Vehicle fuel economy : what is the impact of fuel prices on the level of hybrid and electric vehicle adoption? : empirical evidence from Europe

(1)

Vehicle Fuel Economy:

What is the impact of fuel prices on the level of hybrid and

electric vehicle adoption?

Empirical Evidence from Europe

Master’s Thesis

July 2014

Karim El Hennawi (10622659) Msc in Economics

Specialization: Public Economic Policy Faculty of Economics and Business

(2)

1

Abstract

This study investigates the effect of increasing fuel prices on the level of adoption of hybrid and electric vehicles. For the main methodology, I implement a cross-sectional analysis of hybrid and electric vehicles market share over 8 years across 9 European countries to test that relationship. I also attempt to test the effects of the level of environmentalism and the political composition in each country on the level of adoption. The results of the analyses using different specifications imply that a 1% increase in fuel prices leads to around 3% to 4% increase in the level of adoption whereas no significant effects appear for the political composition and the level of environmentalism.

Acknowledgements:

I would like to thank my supervisor Dr. Carmine Guerriero for all his helpful comments and the assistance he provided me throughout my thesis.

(3)

2

The percentage of energy consumed by transportation out of total energy has been increasing due to the increase in total vehicle kilometers traveled and the insufficient improvements in vehicle fuel economy. This increase in energy consumption in the transportation sector can be compared to relatively faster and wider improvements in different sectors. Due to the fact that high-emission cars along with other types of vehicles are causing various economic and environmental problems, there had to be a suitable replacement that could solve the economic issue and reduce the negative effects on the environment. Hybrid Electric Vehicles are considered one of the most feasible solutions where each vehicle has a completely different performance design opposed to the conventional gasoline and diesel engines one. A hybrid-electric vehicle is a car that utilizes two or more main sources of power. Most hybrids operate through a combination of a fuel powered engine and an electric motor. Different forms of hybrids have become available in the market, and the methods in which they operate can vary. Some of these vehicles are plug-in vehicles where the storage battery used to power the electric motor is charged by connecting a plug to an external electric power source. Other types of hybrid vehicles make use of regenerative braking technology, which utilizes the energy released by the motor when the car brakes in order to recharge the battery. Another type of hybrid vehicles could employ fuel cell technology, which uses hydrogen as a fuel, and the fuel cell operates side by side with the battery in supplying power to the electric motor.

Since their introduction in the European market in early 2000’s, hybrid vehicles have been in increasing demand in several European countries. Sales have risen from less than 2,000 cars in 2001 to around 160,000 in 2013 across Europe (European vehicle market statistics, 2013). Expanding the adoption of hybrid vehicles has been in discussion for a few years mainly since their introduction, specifically with matters such as the efficient use of energy and the depletion of natural resources rising as hot topics. These vehicles have the potential of success in overshadowing commonly used fuel powered vehicles not too far in the future. Hybrid vehicles present the capability to address several environmental and resource externalities that are generally ignored by the existing market, and they could help automobile makers meet new

(5)

4

fleet economy requirements that are being mandated. They are considered one of the most feasible solutions to compete with the conventional gasoline and diesel vehicles and address the environmental concerns.

There has also been an evident increase in fuel prices over the past few years. Diesel prices have risen from an average of €1.23 per liter across Europe in 2005 to an average of €1.49 per liter in 2012. As for gasoline prices, they have risen from an average of €1.4 per liter in 2005 to €1.62 per liter in 2012.1 This increase raises speculation on the relationship between the price of these fuels and new hybrid and electric vehicle registrations in Europe. For several years, European countries have tried to raise fuel taxes in order to persuade consumers to drive less or switch towards vehicles with high fuel economy, of which hybrid vehicles represent a major part. High fuel taxes in Europe compared to those in the US have been believed to contribute to higher fuel economy for new vehicles and to the reduction in CO2 emissions and oil imports (Klier and Linn, 2013). Recent studies in the US have found large effects of fuel prices on fuel economy (e.g. Allcott & Wozny 2010; Busse et al. 2012), whereas studies in Europe have reached widely varying results (Clerides and Zachariadis 2008; Ryan et al. 2009). Furthermore, many studies examining the alternatives of tightening fuel economy regulation and increasing fuel tax have found the latter to be more cost effective. However, the tightening of fuel economy regulations and emission standards has been politically more favorable than fuel tax increase (Beresteanu & Li, 2011).

In this study, the goal is to perform studies similar to those mentioned above in order to observe what effect does the increase in fuel prices have on fleet economy. However, the main focus will be on the effect of gasoline and diesel prices on the percentage of new vehicle registrations (market shares) of hybrid and electric cars in Europe, specifically looking at the levels of adoption in the largest European car markets. Therefore, the aim is to answer the following question, “What is the impact of the increasing fuel prices on the market shares of hybrid and electric vehicles in Europe?” An answer to this question would have various policy implications and would affect regulations to be implemented. A strong relationship between

1

(6)

5

fuel prices and HV adoption would justify the implementation of higher fuel taxes, whereas smaller responses to fuel taxes would imply that more stringent fuel economy standards is the way to go. The method I intend to use is a panel data analysis with time fixed-effects by looking at vehicle registration statistics over time from 9 European countries in order to test the relationship between hybrid adoption and fuel prices in addition to political and environmental variables. As a main source of data, I will use country hybrid and electric car registration data from 2005 to 2012 provided by the “The International Council of Clean Transportation.”

The rest of my paper will proceed as follows. In section 2, I will discuss previous literature done on the adoption of hybrid vehicles, some of the factors affecting the adoption, and studies done on fuel economy. In section 3, I will present a preliminary analysis and my hypotheses. In section 4, I will discuss my methodology and the nature and sources of the data I will use. In section 5, I will present the results of the different specifications used and discuss the results of each of these specifications. In the sixth and final section, I will conclude and mention possible improvements and expansions that can be done in future studies.

(7)

6

2. Literature Review

Previous research on hybrid and electric vehicles has focused on the environmental advantages of these vehicles and the factors that have an effect on their adoption. There are clearly various factors that play a role in the adoption of hybrid vehicles. The impact of government incentives on the level of adoption has been studied, and to a lesser extent the effect of the changing fuel prices. Several studies have discussed the factors influencing fuel economy in general and whether tighter fuel economy standards and increased fuel taxes play any role in the improvement of fuel economy. Other literature was aimed at investigating the reasons that motivate consumers to choose a hybrid vehicle over a typical fuel powered vehicle; these reasons varied between the arising financial benefits, social norms and the willingness to abide by these norms (Ozaki and Sevastyanova, 2011). The different studies have in general found a stronger impact of fuel price compared to the various policy measures such as government incentives and fuel emission standards.

2.1. Fuel prices

In this section I will present what the literature has on the means that play a role in promoting fuel economy in general and hybrid vehicle adoption in specific. The main factors discussed are the increasing fuel prices, implementation of fuel emission standards, and the government incentives. There is ongoing debate on which is the more efficient option that would lead to lower fuel consumption.

In general, it is agreed on that the implementation of standards ensures that the fuel economy of new vehicles will not be worse than previous models despite the fact that there is a general preference for vehicles with better safety and higher energy consumption, which tend to decrease fuel economy. However, it is unclear that the implementation of standards, on its own, will continuously keep improving fuel economy. Moving on to the effect of fuel prices, there are discussions on a number of reasons why the increase in fuel taxes may not always produce the intended results of enhancing fuel economy. Many studies argue that consumers usually tend to under-estimate the potential cost savings arising from the use of fuel efficient

(8)

7

vehicles and make errors evaluating the actual costs and savings over time. This misevaluation of cost savings may diminish the impact of a rise in fuel taxes (Turrentine and Kurani 2006). According to Glazer and Lave (1994), consumers as well as manufacturers opt to take their time before making a purchase decision or proceeding in costly research, leading to a delayed effect of the increase in fuel taxes. Thus, unlike regulations whose effects are immediate, the effect of fuel prices although possibly large may be observed with a delay.

In their study, Clerides and Zachariadis (2008) aim to analyze the change in the fuel economy of several countries. They attempt to isolate the effect that fuel prices and standards have on the evolution of fuel economy. The addition that Clerides and Zachariadis present to the literature is that their analysis is based on data from many countries, which have established fuel economy standards, all over the world. The data used in their study is that of new vehicles as it is the relevant data that allows the observation of the policy effects. They explore an unbalanced panel set of data from Australia, Canada, the EU, Japan, and the US during the years 1975 to 2003. Their analysis is based on reduced form dynamic models, and they analyze each time series separately and perform a difference-in-difference methodology making use of pooled country data. They find that the implementation of regulatory methods leads to an improvement in fuel economy and that the employment of standards decreases the sensitivity of fuel economy to higher fuel prices. The analysis indicates that fuel economy targets have been the main reason behind the improvements noticed in Europe and Japan. Also, increased fuel prices have an impact on improving fuel economy. However, even in the long term price elasticity is small, so fuel prices alone are not enough to cause a considerable reduction in fuel consumption. Finally, it is expected that no improvement in fuel economy is possible without an increase in fuel prices or the tightening of fuel economy standards; if not restricted, consumers tend to shift any advantage arising from fuel savings towards a more comfortable and powerful vehicle. This study analyzes the impacts of change in emission standards in addition to the change in fuel prices, but in my study, as will be discussed further, emission standards witnessed no or very minor change preventing the study of their effect.

(9)

8

In another study, Diamond D. (2009) tries to measure the effect that government incentives have on the promotion of hybrid vehicle adoption, and what policy implications that effect may carry. In his study, he uses a cross-sectional time series analysis across US states in an attempt to measure the relationship between the adoption of hybrid vehicles and a set of socioeconomic and policy variables. This set of variables mainly includes government incentives, fuel prices, and environmentalism measures. Diamond emphasizes the difficulty in separating the impacts of the various factors affecting adoption such as the mentioned variables, or other distinct events that could possibly cause a change in hybrid vehicle sales. Similar to other studies, it is also argued that the share of hybrid vehicles may not necessarily be caused by any of the factors mentioned but simply as a result of a natural increase over time caused by diffusion of new technology. The motivation behind using cross-sectional data is to be able to isolate and be able to observe separately the different factors among each state. The variation in adoption across the states plays a role in isolating the elements that are of influence on vehicle adoption. Diamond’s main variable is the new registrations of hybrid vehicles as a percentage of all new vehicle registrations during a certain time period, which is monthly in this case. His model includes variables such as the value of government incentives, gas prices, and annual vehicle miles traveled. In addition, the model also includes factors such as HOV (high-occupancy vehicle) lane privileges, a measure of the consumers’ environmentalism, and income. Diamond’s model is summarized by the following equation,

A log-log specification is used as it comes handy in two cases. First, it seems to provide better fit to the data than the standard OLS, and secondly the coefficients of each variable can be read as the elasticity of the market share of hybrid vehicle with respect to the independent variable. The results of the analysis seem to prove the presence of a strong positive relationship between gasoline prices and hybrid & electric vehicles market shares. However, the relationship between the adoption of these fuel efficient and environment friendly vehicles with incentive policies appears to be much weaker, and it is also noticed that the means by which the

(10)

9

incentives are implemented also plays a role in the magnitude of their effect. For example, it is noticed that the effect of offering tax rebates or tax exemptions upfront is larger than the cases where these tax credits are delayed. Thus, the most influential method to utilize available government funds is to provide them to consumers up-front.

2.2. Consumer Purchase motivations

Several motives that play a role in hybrid and electric car adoption must be taken into account; these motives can be summed up under a couple of general headings. The first of these headings is the normative and pro-environmental behavior of consumers. The other set of factors can be put under the notion of innovation diffusion and consumption i.e. motives arising from social impact rather than through an individual’s rational choice. Empirical studies with regard to this topic have rendered an extensive list of motivational constructs that lie within the two general labels just mentioned.

First, a motive that can’t be left out is attributed to the various financial benefits of improving fuel economy such as the possibility of saving money on fuel. Moreover, the shift towards hybrid vehicles can be associated with the ability to benefit from certain privileges such as exemption from congestion charges. A second possible motive behind higher levels of hybrid adoption is the idea of environmentalism. Some consumers tend to choose hybrid and electric cars over the commonly used fuel powered vehicles in an attempt to preserve the environment and reduce their carbon footprint. Most of these consumers do it to set an image for themselves that they are “green.” These consumers try to show that the reason behind their choice of vehicle is purely based on their level of environmental awareness and their affinity towards the environment. A different set of consumers are those that are affected by their community and tend to abide by social norms, so their use of hybrid cars gives them a sense of sharing community values. For others, the use of hybrid cars arises from an attraction to new technologies and innovation or simply as an attempt to reduce petroleum use and be free of restrictions by oil producers. The set of motives just mentioned all hold the same value in trying to assess consumer purchase behavior. Ideally, an appropriate measure of all these motives

(11)

10

would be required in order to estimate the effect of consumer purchase motivations on the level of hybrid adoption.

In the study by Ozaki and Sevastyanova (2011), a questionnaire survey and exploratory factor analyses are exploited in order to measure the importance of each of these sets of motives. They find that consumer behavior is highly affected by the arising financial benefits from the purchase of a hybrid car. Moreover, the effect of social norms and pressure also seems to play a big role in influencing consumer behavior when it comes to adoption of a new technology such as hybrid cars. However, the effect of a motive such as environmental benefits tends to be mixed for the reason that environmentalism is a broad topic and its outcome is not immediately observable at a personal level. Finally, the mere fact that hybrid vehicles are a new technology seems to have a large impact on consumers, and people’s interest in technology seems to drive them toward choosing a fuel efficient car. As mentioned earlier, it is essential to account for these motives when attempting to estimate the size and magnitude of any factor influencing the adoption of hybrids though some of them may be hard to quantify. This study uses a reliable methodology that accounts for different dimensions of consumer adoption motivation. It also has implications on transport policy and persuades the notion that financial incentives and emphasizing these benefits would have positive consequences on adoption.

2.3. Political composition

There is also reason to believe that the political composition of a country plays a role in altering the level of adoption of hybrid and electric vehicles. Several studies have been done within one country or among limited number of countries. Dunlap et al. (2001) look at the US comparing the behavior of democrats and republicans in terms of environmental support. That study could have different dynamics than a study looking at environmental behavior of political parties in Europe. However, its results can still be used to argue that support for environmental laws differs from one political party to the other. The results of this paper indicate there is a gap between Republican and Democratic support for pro-environmental rules and this gap has also been increasing over the years. Furthermore, Neumayer (2004) makes use of large cross-national sample from 25 countries over a time span from 1945 to 1998 making use of party

(12)

11

platforms for elections. Results in Neumayer’s study show that a left-wing political party is more likely to promote pro-environmental beliefs and support environmental laws.

(13)

12

3. Preliminary Analysis and Hypotheses

Before beginning the explicit description of the methodology I will be using in my analysis, I will first restate what I aim to test in my study. My goal is to attempt to estimate the sign and magnitude of the effect that changes in fuel prices have on the adoption of hybrid and electric vehicles, which are considered the leading technology in terms of vehicle fuel efficiency. As mentioned earlier, the question I intend to answer is what the impact is, of the recently continuous increase in fuel prices in 9 of the largest European automobile markets on the market shares of hybrid and electric vehicles. According to previous literature on factors affecting fuel economy and the factors affecting consumer behavior, the expected result would be that the higher fuel prices would lead to faster adoption and thus a larger market share.

There are several difficulties to be discussed when analyzing the adoption of hybrid and electric vehicles. The isolation of the effects of specific independent variables is difficult as they have, along with the market share for hybrid and electric cars, exhibited an increasing trend over time. In specific, looking at the annual averages of diesel and gasoline prices and the annual average of market shares, we notice according to figure 1 above that they seem to have undergone steady increases throughout the concerned time period form 2005 till 2012.

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2005 2006 2007 2008 2009 2010 2011 2012 MS G_price D_price

Figure 1 - Yearly hybrid and electric vehicle market shares and average annual gasoline and diesel prices

(14)

13

Furthermore, there is the theory that the increase in the market share of hybrids could be a result of natural increase over time following a trend of the diffusion of new technologies; it is believed that the consumers’ purchase or adoption of a certain product is based on or at the least associated with the number of previous buyers of that product as seen in the study by Bass (1969).

One of the first limitations in my study and perhaps the most important one is the inability to find clear data on the government incentives implemented in the concerned European countries. Country specific lists of the currently implemented incentives are available for some of the countries, but I was not able to find any clear timeline of when and how long these incentives were in effect. Also, for some of the countries there has not been any noticeable change in the level of incentives within the concerned time period. In terms of the unavailability of clear annual data, this is specifically a major setback as it is essential to have year by year data for the separate entities. However this problem might have less of an effect on the results than anticipated as the level of incentives within the concerned time period from 2005 till 2012 has not undergone major changes in the 9 countries concerned. Thus, working with panel data and running a ‘fixed effects’ regression, the problem of homogeneity of incentives across the time periods is taken care of.

To address the problem of the missing incentives data, the inclusion of a political variable could be an appropriate proxy for the government incentives level or the level by which hybrid and electric vehicles are promoted within that country. As discussed by Neumayer (2004), the position which political parties take within the left-right political image tends to play a role in their attitude towards environmental topics and beliefs. The results of the study, analogous with previous studies, indicate that a political party with a left-wing orientation inclines towards the support of pro-environmental issues and they seem to embrace and promote pro-environmental beliefs. Therefore, a variable that measures the percentage of the left wing members in the parliament in each country could be a proxy for the level of support for environmental friendly and energy efficient technologies such as hybrid and electric

(15)

14

vehicles. It is then expected that the higher the percentage of left wing members in the parliament, the higher the level of hybrid and electric vehicle adoption.

(16)

15

4. Data & Methodology

In this section, I will present in detail the nature of the data and the data sources used, in addition to how I will exploit that data. Furthermore, I will go on to discuss the main

specification that I will use in my methodology.

4.1. Data

In terms of the data to be used, I will utilize statistics from the 9 largest vehicle markets in Europe: Austria, Belgium, France, Germany, Italy, Netherlands, Spain, Sweden and United Kingdom. The choice of data was limited to these 9 countries mainly due to the unavailability of data for other European countries for the chief variable of concern which is the market share of hybrid and electric vehicles. Also, these countries would be indicative of Europe as a whole since if combined they represent more than 85% of the total number of car registrations every year across the EU-27 countries (European Automobile Manufacturers Association, 2013).

The data I will use mainly relies on yearly registrations of hybrid vehicles as a percentage of total passenger car registrations, and the yearly averages of diesel and gasoline prices in the chosen European countries. The data for market shares is obtained from European vehicle market statistics provided by “The International Council of Clean Transportation.” The numbers used for new vehicle registrations are based merely on passenger car registrations as they constitute 90% of the market with the remaining 10% being light-commercial vehicles. Data on fuel prices is obtained through the European Commission’s Oil Bulletin under “Market observatory and Statistics” where a history of weekly prices is posted. From this data, gasoline and diesel prices used for transportation in the EU are picked out from the listed weekly fuel prices and are used to calculate annual averages.

An arising shortcoming could be the unavailability of monthly data for market shares since the changes in fuel prices can fluctuate from month to month and that could lead to a bias in the results. However, despite the fact that fuel prices undergo short term fluctuations, monthly data for market share could be an unreliable measure as a result of supply concerns

(17)

16

during times of high demand (Diamond, D., 2009). Thus, the use of yearly data is expected to produce the desired results, and the absence of monthly data should not pose difficulties in the analysis. Furthermore, I would have run into the same obstacle with the other variables I am using seeing that monthly data for those is also not readily available, and for some there is not enough monthly variation in the data which would have been also problematic with the use of panel data. Moreover, an observed limitation in comparison to previous studies, specifically those using US data, is the unavailability of data for specific vehicle models, but rather the use of country aggregate statistics for registrations of all hybrid and electric vehicles in a specific country.

The data for the political variable was collected from the parliamentary results in each of the countries concerned in the study. As this variable relies on the percentage of left wing members in the parliament, it is necessary to obtain numbers for the winning candidates in each round of the elections within the time period ranging from 2005 to 2012. This data was obtained from the archives of the official website of the ministry of interior or any equivalent ministry in these 9 European countries.

As for the data for environmentalism, I made use of survey results done by the European Social Survey (ESS) organization. These surveys are available on their website with the results of various rounds of the survey running every 2 years from 2002 until 2012. The results of the question “How important is it for you to care for nature and the environment” will be utilized. The answers to this survey range from those who care most about the environment answering “Very much like me” and those who don’t think the environment is important to them at all choosing the answer “Not like me.” One limitation could be the unavailability of these survey results for Austria and Italy for rounds further than 2006.

4.2. Methodology

The use of a cross-sectional fixed effects model to estimate the impact of fuel prices and other factors affecting hybrid adoption allows controlling for policy and economic factors that influence adoption but are generally constant over time. In addition to the problem faced in

(18)

17

terms of accounting for government incentives, which was discussed earlier, there is also the matter of vehicle emission standards. These emission standards have changed once during the period in which I am running my analysis. Euro-5 emission standards were introduced in 2009 to replace the Euro-4 standards. In terms of the difference in the emission levels allowed according to these standards, I will assume that the change is minor and plays no or limited role in altering the behavior towards the adoption of these fuel efficient vehicles. The differences between the mentioned standards are displayed in figure 2.

Figure 2 - EU emission limits for gasoline and diesel passenger cars

Further justification behind the use of a fixed effects panel data model are reasons related to the variables at hand. The time-invariant variables that are to be used in the analysis are distinctive for each country which is an important assumption of the fixed-effects model. The level of adoption, the prices of gasoline and diesel, the percentage of left wing parliamentarians, and the level of environmentalism all differ between the 9 countries. Before providing the basic model for the analysis, a description, of each of the independent variables, is necessary. The level of adoption is measured as the percentage of hybrid and electric vehicle new registrations within a certain year out of the total number of new vehicle registrations. The prices of diesel and gasoline are utilized as annual averages of the weekly prices calculated in euros per liter. Whereas the political variable is represented by the percentage of members of

(19)

18

the parliament that are part of a left wing party, or that are part of a party with left wing orientation. As for the environmentalism variable, as stated earlier it utilizes survey results indicating how much people care about nature and the environment. The answers ranged from 1 to 6, so the variable I will use constitutes of an average of the provided answers for each country in each round of the survey. The surveys were run every other year, so for the missing years I used the results of the previous round, which could be considered as one of the limitations in my study. Also, the data for Austria and Italy is only available for the first two rounds, thus from 2006 onwards the environmentalism measure for these two countries remains constant.

The basic model specification I will be using is the following,

(1)

where subscripts indicate an observation in country i and at time t. is an unknown intercept that is to be estimated for each country and can be defined as country fixed effects. The difference in the country fixed effects arises from omitted variables such as government incentives which tend to vary across countries but not over time. The independent variables are those mentioned in the section above. First element is the annual average price of

gasoline in euros/liter for each country. is the annual average price of diesel in euros/liter

for each country. is the political variable measuring the percentage of left wing members

in the parliament. Whereas is the measure of a country’s environmentalism. As

mentioned earlier, the data for the environmentalism variable is based on survey results done by the “European Social Survey.” Seeing that the range of answers to this survey is from 1 to 6, my variable is an average of the results, with the weight for each answer being the percentage of participants that provided that answer2. The vector , as its name indicates, is a

vector of controls that accounts for factors that reflect a country’s economic situation in general and its fleet economy in specific. This vector includes a variable measuring a country’s

2

(20)

19

population, a variable representing a country’s GDP measured in PPP3, and a variable measuring the total number of kilometers per year traveled by passenger cars in each country. These controls are exogenous as they are not affected by other variables in the model. There is no causal link from within the model leading to the population, GDP, and vehicle kilometers traveled in a country within a certain year. The dependent variable in the analysis “ ” is the

market share of hybrid and electric vehicles in each country. This variable measures the percentage of hybrid and electric vehicle registrations out of the total number of new car registrations during a single year.

3

PPP: Purchasing Power Parity, which takes into account the relative costs and the inflation rates of the distinct countries, rather than simply using exchange rates which may distort the real differences in income.

(21)

20

5. Results and Discussion

The estimates obtained from the different models and specifications used are presented in this section. Along with the reported results is an analysis of the OLS estimates, log-log estimates, and instrumental variables estimates.

5.1. OLS Estimates

The primary regression performed is the basic panel data OLS regression which was run as described in the previous section. The regression analysis produced the results displayed in the table 1 below. In this analysis all data points were used for all countries and the prices for gasoline and diesel were included as separate independent variables. The exact same regression was run again, but with the use of gasoline and diesel prices that were adjusted for inflation. The second column of table 1 shows the results of the regression that was run with these inflation adjusted prices.

Table 1 – OLS regression results

(1) (2)

Dependent Variable

HEV market share HEV market share

Gasoline Price 2.11 (1.88) -4.14 (3.64) Diesel Price 0.128 (1.14) 4.31 (2.63)+ Political Composition 0.0888 (0.00855) -0.00600 (0.0192) Environmentalism -1.52 (1.37) -0.291 (0.382) Vehicle km traveled 0.00675 (0.00595) 0.00554 (0.00983) GDP in PPP -0.447 (0.0194)** -0.0162 (0.0206) Population 4.63e-08 (1.10e-07) 2.70e-07 (1.13e-07)** Constant 5.31 (11.7) -8.36 (7.35)

Estimation Procedure Fixed effects

R2 (within) 0.299 0.230

Number of observations 72 72

Notes: Robust standard errors are reported in parentheses ***denotes significant at the 1% level; **, 5% level; *, 10%; +, 15%.

(22)

21

Before performing the regression analysis, in order to check for heteroskedasticity, a “Modified Wald test for groupwise heteroskedasticity in a fixed effect regression model” was run4. The null hypothesis for this test is homoscedasticity; here the null is rejected and I conclude the presence of heteroskedasticity among the variables. Therefore, to address this problem clustered standard errors were used. In relation to panel data, each cluster consists of an entity, so clustered standard errors allow for heteroskedasticity and for auto-correlation within an entity, but treat the errors as uncorrelated across entities (Stock & Watson, 2011). Moving on to the results, first I will discuss the difference between the two regressions. None of the variables of interest produces a significant estimate in both regressions which tends to be surprising. Thus no real conclusions can be made from these obtained results. Furthermore, the R2 drops from 0.299 to 0.230 after adjusting prices for inflation, and that could be a result of the fact that the prices actually observed by consumers could be more indicative of the market rather than the use of real prices. An additional explanation for the obtained results could be a result of using fuel prices as separate variables. That could be a problem specifically in Europe, seeing that many European countries use both fuels, diesel and gasoline, and their use differs in the percentage from total fuel consumption from one country to the other. Some European countries have a higher percentage of gasoline operated passenger cars in the market while other countries rely mainly on diesel as their main source of fuel. This is an issue that does not come up in studies performed in other major markets such as the US, China or Japan as their main fuel source is gasoline and very few diesel operated vehicles or even none at all are available in these markets (European vehicle market statistics, 2013). One way to deal with this problem is to make use of the percentages of gasoline and diesel operated cars in the market within a given country. Obtaining those percentages allows calculating a new independent variable, that variable being a weighted average of the annual averages of gasoline and diesel prices. The results of the regression after the inclusion of the new variable, the weighted average of fuel prices, are shown in table 2 below.

4

‘Modified Wald test for groupwise heteroskedasticity in a fixed effect regression model’ is run through user written program in Stata called xttest3

(23)

22

The estimates obtained indicate the presence of a positive and significant effect of fuel price on the market share of hybrid and electric vehicles, the coefficient is statistically significant at the 1% level. With just a minor decrease in R2 from 0.299 to 0.258 after the use of the weighted average and the large significance of the coefficient obtained, this justifies the use of a weighted average over the use of separate variables for gasoline and diesel prices. Moving on to the political composition variable, we notice that the coefficient is very small and not significant. As for the coefficient for environmentalism it is surprisingly negative, but also not significant. Therefore, no conclusions can be derived on the actual effects of the two latter variables.

Figure 3 – Annual average of hybrid and electric vehicle market shares and weighted average of fuel prices

The relationship between the market share of hybrid and electric car registration and the weighted average of fuel prices is shown in figure 1 above. It is clear that both measures exhibit an increasing trend over time and tend to follow the same pattern i.e. an increase in fuel price is accompanied by an increase in the level of adoption of these vehicles. It is however noticeable that there is an evident sudden drop in fuel prices in 2009, and that could be attributed to the great recession that started in the last quarter of 2008. The increasing trend in the average fuel price continued from 2009 onwards. Thus, the need arises to control for this recession by the inclusion of an additional variable. The variable to be used to account for this

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2005 2006 2007 2008 2009 2010 2011 2012 MS Weighted price

(24)

23

recession is a simple dummy variable (Burkhauser et al., 2000). This dummy is set to 1 for the year 2009 in which prices were mainly affected by the recession and it is 0 for all other time periods. The regression is run with the inclusion of this dummy, and the results are observed in 2nd column of table 2. The inclusion of this dummy variable leads to a decrease in the level significance of coefficient of the weighted fuel price from the 1% level to the 10% level with the increase in the p-value from 0.009 to 0.054.Although the coefficient of the recession dummy variable on the market share is not statistically significant, we notice that the R2 of the regression increases from 0.258 to 0.323 indicating that the inclusion of this dummy variable has provided a better fit of the data.

Table 2 - Regression results using weighted fuel price

(1) (2)

Dependent Variable

Weighted fuel price 1.90

(0.554)*** 2.69 (1.19)* Political Composition 0.00587 (0.0105) 0.00774 (0.0141) Environmentalism -1.28 (1.15) -1.52 (1.41) Vehicle km traveled 0.00612 (0.00632) 0.00171 (0.00934) GDP in PPP -0.374 (0.0162)** -0.0434 (0.0222)* Population 9.53e-08 (7.90e-08) -2.96e-08 (1.63e-07) Recession --- 0.586 (0.507) Constant 5.31 (11.7) 9.82 (16.2)

R2 (within) 0.258 0.323

Notes: Robust standard errors are reported in parentheses

(25)

24

5.2. Log-Log Estimates

As mentioned earlier in the study done by Diamond (2009), the use of a log-log specification could lead to a better fit for the data, and it also allows the interpretation of the coefficient as the percentage change in the independent variable that would lead a one percent change in the dependent variable. The use of a log-log specification also deals with the heteroskedasticty problem. Thus, I opted to run a regression using log values for all the variables available with the exception of the recession dummy variable. Clustered standard errors are still used as they are valid whether or not there is heteroskedasticity (Stock & Watson, 2011). The results of the regression are displayed in table 3 below. The first column is the regression run without the recession dummy variable, and column 2 includes that dummy.

Table 3 - Regression results using log-log specification

(1) (2)

Dependent Variable

Log(Weighted fuel price) 2.52

(0.923)** 4.06 (0.715)*** Log(Political Composition) 0.116 (0.793) 0.240 (0.886) Log(Environmentalism) -4.55 (4.85) -5.70 (5.38) Log(Vehicle km traveled) 5.37 (8.52) 3.28 (8.13) Log(GDP in PPP) -2.27 (5.32) -2.71 (4.72) Log(Population) 22.2 (10.2)* 14.2 (7.13)* Recession ---- 0.283 (0.113)** constant -172 (65.7)** -106 (49.6)**

R2 (within) 0.629 0.677

***denotes significant at the 1% level; **, 5% level; *, 10%; +, 15%.

As clear from the results, the use of a log-log specification leads to a substantial increase in the R2 from 0.26 to 0.63. Clearly this model provides a better fit for the data, but the

(26)

25

obtained coefficients do not vary much from the previously obtained results. Before accounting for the recession, we notice that the a 1% increase in the average fuel price leads to a 2.52% increase in annual hybrid and electric vehicle registrations, and this coefficient is statistically significant at the 5% level. The other variables of concern, i.e. the political composition variable and the environmentalism variable, are both clearly statistically insignificant, and the environmentalism variable shows that a 1% increase in environmentalism leads to a 4.5% decrease in hybrid adoption which is inconsistent with expectations and with previous literature on fuel economy. The inclusion of the recession dummy variable as observed earlier also results in a minor increase in the R2 from 0.63 to 0.68. The effect of the weighted average of fuel prices on the adoption level seems to become larger after the addition of this dummy variable as a 1% increase in in the fuel causes a 4% increase in market shares and that value is statistically significant at the 1% level with a very small p-value .There is clearly no change in the coefficients of other variables of interest; however, the recession dummy variable seems to have a significant positive effect on the level of adoption. The effect of the recession on the level of adoption could be explained in two ways. The first, which is a direct effect, is a positive one and could have resulted from the decrease of the prices of the vehicles themselves thus increasing the number of new hybrid and electric vehicle registrations. The second effect of the recession is an indirect one and not explicitly observable, and this effect is expected to be a negative one resulting from the decrease in fuel prices in the market, making the purchase of a fuel efficient vehicle less attractive to consumers.

5.3. Instrumental Variable Estimates

As mentioned earlier, both the political composition and the environmentalism variables produce unexpected results. First, the political composition although positive in all the regressions performed, it does not appear to be statistically significant in any of them with p-values exceeding 0.5. Furthermore, the coefficient for the environmentalism variable appeared to be negative for the regressions done, yet it also was not significant in any of the specifications used. A possible explanation for those results could be the fact that the environmentalism variable varied only every other year, and for both Italy and Austria the same

(27)

26

value was used across all the time periods due to unavailability of the survey results in following rounds. In an attempt to tackle this problem, I ran the regression excluding this environmentalism measure. However, there was no effect on the results of the panel regression as the coefficient for the average fuel prices remained positive and statistically significant at the 1% whereas a minor increase in the political composition coefficient is observed though still not statistically significant.

Another possible explanation for the obtained results is that the effect of environmentalism is not a direct effect on the level of adoption, but rather has an effect on the political composition of a certain country. The level of environmentalism in a country is expected to play a role in the people’s decisions, and thus on their political stands. Dunlap et al. (2010) mention that if voters take a stand that is pro-environmental when selecting their preferred candidates based on those candidates’ levels of support for the environment, then that will have an effect on results of the elections and the policies that will emerge as a consequence of these results. List and Sturm (2004) find that the composition of the population plays a role in determining the political composition of that country and thus the environmentally friendly policies to be implemented and also that politicians tend to impact the decisions of voters through secondary policies mainly environmental policies which seem to have a larger impact on electoral decisions and politicians find easier to control. Based on that relationship between the level of environmentalism and the political composition, in addition to the fact people’s electoral decisions are highly affected by politicians’ environmental actions, the level of environmentalism therefore plays a role in determining the political composition. Thus, I will use the environmentalism variable as an instrument for the political variable, and observe what effect it has on the coefficients. I assume that the level of environmentalism in a certain country is exogenous. People’s views on the environment and the pro-environmental stands they take are based on their own beliefs. Despite higher levels of awareness in some societies, environmental behavior relies solely on their personal attitudes. The endogenous variable is the country’s political composition.

(28)

27

Table 4 - Two stage least square regression results

(1) (2)

Dependent Variable

Log(Weighted fuel price) _(1.48)**3.31 _(2.78)**6.65

Log(Political Composition) _(4.67)3.85 _(5.94)6.04 Log(Vehicle km traveled) _(6.71)8.46 _(7.06)5.27 Log(GDP in PPP) _(9.31)-9.06 _(11.5)-12.8 Log(Population) _(6.83)***23.8 _(9.30)10.2 Recession ---- _(0.224)**0.501 Constant _(52.1)***-186 _(73.0)-73.6

Estimation Procedure Instrumental variables and two-stage least squares for panel-data _models

R2 (within) 0.456 0.326

***denotes significant at the 1% level; **, 5% level; *, 10%; +, 15%.

The results of the two-stage-least square regression with environmentalism as an instrument for political composition of a country are displayed in table 4 above. The variables used in this regression are all log values since as explained earlier this specification deals with the heteroskedasticity problem and provides better fit for the data. As the results in the earlier regressions, the coefficient for fuel prices appears to be positive and statistically significant at the 5% level with a clearly higher coefficient when accounting for the recession. The coefficient for the recession dummy variable is also positive and significant, and as explained earlier this positive effect is likely as a result of the drop in the price of the vehicles. Moving on to the variable being instrumented which is the political composition variable, we notice an evident increase in the obtained coefficient compared to the simple cross-sectional analysis. Although these coefficients for the political composition are still not statistically significant, it is noticed that there is a decrease in the standard error relative to the coefficient and thus a decrease in the p-value. The observed p-value for this coefficient is around 0.3 which is much lower compared to values obtained in previous regressions which ranged between 0.6 and 0.8. The

(29)

28

obtained coefficients suffer from the same limitations that were mentioned earlier which could be the cause of a bias in the results and which could have prevented a suitable fit for the data. The political composition variable did not vary every time period but rather every two or three years depending on the frequency of the parliamentary elections in each country. Also, the environmentalism variable only varied every other year and even this variation was a small one. These predictors had little or much slower variation over time compared to average fuel prices which increased at a faster rate than any of the other dependent variables, explaining the larger and more significant coefficients obtained compared to the fixed effects specification.

In order to test for instrument validity I observe the results of the uderidentification test5. In the ﬁrst-stage regression results, the “Anderson Canon. Corr. LM” statistic fails to reject the null hypothesis of underidentification at the 95% level. This suggests that even for overidentiﬁcation with the order condition, the instruments may be inadequate to identify the equation. Now I look at the Weak-instrument-robust inference test which tests for the significance of the endogenous regressor in the main equation. The Anderson–Rubin Wald test and Stock–Wright LM test reject their null hypothesis at the 10% level and indicate that the endogenous regressors are relevant. However, the null hypotheses are joint tests of irrelevant regressors and appropriate overidentifying restrictions. Thus, the evidence presented is not promising in terms of the adequacy of the instrument.

5.4. Alternate Specifications

In this section, I consider robustness checks. First, I investigate the effect of the addition of time dummies to the panel regression. Second, I consider running standard OLS regression rather than a cross-sectional analysis.

I perform the fixed effects regression with additional regressors; these regressors being time period dummies in order to observe whether inclusion of time effects has any effect on the results. For this regression I also used log values for all independent variables excluding the dummy variables which always take a value of 0 or 1. This addition leads to a minor increase in the R2, implying a possibly better fit of the data. However, none of the time dummies seems to

5

(30)

29

have any statistically significant effect. Also, the coefficient for the fuel prices appears to be much smaller, but it is not statistically significant. As for the other variables of concern, all coefficients are statistically insignificant in particular the environmentalism and political composition variables. Furthermore, I perform the analysis using standard OLS with each observation as a separate data point rather than pooling the data according to country and time period. The R2 is also a bit higher compared to the panel regression with a rise from 0.677 to 0.699. The results obtained are very similar to those of the fixed effects regression. The coefficient for fuel prices is positive and statistically significant. However, environmentalism variable is surprisingly negative and statistically significant. On the other hand, the coefficient for the political composition variable is small and not statistically significant.

5.5. Discussion

From the obtained results a few inferences can be drawn on the different specifications and the choice of independent variables. The fixed effects regression accounts for the missing policy effects. Variables that are not observable and have small or no change over time such as government incentives are taken care of using the panel data fixed effects model. The use of the average fuel price of gasoline and diesel rather than separate variables for these prices clearly is a better estimate. This average leads to a better fit for the data in addition to a more accurate representation of the observed fuel prices in the different markets. Furthermore, the inclusion of the recession variable proved to be necessary. It clearly leads to a better fit, and has a significant effect on the level of hybrid adoption. Most importantly, the log-log specification appears to be the most suitable specification. This specification deals with the heteroskedasticity problem and provides a better fit for the data. As for the instrumental variable method the results may not be as reliable. That is due to the fact that the instrument used appears to be a weak one. The invalidity of the instrument could be caused by the limitations mentioned concerning the environmentalism measure6.

6

These limitations include missing data for some countries, and the fact that the survey results are available only every other year.

(31)

30

6. Conclusion

After the analysis of the different factors that may play a role in promoting hybrid and electric vehicle adoption not very strong conclusions can be made. One thing that is clear is that fuel prices have a large impact on consumer decisions and can be seen that a 1% increase in fuel prices can lead to around 4% increase in hybrid car registrations each year. However, the effect of the political composition seems to be a smaller one and generally insignificant which may indicate that the percentage of left wing parliamentarians has less of an effect on the level of hybrid adoption. Moreover, Environmentalism variable seems to present mixed and insignificant results which could be due to the limitations discussed above. In general, no major policy implications can be derived from this analysis seeing that the effect of government incentives was not also studied. Nevertheless, I believe it is safe to confirm that an increase in fuel taxes would encourage car users to shift towards vehicles that are less fuel reliant such as hybrid and electric vehicles. Also, it should be noted that even though consumers may under-estimates the cost benefits of hybrid cars, they could still be geared towards a hybrid car as a guarantee against future fluctuations in fuel prices.

Despite the fact that an inference on the impact of the level of fuel prices can be taken from this study, there are wider areas through which the elements of hybrid adoption can be studied. A possible expansion of this study would be the possibility of using data on separate hybrid models available in the European in order to observe any differences in the reaction to certain factors. Furthermore, one of the major limitations I believe was the environmentalism variable used. The same analysis done with an accurate measure of environmentalism that varies on an annual basis could provide much more reliable results. Also, the market penetration of hybrid and electric vehicles in Europe is clearly still in its prime, and percentage of hybrid registrations is still low and even non-existent in some European countries which were not included. Therefore, future research could use a wider dataset in terms of additional time periods and more countries across Europe for a better representation. Finally, the expansion of the dataset could also be used to include time periods in which interpretations can be made on the effect of a change in fuel emission standards on the level of adoption.

(32)

31

References

Allcott, H., Wozny, J. (2010). Gasoline Prices, Fuel Economy, and the Energy Paradox. Working Papers 1003, Massachusetts Institute of Technology, Center for Energy and Environmental Policy Research.

Bass, F. M. (1969). A New Product Growth for Model Consumer Durables. Management Science, Vol.15, Theory Series, 215-227.

Beresteanu, A., Li, S. (2011). Gasoline Prices, Government Support, and the Demand for Hybrid Vehicles in the United States. International Economic Review, Vol. 52, No. 1, pp. 161-182. Burkhauser, R. V., Couch, K. a., & Wittenburg, D. C. (2000). A Reassessment of the New Economics of the Minimum Wage: Literature with Monthly Data from the Current Population Survey. Journal of Labor Economics, 18(4), 653–680.

Busse,M., Knittel, C., Zettelmeyer, F. (2013). Are Consumers Myopic? Evidence from New and Used Car Purchases. The American Economic Review, 103(1), pp. 220–256.

Cansino, J. M., Pablo-Romero, M. D. P., Román, R., & Yñiguez, R. (2010). Tax incentives to promote green electricity: An overview of EU-27 countries. Energy Policy, 38(10), 6000–6008. Clerides, S., Zachariadis, Theodoros, 2008. The effect of standards and fuel prices on automobile fuel economy: an international analysis. Energy Economics Vol. 30, PP. 2657–2672.

Diamond, D. (2009). The Impact of Government Incentives for Hybrid-Electric Vehicles: Evidence from US States. Energy Policy, Vol. 37, pp. 972-983.

Dunlap, R. E., Xiao, C., & McCright, a. M. (2001). Politics and Environment in America: Partisan and Ideological Cleavages in Public Support for Environmentalism. Environmental Politics, 10(4), 23–48.

European Automobile Manufacturers Association (2013). New Passenger Car Registrations by country [excel file]. Retrieved from http://www.acea.be/statistics/tag/category/by-country-registrations

European Commission (2014). Oil Bulletin [excel file]. Retrieved from http://ec.europa.eu/energy/observatory/oil/bulletin_en.htm

European Commission (2013). Statistical Pocket Book 2013, Mobility & Transport Statistics. Retrieved from http://ec.europa.eu/transport/facts-fundings/statistics/pocketbook-2013_en.htm Glazer, A., & Lave, C. (1994). How Regulations Can Succeed Where Taxes Do Not : An Examination of Automobile Fuel Efficiency. EconWPA, Public Economics 9406002. Also at http://ideas.repec.org/p/wpa/wuwppe/9406002.html

(33)

32

Klier, T., Linn, J. (2010). The Price of Gasoline and New Vehicle Fuel Economy: Evidence from Monthly Sales Data. American Economic Journal: Economic Policy, Vol. 2, No. 3, pp. 134-153 Klier, T., & Linn, J. (2013). Fuel prices and new vehicle fuel economy—Comparing the United States and Western Europe. Journal of Environmental Economics and Management, 66(2), 280– 300.

Klier, T., Linn, J. (2013). Fuel Prices and New Vehicle Fuel Economy in Europe. Journal of Environmental Economics and Management, Vol. 66, Issue 2, pp. 280–300.

Li, S., Timmins, C., H. von Haefen, R. (2009). How Do Gasoline Prices Affect Fleet Fuel Economy? American Economic Journal: Economic Policy, 1:2, pp. 113–137

List, J. A., Sturm, D. M. (2004). How Elections Matter: Theory and Evidence from Environmental Policy .National Bureau of Economic Research. NBER working paper series, Working Paper 10609.

Mock Peter, “European Vehicle Market Statistics, 2013,” Published Mon, 2013.10.28

Neumayer, E. (2004). The environment, left-wing political orientation and ecological economics. Ecological Economics, 51(3-4), 167–175.

Baum, C. F., Schaffer, M. E., Stillman, S. (2007). Enhanced routines for instrumental variables/generalized method of moments estimation and testing. The Stata Journal, 7(4), 465-506.

Ritsuko Ozaki, R., Sevastyanova, K. (2011). Going Hybrid: An Analysis of Consumer Purchase Motivations. . Energy Policy, Vol. 39, pp. 2217-2227.

Ryan, L., Ferreira, S., Convery, F., 2009.The impact of fiscal and other measures on new passenger car sales and CO2 emissions intensity: evidence from Europe. Energy Economics 31, 365–374.

Schipper, L. (2011). Automobile use, fuel economy and CO2 emissions in industrialized countries: Encouraging trends through 2008? Transport Policy, 18(2), 358–372.

Stock J, Watson M. Introduction to Econometrics (3rd edition). Addison Wesley Longman, 2011.

Turrentine, T. S., & Kurani, K. S. (2007). Car buyers and fuel economy? Energy Policy, 35(2), 1213–1223.

Vance, C., Mehlin, M. (2009). Tax Policy and CO2 Emissions – An Econometric Analysis of the German Automobile Market. Ruhr-Universität Bochum (RUB), Department of Economics.

Vehicle fuel economy : what is the impact of fuel prices on the level of hybrid and electric vehicle adoption? : empirical evidence from Europe