technologies become more widely adopted, the subsidy cost per additional net tonne of tailpipe-CO 2 avoided is expected to fall in the future.

CONCLUSION
We find that reducing tailpipe-CO 2 emissions by PEV adoption subsidies is expensive.Combining the extent of tailpipe-CO 2 avoided with the subsidy costs, we find that a conservative estimate of the cost per tonne of tailpipe-CO 2 avoided can be as high as $1,600 (on PPP basis in 2010 $).Considering the variation across different countries, we find that the cost per tonne varies almost linearly with subsidy (as a percent of vehicle price) for PEVs, with the highest cost per tonne seen for China, followed by Denmark and Norway.The policy cost is more than an order of magnitude higher than the social cost of carbon.The estimated cost per tonne of tailpipe-CO 2 avoided becomes even higher when taking into account the actual extent of electric vehicle sales induced by the subsidies, i.e., elasticity of PEV market share with respect to PEV subsidies.
The high tailpipe-CO 2 emissions reduction costs of PEV subsidy policies warrants research into and adoption of innovative subsidy designs to improve their cost-effectiveness.Innovative targeted PEV subsidy designs could be incorporated in the COVID-19 economic stimulus packages that are being currently considered in different parts of the world for promoting PEV adoption.Targeted designs based on either consumer income or vehicle price represent viable cost-effective alternatives.

Figures 1 ,
Figures 1, 2 and 3 provide summary statistics for various vehicle attributes broken down by country and year.In particular, figure 1 includes new vehicle sales-weighed fleet dimensions including: (a) length, (b) width, (c) height and (d) wheelbase.Figure 2 shows new vehicle sales-weighed fleet average for the attributes including: (a) tailpipe-CO 2 (g/km), (b) MSRP, (c) kerb weight and (d) horsepower.Figure 3 shows the evolution of the (a) PEV, (b) battery electric vehicle (BEV) and (c) plug-in hybrid electric vehicle (PHEV) market shares in each country over time together with the average PEV subsidy amounts.Among the different countries, the U.S. new vehicle fleet is the largest, tallest, heaviest, strongest (in terms of horsepower) and dirtiest (in terms of tailpipe-CO 2 emissions).On the other hand, the fleets in Denmark and the Netherlands tend to be the smallest, shortest, lightest, weakest (in terms of horsepower) and cleanest (in terms of tailpipe-CO 2 emissions) for most years.For most years, the fleets in Norway, Sweden and Germany tend to be the priciest (excluding taxes).The U.S. fleet became more expensive over time (excluding taxes) and overtook its European counterparts in 2015 and 2016.While all the fleets are becoming greener over time (in terms of PEV market share), the Norway fleet is the greenest followed by the Netherlands and Sweden.While the market shares for both BEVs and PHEVs are rising over time, there is no clear trend of one of these powertrain types dominating the other.The respective plots summarizing the statistics for the PEV fleets in the different countries are given in the online appendix.Based on figures A1-A2, in recent years, the Chinese PEV fleet is the lightest, cheapest, cleanest and smallest amongst all.On the other hand, the PEV fleets in the U.S. and the Netherlands are relatively heavier, costlier, dirtier and bigger.Looking at the ratios of the PEV fleet to the overall fleet (figures A3-A4) suggests that the PEV fleet is costlier and cleaner than the overall fleet in all countries across all years.For European countries, we obtain the average subsidy amounts per PEV fromMünzel et al. (2019).For the U.S., both the federal and state level subsidy amounts per PEV were obtained from the U.S. department of energy websites (U.S. DOE-EPA , AFDC).For China, the federal subsidy amounts were obtained fromQian (2018) andICCT (2017) and the 2015 province level subsidy amounts including registration incentive were obtained from ICCT (2018) and were assumed to be the same from 2013 to 2017.A Chinese vehicle tax exemption of 10% for PEVs was also included(Argus 2020).The average subsidy numbers for all the countries from 2010-2017 are plotted in figure 3d.Denmark, Norway and China lead in the total amount of subsidies offered for PEVs.While the generous waivers of point of sales tax dominates in Norway and Denmark, for China it is the combination of federal, provincial and registration incentives in vehicle ownership-restricted cities.

Figure 1 :
Figure 1: Evolution of sales-weighted average new vehicle fleet summary statistics for: (a) length, (b) width, (c) height and (d) wheelbase.

Figure 3 :
Figure 3: Evolution of market share for: (a) PEV, (b) BEV and (c) PHEV and (d) average subsidy.*The y-axis, which depicts market shares in figures (a, b, and c), is on a log scale.

Figure 4 :
Figure 4: Ratio of average new vehicle fleet characteristics for PEV fleet (y-axis) and replaced fleet (x-axis); both normalized by non-PEV fleet for: (a, c) tailpipe-CO 2 , (b, d) manufacturer suggested retail price (MSRP).The y-axis in 4(c) is a log scale.The size of the bubble for each country is proportional to the market share of electric vehicles in new vehicle sales in that country corresponding to each year over the 2010-2017 time period.The largest bubble size represents a market share of ~38%.

Figure 5 :
Figure 5: Variation in (a) cumulative tailpipe-CO2 avoided (y-axis) with time, (b) percentage of tailpipe-CO 2 avoided (y-axis), and (b) percentage variation from the special case curve (y-axis); with respect to the electric vehicle market share (x-axis) for the new fleet corresponding to each year over the 2010-2017 time period.

Figure 6 :
Figure 6: Variation in subsidy cost per tonne of tailpipe-CO 2 avoided (y-axis) with subsidy percentage on electric vehicles (x-axis).The size of the bubble for each country is proportional to the market share of electric vehicles in new vehicle sales in that country corresponding to each year over the 2010-2017 time period.The largest bubble size represents a market share of ~38%.