The price effects of the EU ETS in the heat and power sectors

The price effects of the EU ETS in

the heat and power sectors

Carlotta Masciandaro

Supervisor: Stefan W¨

ohrm¨

uller

Bachelor thesis, Economics and Business Economics

Major: Economics

Student id: 11844957

June 29, 2020

Abstract

This thesis investigates the price effects of the European Union Emission Trading Scheme (EU ETS) in the heat and power sectors for 27 participant countries over the period from 2008 to 2018. I employ a Country and Time Fixed Effects regression analysis on realized historical data to estimate the impact of the European Allowances price on the Harmonized Index of Consumer Prices. Notably, poorer households spend a larger share of their income on energy and, thus, a policy that raises heat and power prices is by itself regressive. My findings suggest that the EU ETS had a statistically significant yet small regressive impact and that this effect varied considerably across geographical regions and over the national income per capita distribution.

Keywords: European Union Emission Trading Scheme; Cap-and-trade; Price Effects; Distributional Impacts; Panel Data Analysis; Country and Time Fixed Effects.

Statement of Originality

This document is written by Carlotta Masciandaro who declares to take full responsi-bility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Contents

1 Introduction 2 2 Policy Description 3 3 Literature Review 5 4 Data Description 7 5 Methodology 12 6 Results 15

7 Conclusion and limitations 21

8 References 22

9 Appendix 25

9.1 Countries included in the study . . . 25

9.2 Multicollinearity Diagnostics . . . 26

List of Figures

1 EU ETS Phase 2. Data from the European Climate Exchange. . . 3

2 EU ETS Phase 3. Data from the European Climate Exchange. . . 4

3 Price index average over the participant countries in the time period of interest.. 9

4 Crude oil price trend between 2008 and 2018. . . 10

5 Euribor one month trend between 2008 and 2018. . . 11

6 Output of the Hausman Test . . . 13

7 Output of the test for heteroskedasticity . . . 13

8 Output of the test for time dummies joint significance . . . 14

List of Tables

1 Summary statistics table for the variables described in this section. . . 8

2 Regressions results . . . 16

3 Regional comparisons . . . 19

4 Poorest and Richest EU countries comparison . . . 20

5 Countries in the sample and respective regions and income classification . . . 25

6 VIF values for OLS . . . 26

7 VIF values for Country FE . . . 27

1

Introduction

The high level of greenhouse gases that human activities have been emitting since the Industrial Revolution is generating catastrophic consequences (NASA, 2019). Several measures have been taken to reduce emissions both at the national and at the international level, ranging from the carbon tax applied in the British Columbia province of Canada to the Kyoto Protocol, signed by 84 countries. To comply with the latter, the European Union instituted the European Emission Trading Scheme (EU ETS). It is the world’s largest cap-and-trade system and it has proven quite effective in meeting the environmental targets set by the EU. The extent of this policy -more than 446 million people are affected by it- make it crucial to evaluate its desirability taking into account its distributional effects as well as its effectiveness. In light of this, I estimate the price effects generated by the EU ETS using actual historical data and I find them to be slightly positive.

From a theoretical point of view, carbon pricing represents a solution to the tragedy of open access or, as Hardin termed it in 1968, the ”tragedy of the commons”. The absence of property rights over the limited ability of the atmosphere to absorb CO2 allows for the overuse of this

scarce resource, which produces an externality for the entire society (Boyce, 2018). Then, the implementation of a carbon price creates these property rights, which in the EU ETS are shared between the firms that received the permits for free and the state, when those are auctioned off.

Caney and Hepburn (2011) identify four main advantages of cap-and-trade systems. Firstly, this policy ensures the achievement of emission targets, such as the Paris Agreement’s goal of holding the rise in global mean temperature to 1.5–2◦C above pre-industrial levels. Secondly, it is highly politically feasible, especially when the permits are given away for free. Thirdly, this system is cost-effective, as it minimizes the waste of resources and simultaneously induces cost-saving innovations. Lastly, it gives activists the possibility of buying permits to tighten the cap. On the other hand, they point out that the distributional consequences of these systems are a substantial disadvantage, in particular when the permits are allocated at no cost.

Although cap-and-trade systems achieve several beneficial environmental targets cost-effectively, the past literature has mostly found them to have regressive impacts. This is because people at the lower end of the income distribution spend a larger proportion of their income in energy bills and, as a result, an increase in the prices of heat and power is per se regressive. Nonetheless, very few studies have focused on this aspect of the EU ETS. In this thesis, I use a Country and Time Fixed effects regression analysis to evaluate the impact of the price of European allowances (EUA) on the index of consumer prices over the period ranging from 2008 to 2018 in the utility sector. My findings suggest that the EU ETS is regressive, however, this effect is quite small. Furthermore, my results hint that this regressive impact differs substantially across geographical regions and over the GDP per capita distribution. My main contributions are the use of realized historical data and the consideration of this effect over most of the countries participating in the programme. My findings substantiate those of previous studies in this field even though they are obtained with a fairly different methodology.

The next section presents the EU ETS, describing how it works, which countries participate in it and how it evolved over time. Section 3 offers a review on the current literature on the distributional consequences of cap-and-trade systems and, specifically, of the EU ETS. Section 4 describes the data employed in the study and its sources, while Section 5 discusses

the methodology used. I present my results in Section6 and conclude in Section 7with some considerations on my findings, the limitations of my model, and some suggestions for future research.

2

Policy Description

The European Union Emission Trading System (EU ETS) is a cap-and-trade system that covers all EU member countries plus Iceland, Liechtenstein and Norway (European Commission Climate Action,2016). The system places a ceiling on total emission allowed and allocates, for free or through an auction, permits to pollute. These permits are called EU allowances (EUA) give the holder the right to emit one ton of CO2. Then, companies have the freedom to trade these

permits, effectively imposing a price on carbon. The ability to trade allowances is what makes this policy cost-effective: put simply, firm A which has a lower marginal abatement cost, M ACA,

will reduce emissions and sell its surplus permits to firm B which has a higher marginal abatement cost, M ACB, at a price above M ACA and below M ACB (Tietenberg & Lewis, 2013). Since

2013, there is a single cap for all participating countries in the EU ETS, which is reduced by 1.74 percent every year (European Commission Climate Action, 2016). The limited supply and the differing marginal abatement costs of firms generate a positive carbon price which creates incentives for firms to invest in green technologies and clean energy.

Figure 1: EU ETS Phase 2. Data from the European Climate Exchange.

The EU ETS has developed over four different phases and the time period I employ in this thesis allows for observations regarding both the second phase, from 2008 to 2012, and part of the third phase, from 2013 to 2018. Specifically, the first phase ran from 2005 to 2007, it was a pilot phase, and its main scope was to acquire reliable emission data and determine the

appropriate emission cap for the next phase (European Commission Climate Action, 2015). The allowances price during this phase peaked in 2006 exceeding 30€ and then fell to around 0€ in 2007, as it became clear that permits were over-allocated. The second phase ran from 2008 to 2012 and it coincided with the Kyoto Protocol first commitment period. As visible from Figure1, the price of the EUA decreased considerably during this phase, which was mainly due to the global financial crisis and the sovereign debt crisis that burdened Europe in this period.

The third, and current, phase started in 2013 and will last until the end of 2020. This phase corresponds to the second commitment period under the Kyoto Protocol and is characterized by the harmonization of the EU ETS among the member countries (European Commission Climate Action, 2015). Moreover, from the beginning of this phase, all permits for the power generation sector have been auctioned off, while the proportion of free allocated allowances in the heat sector has been decreasing from eighty percent in 2013 to thirty percent in 2020. Figure 2

shows the trend of the EUA price in this phase, which has been overall positive, even if the price was still below that of 2008. The next and fourth phase will run from 2021 to 2030 and it aims at achieving a 43 percent emission reduction compared to 2005 levels by 2030 (European Commission Climate Action, 2020). Furthermore, in the fourth phase, the cap will decline at a rate of 2.2 percent instead of 1.74 percent of the previous phase (European Commission Climate Action,2016).

Figure 2: EU ETS Phase 3. Data from the European Climate Exchange.

Hintermann (2010) investigates the permit price drivers in the pilot phase of the EU ETS. His contribution is particularly important for my research to gather whether the EUA price was driven by the marginal abatement cost determinants and, thus, whether the allowances market was efficient. Hintermann (2010) finds that the permits price did not reflect the marginal

abatement cost in the first 16 months of the EU ETS. However, he shows that the market corrected itself after the first round of emission verifications. Hintermann (2010) identifies four possible reasons behind the initial inefficiency of the EUA market: a carbon price bubble, market power exerted by utility companies, firms hedging against the probability of having to pay a noncompliance fine, and inflating of historical emissions for next period allocations. The author finds that this was only an initial and transitory inefficiency that was corrected when actual emission data were published. This findings provide some confidence over the efficiency of the allowances market beyond 2006, and specifically in the second and third EU ETS phases that I take into considerations in this thesis. Namely, an efficient EUA price ensures that (1) the price only reflects its true determinants and that (2) the futures prices convey all the necessary information to the firms buying them (see Section 4).

3

Literature Review

The literature on the distributional consequences of environmental policies is much less extensive than that on their effectiveness. Nonetheless, most of these studies find climate policies to be regressive. For instance, carbon taxes have been estimated to have a regressive impact unless their revenues are recycled: the proceeds may be used for reductions in the capital income tax (Metcalf, 2005), recycled through a uniform lump-sum transfer (Klenert & Mattauch, 2016), or through an increase of the personal income tax credit (Callan et al., 2009). Similarly, most studies of cap-and-trade systems agree on the policy being regressive. However, there are several different views on the policy design that could reduce this regressive impact and there is much disagreement on the appropriate way to model this problem. As a matter of fact, there is much unresolved complexity on what is the behaviour of the government and the firms and households involved. Furthermore, there is quite some dissent in previous research on the extent of the regressive impact of a cap-and-trade system.

Farber (2012) presents a theoretical discussion of the fairness of cap-and-trade systems to low-income consumers. He highlights the importance of direct energy costs when studying the distributional consequences of emission trading. A raise in direct energy costs is visible in a utility bill, while an increase in indirect energy costs represents a higher price for energy-intensive goods or services(Farber, 2012). Given a carbon price, direct energy costs will rise more than indirect energy costs, due to a shorter supply chain, a higher market concentration, and a higher cost pass-through rate. In this thesis, I will only consider the direct energy costs by looking at the effect of the EUA price on the index of heat and power sectors consumer prices. A key concept for this analysis is that an increase in energy prices is by itself regressive: it affects lower-income people more than higher-income people because poorer households spend a larger share of their income in power and heat (Farber,2012; Kaswan, 2009; Rausch et al.,2011). However, Farber (2012) argues that this regressive effect can be softened by subsidizing energy efficiency measures,

such as wall insulation and new efficient boiler installation. Moreover, if permits are auctioned the revenues can be redistributed in the form of income tax cuts or through an expansion of the earned income tax credit (Farber, 2012). On the other hand, free allocation of allowances tends to be more regressive as it benefits the shareholders of the energy companies that sell the permits and those shareholders have, generally speaking, a relatively high income (Farber, 2012).

Andersson (2018) conducts a study that evaluates the distributional consequences of carbon pricing based on inflation effects, similarly to what I do in this paper. In particular, Andersson (2018) uses a scenario analysis to estimate the welfare effects of a global carbon price. He finds that inflation effects are very small in all developed countries, even under the strong assumption of a 100 percent cost pass-through rate. Moreover, he claims that if the carbon price was introduced gradually over a period of two to three years the welfare effects would be small even in the short-run, which would make the measure politically more feasible. Andersson (2018), then, decomposes the inflation effects by sector. He finds those to be negligible in all sectors but the utilities. He argues that the main reason for this is the lack of competition. Firms in an oligopolistic market have the ability to pass their higher costs onto higher prices for consumers, which are reflected in his estimation of higher inflation effects in this sector. Low levels of competition reduce the incentives to move production to more carbon-efficient technologies, thus making the carbon price less effective. Based on this conclusion, I include a market concentration proxy in my regression model to control its effect on the price level.

Dinan and Rogers (2002) show that the incidence of a cap-and-trade system is mostly determined by policy design. The government influences the extent of the deadweight loss and the burden on households through its decisions over permits allocation and revenue recycling. Dinan and Rogers (2002) claim that the government would earn some allowances rent even if the permits were given away for free, because the free allowances cause the receiving firms to make higher profits, which, in turn, result in higher tax revenues. These revenues could be used to neutralize the regressive effects of emission trading or to reduce pre-existing distortionary taxes. Dinan and Rogers (2002) argue that the latter objective would bring a smaller cost to the economy than the former, yet auctioning the permits may enable the government to reach both these goals.

The literature especially on the distributional consequences of the EU ETS is not very large and only a few studies have estimated the effects across the multiple participating countries. Sager (2019) investigates the EU ETS cost to consumers using a non-homothetic gravity approach. Simulating a constant carbon price of 30$/ton, he finds the EU-wide carbon price to be regressive. Specifically, he estimates the cost to the bottom ten percent of the income distribution to be around two percent of their budget, while less than one percent for the top half. Yet, Sager (2019) shows that this regressive effect varies considerably across countries and that Eastern-European and Baltic nations are the ones that suffer the largest welfare loss. To test whether this is true for my model as well, I run regressions on restricted samples by geographical regions and by countries’ incomes and I, indeed, find substantial differences on the estimated price effects. Sager (2019) argues that two mechanisms cause the EU ETS to be regressive, lower-income consumers consume more emission-intensive products and lower-income countries have more emission-intensive firms. Lastly, Sager (2019) claims that climate mitigation benefits may reduce the regressive effect of this policy, especially for countries that are both hot and poor - where climate mitigation has a strong positive impact.

Beznoska et al. (2012) compute the incidence of the EU ETS on German households. They find that this policy has a regressive effect on private households, yet this effect is quite small. They estimate the burden of the EU ETS, quantified through Compensating Variation, to be 0.8 percent of total consumption. Moreover, they construct a consumption distribution that they

employ as a proxy of the lifetime income. Beznoska et al. (2012) find that the lowest decile of this distribution bears a burden of 1.1 percent while the highest decile of 0.7 percent. Their approach is quite different from previous literature as they employ the ”Almost Ideal Demand System” consumer behaviour model to estimate elasticities, which they use to compute the effect of the EU ETS on household income. Then, they estimate the effect of revenue recycling and conclude that it can neutralize the regressive effect and even make the policy progressive depending on the design.

The main difference of this paper from the previous literature is that I use ex-post actual data to evaluate the relationship between the emission permits price and the index of consumer prices. Contrary to the simulations and predictions models most commonly used in this field, I do not need to make strong assumptions about the behaviour of the agents involved since I use historical EUA prices and realized price levels. Moreover, I estimate the distributional consequences of the EU ETS over both the second and the third phase with a sample that covers most of the participating countries. I build upon the design by Andersson (2018) of estimating the inflation effects of carbon pricing, but I apply it to the specific case of the EU ETS. Additionally, my findings are in line with previous literature, even though I use a quite different and simpler approach.

4

Data Description

My data set consists of annual observations for the time period ranging from 2008 to 2018 for 27 countries (the full list is in Appendix9.1). I consider only the countries that have been part of the EU Emission Trading Scheme for the entire time period in question and are part of the EU. I do not include Croatia, because it entered the EU in 2013. My time period allows for the inclusion of the United Kingdom, as it officially left the EU on the 31st of January 2020. Although Iceland, Liechtenstein, and Norway participated in the EU ETS for the entire period, I do not include them due to lack of data 1.

I estimate the effect of the emission permits price on the price index, and I include several control variables. Firstly, I add a market concentration proxy that would otherwise cause Omitted Variable Bias, since (low) competition is a determinant of final consumer prices and strongly correlates with EUA price, as shown in the literature review section (see Andersson, 2018). Secondly, the energy intensity controls for final demand for energy, and specifically, for the structural and behavioural changes in the final consumption of energy that are not due to a change in the price of emission permits. Thirdly, the Gross Domestic Product (GDP) captures the general macroeconomic trend in the period considered, which is crucial given that both the Great Recession and the Sovereign Debt crisis hit during the time period of interest.

Furthermore, I add input fuel prices over concerns about Omitted Variable Bias, since they affect firms’ demand for those fuels, which directly influence firm’s demand for emission permits, and they are reflected into final consumer energy prices through firms’ variable costs. Inflation and unemployment have a strong negative relationship in the short-run dictated by the Phillips

1

These three countries are part of the European Economic Area (EEA) but not of the EU. The lack of data can be fully attributed to the fact that the Eurostat publishes data collected by the European Member States statistical authorities and, thus, only seldom reports data for countries outside the EU (Eurostat,2015). Hence, this data is missing at random and is not expected to result in bias.

curve, which suggests that adding the unemployment rate to my regression should increase the accuracy of my estimates. Additionally, I control for the negative relationship between inflation and the interest rate. This relationship has been proven to be empirically sound and it is at the core of Central Banks’ common practice of controlling inflation by influencing the interest rates through open market operations, standing facilities rates, and reserve requirements (Castillo-Martinez & Reis, 2019; Mishkin et al., 2013). Lastly, I include the cooling and heating degree days, as I expect omitting these variables would bias my estimates since a larger number of cooling or heating days raises utility bills and, thus, affects the price level as well as they determine the EUA price based on the price drivers model by Hintermann (2010).

Mean Standard Dev. Minimum Maximum HICP 97.9663 11.30157 56.95 145.27 EUA 10.68882 5.700459 4.331 23.458 Competition 196.1751 394.9068 1 2399 Intensity 581.6869 201.2993 156 1095 Crude oil 80.92182 24.5304 43.64 111.63 Coal 10.90213 3.110297 6.966674 18.16386 Natural gas 3.936364 1.690376 2.52 8.86 GDP 97.44068 5.789241 74.118 112.054 Unemployment 9.009428 4.582662 2.2 27.5 Euribor .0076673 .0136059 -.00368 .04239 Cooling 113.2388 182 0 805.71 Heating 2851.9 1163.953 322.12 6190.94 N 297

Table 1: Summary statistics table for the variables described in this section.

Index of Consumer Prices The dependent variable is the annual Harmonised Index of Consumer Prices (HICP) measured according to the subclass ”electricity, gas and other fuels” (COICOP 04.5) - with base year 2015 equals 100. The price index is computed as a fraction

of the consumer expenditure in one year over the consumer expenditure for the same basket of goods in the base year, which is, then, multiplied by a hundred. Moreover, this index is calculated using the same methodology in all EU members, which ensures comparability across countries. The data on the HICP is retrieved from the Eurostat database (Eurostat, 2020) and is presented in Figure 3to show the general trend in the EU. I opt for using the price index over the inflation rate due to data availability and completeness.

European Allowances Price The predictor variable of interest, called EUA, is the European carbon price. Specifically, this price is the annual average of historical settle price of European Allowances (EUA) continuous futures. The EUA futures are traded on the European Climate Exchange, which is currently owned by the Intercontinental Exchange (ICE) and is the source of my data. The choice of using futures over spot prices was determined by the volume of each market. Spot trading for EUA has been very rare until 2018 since the regulation imposed by the Markets in Financial Instruments Directive (MiFID) only covered derivatives trading (European Commission Climate Action,2015). This directive, which applied from 2007 to 2018,

Figure 3: Price index average over the participant countries in the time period of interest.

had as its main goals the harmonization of investor protection and market transparency across all EU members (European Securities and Markets Authority, 2020). Thus, to protect market participants, carbon exchanges started offering financial instruments for the delivery of emission permits, such as daily futures (European Commission Climate Action, 2015). The need for these derivative contracts disappeared in 2018 when the MiFID II entered in force, as the latter establishes that similar rules apply to both spot and derivative trading (European Securities and Markets Authority, 2020). Furthermore, the Efficient Market Hypothesis states that prices are efficient if they convey all available and relevant information and, as a consequence, the difference between spot and derivatives price should depend exclusively on their distinct risk - Hintermann (2010) shows that the EUA price became efficient in 2006 (see Section 2).

Market Concentration The variable termed Competition is a proxy for market concentration. It is calculated as the sum of the total number of retailers that sell electricity to final consumers plus the total number of retailers that sell natural gas to final consumers. The data is obtained from the Eurostat database (Eurostat, 2020), the CEER Monitoring Report (2019), national reports (Competition Authority Estonia, 2019; Energy Authority Finland, 2018; European Commission, 2014b), and national agencies data (Regulatory Office for Network Industries Slovakia, 2020a,2020b). The cases of Cyprus and Malta constitute an exception as natural gas is not sold to final consumers, and, thus, a retail market does not exist (European Commission,

2014a, 2017; Eurostat, 2019). Additionally, both these countries have only one retailer of electricity for the entire time period considered. On the contrary, Germany and Italy have the highest number of retailers overall. All in all, larger values for this variable imply a fiercer competition, and the data shows that most of the countries have highly concentrated retail markets for electricity and natural gas.

Energy Intensity Households energy intensity is defined as final households energy consump-tion divided by the populaconsump-tion. The data is collected from the Eurostat database (Eurostat,

2020). This variable is a measure of the amount of energy needed to generate one unit of GDP (European Environment Agency, 2020). Energy efficiency improvements lower the value of this variable, but so does importing most energy-intensive manufactured products from abroad. A negative trend in energy intensity indicates a shift towards the decoupling of economic growth and energy consumption (European Environment Agency, 2020) and is considered a desirable environmental outcome.

Figure 4: Crude oil price trend between 2008 and 2018.

Input Fuel Prices The input fuel prices are the wholesale prices of crude oil, coal, and natural gas. The crude oil price is the Europe Brent spot price (free on board) measured in US dollars per barrel. The data is obtained from the U.S. Energy Information Administration (2020a). The trend in the crude oil price in the relevant time period is displayed in Figure 4, where it is visible that it shares quite some similarity with the price level trend of Figure 3 -this reinforces the hypothesis that omitting this variable would bias the regression estimates. The coal price is the Northwest Europe marker price measured in current US dollars per megawatt-hour (MWh). The data is obtained from the Our World in Data project (Our World in Data,2019). The natural gas price is the Henry Hub natural gas spot price measured in dollars per million of British Thermal Unit (BTU). The data is obtained from the U.S. Energy Information Administration (2020b). Contrary to crude oil and coal, natural gas trade is still mainly regional and highly regulated (Mazighi, 2005). As a consequence, no official international price reference for natural gas exists and most of the gas traded internationally is indexed to crude oil. However, in the past few years, the Henry Hub price has emerged as one of the most relevant globally thanks to the level of trade of this hub and its use as a benchmark on the New York Mercantile Exchange

(NYMEX). Thus, in this paper, I will assume that the Henry Hub price is a good approximation of the natural gas price for industrial consumers in the countries in the data set.

Gross Domestic Product The GDP is the real Gross Domestic Product at market prices measured with the implicit price deflator (base year 2015) and expressed in euros. The data is obtained from the Eurostat database (Eurostat, 2020).

Unemployment The unemployment rate is the share of unemployed people as a percentage of the labour force. The European Commission Regulation defines unemployed persons as all persons between the age of 15 and 74 who (1) did not work in the reference week, (2) were available to start working within a two weeks period, and (3) were actively looking for a job in the month before the reference week. The labour force is defined as the total number of employed and unemployed people. The data is retrieved from the Eurostat database (Eurostat,2020).

Interest Rate The interest rate is the Euro Interbank Offered Rate (Euribor) with maturity of one month. Data is obtained from the Euribor website (Euribor, 2020). The Euribor rates are the most used reference rates on the European money market, as they serve as a benchmark for financial products ranging from futures to mortgages. The Euribor is calculated as the average interest rate at which a large panel of banks in the EU borrows funds from each other. As shown in Figure 5, the interest rate decreased rapidly during the Great Recession and kept a negative trend over most of my sample. The decrease during the financial crisis is mainly attributable to the expansionary monetary policy put in place by the European Central Bank (ECB) to counteract the credit-crunch and to stimulate the economy.

Cooling and Heating Days Cooling and heating degree days are measures of the need for air-conditioning and heating in buildings. They are computed based on the outdoor temperature and the average room temperature and they are strong predictors of energy consumption. The data is produced monthly by the Joint Research Centre, and, then, is published by the Eurostat in annual form - yearly data are the sum of monthly observations (Eurostat,2020). The European Environment Agency (2019) reports that the annual population-weighted heating degree days declined by 6 percent in Europe between the periods 1950–1980 and 1981–2017. In the same time span, the population-weighted cooling degree days increased by 33 percent, a trend that is expected to continue in the future and which may threaten the stability of electricity network in events such as summer heatwaves (European Environment Agency,2019).

5

Methodology

I employ a panel data setting with large N, 27 countries, and small T, 11 years. The main benefits of using this framework in my study are the possibility to control for country heterogeneity, and the increased variability and efficiency compared to a time-series or a cross-section (Baltagi,

2008). On the other hand, data collection issues and the short time period setting pose some limitations. Below, I present my regression model, Equation (1), and discuss the estimation techniques I apply to this panel data framework.

HICPit= β0+ β1∗ EUAit+ β2∗ Competitionit+ β3∗ Energy Intensityit

+ β4∗ Crude Oilit+ β5∗ Coalit+ β6∗ Natural Gasit

+ β7∗ GDPit+ β8∗ Unemploymentit+ β9∗ Interest Rateit

+ β10∗ Coolingit+ β11∗ Heatingit+ αi+ γt+ it

(1)

Equation1shows my main specification. αi captures the unobservable time-invariant country

heterogeneity, while γt is the unobservable time-specific country-invariant heterogeneity, and it

is the idiosyncratic error.

An Ordinary Least Squares (OLS) regression would result in biased and inconsistent estimates of my coefficients as a result of the omission of the unobserved country effect, which is very likely to be correlated with the included regressors and with the regressand. Two different approaches can be adopted to correct for this, first difference or fixed effects. Foremost, both these methods require each explanatory variable to have some variation over time, and this assumption is fulfilled in my model (yet, the market concentration proxy changes quite slowly over time). While these two econometric techniques produce the same estimates when two time periods are taken into consideration, there are substantial differences in efficiency when the sample contains three or more time periods. Firstly, first difference consists in differencing each variable over time such that αi, which is time-constant, drops out of the equation. Secondly, the

fixed effects technique applies a time-demeaning transformation to the sample data or includes a dummy variable for each country in the regression. The main distinction between the two models is in the number of observations, employing the first difference method results in one time period less than the fixed effects method. This is of particular relevance in my study because of the very short time dimension considered. Wooldridge (2012) shows that fixed effects produce more

efficient estimates than first difference if the idiosyncratic errors are serially uncorrelated, whereas the opposite is true when the it exhibits considerable, positive serial correlation (a case termed

”Random Walk”). Serial correlation is very unlikely to be a problem in short time period panel data (Wooldridge, 2010), and, given that my T is only 11, I assume negligible autocorrelation in my sample. Thus, the fixed effects estimators are more efficient than the first difference ones.

I expect the unobservable time-invariant country effect to be correlated with both the dependent and the independent variables and, thus, that its omission would cause Omitted Variable Bias. On the contrary, if there were unobservable characteristics that affected only the dependent variable, the estimates of the error would be incorrect but there would not be any Omitted Variable Bias. In the latter case, employing random effects is the correct approach to take. I formally test whether fixed effects are more appropriate than random effects with the Hausman test. Figure6shows the output of the test, which rejects the null hypothesis confirming my belief that fixed effects better suit the model.

Figure 6: Output of the Hausman Test

The Gauss-Markov assumption of homoskedasticity implies that the variance of the unobserv-able error it, conditional on the independent variables, is constant (Wooldridge, 2012). In panel

data, this assumption is often invalid, causing standard errors to be incorrect. To determine whether my model suffers from heteroskedasticity, I run the modified Wald statistic for group-wise heteroskedasticity in fixed effects models (”xttest3” by Greene, W.). As visible from Figure 7, the null hypothesis of homoskedasticity is rejected. Thus, I correct for heteroskedasticity by using heteroskedasticity-robust standard errors in my specification.

Figure 7: Output of the test for heteroskedasticity

The omission of time-specific effects that correlate with both the regressand and the regressors will bias the regression estimates, just as the country effects. There are two main events in my sample that would cause Omitted Variable Bias if they were not controlled for: the Great Recession and the Sovereign Debt Crisis, which struck Europe from 2008 to 2012. As discussed in Section 2, the drop in the EUA price during the second phase is largely attributable to the economic downturn. Moreover, the price level increased in this period (see Figure3) mainly due to the rise in oil prices (see Figure 4), the increase of taxes due to the ”austerity” measures, the Quantitative Easing, and the zero interest rates. To account for this, I include a binary variable

that takes the value of one for the years 2008 to 2012 and zero otherwise. One could argue that there are other events or unobservable time-specific characteristics that should be taken into account to prevent Omitted Variable Bias. If that was the case, it would be appropriate to include a dummy variable for each year in the sample (Time Fixed Effects model). Note that all the variables in the model vary over time but some do not vary over the different countries. Specifically, the EUA price, the input fuel prices, and the interest rate are country-invariant. Since Time Fixed Effects control for factors that are constant across countries but change over time (Stock & Watson,2015), some of the year dummies will be thrown out of the regression due to perfect multicollinearity. Figure 8 shows the result of a Wald test with the null hypothesis that the year dummies are jointly equal to zero. I reject the null and conclude that Time Fixed Effects are useful in my specification even in the presence of perfect multicollinearity.

Figure 8: Output of the test for time dummies joint significance

Imperfect multicollinearity arises with the inclusion of a variable that highly correlates with the independent variables already present in the model and that does not contribute much to explaining the variation in the outcome variable (Stock & Watson, 2015). This inflates the variance and, thus, the standard errors of the coefficients, resulting in imprecise -though unbiased-estimates. A commonly used measure of collinearity is the variation Inflation Factor (VIF), which is computed by regressing each independent variable j against every other independent variable to obtain R2_j and plugging it in the formula V IFj = 1/(1 − R2j). Wooldridge (2012)

suggests a VIF value of 10 as a threshold for problematic multicollinearity, though he argues this is quite arbitrary. In my study, multicollinearity is not an issue when estimating the model with OLS, but VIF values become quite large when introducing Country Fixed Effects and they become even more substantial when introducing Time Fixed Effects (see Appendix 9.2for the VIF tables). Another issue that increases the variance of the coefficients in the model is small sample size. The small sample I work with and the relatively high VIF values suggest that the estimators will be quite imprecise and it will be harder to find statistical significance.

In conclusion to this section, I present my expectations regarding the results of the regression. I expect the coefficient of interest, β1, to be positive, since my hypothesis is that an increase in

the price of emission permits causes the price index to increase. On the other hand, I expect the coefficients of my market concentration proxy, β2, to be negative, because a larger number of

retailers implies fiercer competition which should lower energy prices and, thus, the price level. I expect a positive coefficient for energy intensity, β3, as a decline in energy intensity implies a

index. The coefficients of the input fuel prices, β4, β5, and β6, are expected to be positive since

the higher the price the producer or wholesaler faces for fuels the higher the final consumer prices. Moreover, I anticipate a positive effect of GDP, with coefficient β7, on the price index once the

effect of the 2008 to 2012 crisis is controlled for. I expect the coefficient for the unemployment rate, β8, to be negative based on the Phillips Curve. Then, I anticipate a negative relationship

between the price level and the interest rate, with coefficient β9, based on the empirical evidence.

Lastly, I expect the coefficients of cooling and heating degree days, β10 and β11, to be positive

because an increase in either of these variables raises energy bills and, hence, the price index.

6

Results

Table 2presents the estimated coefficients of Equation (1) using different econometric methods. The first column reports the results of a simple Ordinary Least Squares regression. The coefficient of interest is statistically insignificant and has the opposite sign compared to the original expectations, which is unsurprising given the Omitted Variable Bias produced by disregarding country heterogeneity. Adding the crisis dummy - this is a binary variable that takes up value of one in the years 2008 to 2012 and of zero otherwise - controls for the effect of the Great Recession and Sovereign Debt Crisis and has the immediate effect of turning the coefficient of interest positive, though still insignificant. The crisis dummy has a significant large negative effect on the price index. A few other control variables are statistically significant, such as Competition that has the intended negative sign. Interestingly, the coefficient for the price of natural gas is significant at the one percent significance level, yet it has a negative sign. The latter seems to persist across all the different estimation techniques employed which could be explained by (1) the Henry Hub price is not an appropriate reference price for natural gas wholesale trade in Europe, or (2) providers of natural gas to consumers have lowered its price to increase demand since gas is a cleaner fuel. This second reasoning is based on the current global trend for the electricity generation sector to use natural gas over more polluting fossil fuels (International Energy Agency,

2019), such transition benefits European energy providers through a lower EU ETS cost and possibly national subsidies. Similarly, Hintermann (2010) finds that in the first phase of the EU ETS the preferred abatement method has been switching from coal to natural gas in the power generation industry. An additional point is that the EU is investing in stabilizing gas supply in Europe through new gas corridors and gas-related infrastructures (European Commission,

2020). Overall, I believe a combination of the above factors can justify the negative effect of the natural gas price on inflation, but a more detailed investigation should be made in future research through the collection of data on regional gas prices and national and European subsidies to companies.

Models 3 and 4 are estimated using Country Fixed Effects, model 4 additionally controls for the effect of the 2008 to 2012 economic downturn. Just as in the OLS, adding the crisis dummy makes the coefficient of interest change sign. Moreover, the coefficient for the price of the EUA is statically significant at the five percent significance level in model 4. Thus, after controlling for the country effect and the economic crisis, a one-euro increase in the price of emission permits is associated with an increase in the price index of 0.22 points, ceteris paribus.

Dependent Variable: Price index for electricity, gas and other fuels (1) (2) (3) (4) (5) (6) (7) (8) EUA -0.255 0.0772 -0.0487 0.222** -0.149 -0.150 0.310** 0.183** (0.165) (0.136) (0.0837) (0.0934) (0.191) (0.196) (0.145) (0.0906) Competition -0.00150** -0.00134** 0.0110 0.00758 -0.000477 0.00587 0.00587 0.00633** (0.000708) (0.000613) (0.00818) (0.00675) (0.00103) (0.00660) (0.00660) (0.00307) Intensity 0.0108*** 0.0121*** -0.0407 -0.0297 0.00995 -0.0203 -0.0203 -0.00448 (0.00406) (0.00390) (0.0308) (0.0280) (0.00831) (0.0284) (0.0284) (0.0163) Crude oil 0.199*** 0.245*** 0.167*** 0.205*** 0.448*** 0.448*** 0.0812 0.168*** (0.0253) (0.0251) (0.0440) (0.0355) (0.114) (0.115) (0.0752) (0.0377) Coal -0.329 0.449 -0.0189 0.739* 1.660** 2.277*** 2.169*** -0.315* (0.361) (0.423) (0.299) (0.406) (0.700) (0.768) (0.737) (0.186) Natural gas -0.328 -3.602*** -0.854 -3.851*** -12.37*** -14.34*** -4.508* -0.185 (0.731) (1.013) (0.512) (0.655) (4.180) (4.662) (2.355) (0.528) GDP 0.286** 0.0306 0.494* 0.177 -0.0945 0.0347 0.0347 -0.0414 (0.139) (0.149) (0.266) (0.285) (0.305) (0.307) (0.307) (0.139) Unemployment -0.333*** -0.216* 0.725 0.714 0.245 0.705 0.705 0.702** (0.127) (0.118) (0.490) (0.431) (0.271) (0.427) (0.427) (0.280) Euribor -274.1*** 84.60 -171.3** 140.0** 638.4 837.9* 1300.0** 119.6* (100.0) (127.8) (77.17) (67.34) (395.9) (434.0) (588.2) (62.18) Cooling 0.0187*** 0.0215*** 0.0115 0.00520 0.0131* -0.00992 -0.00992 0.0112 (0.00580) (0.00581) (0.00942) (0.00973) (0.00692) (0.0131) (0.0131) (0.00793) Heating -0.0000804 0.000216 -0.00105 0.00362 -0.000228 0.00127 0.00127 -0.000102 (0.000803) (0.000767) (0.00432) (0.00359) (0.00190) (0.00382) (0.00382) (0.00227) Crisis -11.97*** -11.79*** -70.56** 3.096* (2.729) (3.432) (27.91) (1.615) cons 58.75*** 80.52*** 58.46* 71.66** 99.04*** 97.04*** 87.58*** 1.355** (14.36) (14.95) (29.98) (28.74) (30.88) (32.25) (31.04) (0.579)

Country FE NO NO YES YES NO YES YES FD

Time FE NO NO NO NO YES YES YES NO

N 297 297 297 297 297 297 297 270 R2 0.430 0.485 0.519 0.566 0.584 0.584 0.392 Adj. R2 _{0.408 0.463 0.500 0.548 0.561 0.561 0.363} R2within 0.519 0.566 0.560 0.584 0.584 R2_{between 0.00620 0.102 0.149 0.270 0.270} R2overall 0.0763 0.165 0.462 0.192 0.192 F 18.39 19.45 23.54 30.91 52.26 52.26 16.09

Standard errors in parenthesis * p < 0.10, ** p < 0.05, *** p < 0.01

This effect is quite small, especially when compared to the trend of the price level in the period considered (see Figure 3, where yearly changes are fairly large). When investigating the control variables, the interest rate coefficient has an unexpected behaviour, as it is positive and significant. Interestingly, this coefficient is positive in all the regressions that control for the effect of the recession or for the time effect or both. This suggests the presence of collinearity of the interest rate with the crisis variable and with the time dummies, an issue that can easily affect the signs of highly correlated variable coefficients (Kumar, 2020). To test whether this is the case, I compute the correlation between these variables: it is 0.6999 with the crisis binary and 0.4198 with the 2009 year dummy, both are significantly different from zero at the one percent significance level. I infer that there is a negative relationship that links the interest rate to the price index, yet it is misidentified in the model due to the even stronger correlation of the interest rate with the crisis dummy and the 2009 year dummy.

Model 5 presents the result of estimating Equation (1) with Time Fixed Effects exclusively. The year dummies from 2014 to 2018 are dropped due to multicollinearity (see Section5 for a more extensive discussion on this). The coefficient of interest is negative and insignificant, due to the Omitted Variable Bias generated from the exclusion of the country effects and the economic downturn control. Model 6 shows that adding Country Fixed Effects alone does not improve the results confirming that the time effects do not capture the full impact of the economic crisis. However, they do control for a part of this effect since the coefficients of the year dummies turn positive when adding the Crisis variable to this model (from Column 6 to Column 7, these coefficients are omitted from the table). Model 7 returns a coefficient for the emission permits price that is positive and significant at the five percent significance level. The magnitude is larger than that estimated in Model 4, a one-euro increment in the allowances price is associated with a 0.31 points increment in the price index, everything else held constant. Most of the control variables lose significance in this model due to the high level of multicollinearity.

Model 8 is estimated using first difference. As discussed in Section 5, first differencing reduces T by one and the estimators are less efficient relative to the Country Fixed Effects model. Wooldridge (2012) claims that it is good practice to check whether the results are sensitive to the use of first difference instead of Country Fixed Effects, since a low sensitivity suggests that the assumption of no serial correlation in the sample is not a particularly strong one. The coefficient of interest estimated with first difference is significant, positive, and its magnitude very similar to that estimated in model 4. The effect of the EUA price on the price index is estimated to be 0.039 points smaller when using first differencing compared to using Country Fixed effects. Thus, the sensitivity of the results to these different estimation approaches is very low. A last note on Model 8 is that coefficient for the crisis binary variable is positive because this variable is not differenced over time, while all others are. Hence, Model 8 estimates that being in a time period during the economic downturn of 2008 to 2012 is associated with an increase in the one-period difference in the price index of 3.096 points, as opposed to Model 4 which estimates that being in a time period during the economic downturn of 2008 to 2012 is associated with a decrease of 11.79 points in the price index.

Model 7 is the most appropriate model to estimate my relationship of interest as it is very unlikely to suffer from Omitted Variable Bias and its estimators are the most efficient. Moreover, the F-test null hypothesis that the regressors are jointly insignificant is rejected at the one percent

significance level. The within R2 shows that 58.43 percent of the variation in the price index within countries is captured by the model. This model estimates that a one-euro increase in the emission permits price is associated with a 0.31 points increase in the price index on average, holding everything else constant. To put this result into perspective, the price index in the countries and time periods considered in this thesis experienced an average annual change of 1.84 points overall and my findings suggest that only a change of 0.31 points can be attributed to the EU ETS. Hence, my results indicate that the EU ETS per se has a regressive impact in the heat and power sectors, yet the extent of this impact is quite small. This regressive effect may have been counteracted with measures such as revenue recycling programs, but whether those were in place and were effective during the time period of interest is beyond the scope of this thesis. The small regressive impact of this cap-and-trade system in the energy sector is consistent with most of the previous literature, with the main difference that this thesis employed Country and Time Fixed Effects on historical data to estimate it.

Sager (2019) finds the distributional consequences of EU ETS to be very different across countries and to be the most regressive in Eastern European and Baltic countries. To test whether this holds for my sample as well, I run the regression with Country and Time Fixed Effects on restricted samples by region. I use the M49 Standard of the United Nations (United Nation Statistical Division, 1999) to divide the countries in my sample into four regions: North, West, East, and South (the full list of countries in each region is presented in Appendix 9.1). The results, presented in Table3, show that the effect of the emission permits price on the price index has been significant and large for the Northern European countries, larger than the unrestricted sample effect. On the other hand, the coefficients of interest for the other three regions are statistically insignificant, and the one for the Southern European countries is very small, smaller than the unrestricted sample estimate. The within R2 is above 0.67 for all four regressions, showing that more than 67 percent of the variation in the price index within countries is captured by each regional model.

A further comparison can be done by restricting the sample based on the countries’ income. I obtain data on Gross Domestic Product per capita expressed in Purchasing Power Standards for the year 2018 for all the countries in my data set from the Eurostat database (Eurostat, 2020). Then, I create three subsamples: the poorest five countries (Bulgaria, Greece, Latvia, Poland, Romania), the richest five countries (Austria, Denmark, Ireland, Luxembourg, Netherlands), and the five countries in the middle of this national income distribution (Cyprus, Czech Republic, Italy, Malta, Spain). As shown in Table 4, I find that a one-euro increase in the EUA price is associated with a price index increase of 0.612 points in the five poorest countries and 0.480 points in the five richest countries. Both of these results are significant at the five percent significance level. On the contrary, the coefficient of interest for the middle five countries of the national income distribution is insignificant and negative. These findings suggest that (1) the regressive impact of the EU ETS is much larger for poorer countries, which is in line with Sager (2019), and (2) the countries in the middle of the GDP per capita distribution experience an

Dependent Variable: Price index for electricity, gas and other fuels

(1) (2) (3) (4)

North West East South

EUA 0.590** 0.317 0.623 0.148 (0.195) (0.407) (0.409) (0.364) Competition 0.0773 -0.000212 0.0312 0.00938 (0.0680) (0.00826) (0.0217) (0.0182) Intensity -0.0343 0.00190 0.0899 -0.128 (0.0322) (0.0248) (0.0859) (0.0770) Crude oil 0.0496 0.149 -0.00745 0.0824 (0.0909) (0.213) (0.169) (0.205) Coal 1.159 0.412 0.809 3.521 (0.845) (1.740) (1.324) (1.985) Natural gas -1.377 -1.361 -2.213 -5.611 (2.251) (10.63) (2.415) (6.807) GDP 0.0877 -0.926 0.0440 -0.316 (0.271) (0.908) (0.528) (1.046) Unemployment 0.202 -1.251 2.473 0.341 (0.642) (1.271) (2.361) (0.771) Euribor 1354.6 -377.1 -706.2 2534.3 (771.5) (2034.6) (1184.9) (1962.3) Cooling -0.0658 0.0524 -0.0123 0.00993 (0.0676) (0.0794) (0.0206) (0.0289) Heating 0.0111 -0.00365 -0.0112 0.0131 (0.00672) (0.0104) (0.00980) (0.00814) Crisis -86.87** -5.208 14.22 -130.5 (34.18) (66.77) (46.30) (89.79) cons 53.98** 198.1* 57.96 135.0 (20.24) (93.01) (33.67) (113.2)

Country FE YES YES YES YES

Time FE YES YES YES YES

N 88 66 66 77 R2 0.753 0.679 0.676 0.677 Adj. R2 _{0.697 0.575 0.570 0.590} R2_{within 0.753 0.679 0.676 0.677} R2_{between 0.262 0.0488 0.425 0.00132} R2overall 0.438 0.519 0.436 0.257

Standard errors in parenthesis * p < 0.10, ** p < 0.05, *** p < 0.01

Dependent Variable: Price index for electricity, gas and other fuels

(1) (2) (3)

Top 5 Poorest Top 5 Richest Middle 5

EUA 0.612** 0.480** -0.323 (0.191) (0.116) (0.360) Competition 0.0433 0.107 0.0118 (0.0414) (0.0853) (0.0118) Intensity -0.0722 -0.0156 -0.203 (0.0442) (0.0179) (0.120) Crude oil -0.0941 -0.00242 0.252 (0.0759) (0.0793) (0.199) Coal 2.869 2.528 5.917** (1.443) (2.034) (2.055) Natural gas -2.533 -3.864 -17.93* (2.823) (7.119) (6.531) GDP 0.113 -0.336 -0.0952 (0.180) (0.367) (0.746) Unemployment 1.199** -1.129* -0.810 (0.289) (0.433) (0.992) Euribor 1447.4 3259.0 4349.3 (1313.7) (1866.1) (2085.7) Cooling 0.00771 -0.0692 -0.0162 (0.0331) (0.0357) (0.0338) Heating 0.00236 0.00619 0.00555 (0.00619) (0.00428) (0.0100) Crisis -97.52 -159.6* -159.0 (56.04) (62.34) (104.4) cons 80.57* 116.3* 191.0 (30.12) (49.84) (104.0)

Country FE YES YES YES

Time FE YES YES YES

N 55 55 55 R2 _{0.888 0.765 0.667} Adj. R2 _{0.840 0.666 0.527} R2within 0.888 0.765 0.667 R2_{between 0.0800 0.0784 0.180} R2_{overall 0.606 0.0979 0.210}

Standard errors in parenthesis * p < 0.10, ** p < 0.05, *** p < 0.01

The estimates of the coefficient of interest are insignificant for the middle five countries of the national income distribution and for three out of four geographical regions, which might be due to the small size of the restricted samples or to multicollinearity. Moreover, data on the emission intensity of supply of those regions and the middle five countries does not seem to provide an economically sound explanation for these results, which remain quite puzzling. On the other hand, the significant estimators show that the regressive effect of this cap-and-trade system might be quite larger for some regions or groups of countries than the overall effect. This should be taken into account by policymakers when considering redistributions and revenue recycling programmes intended to improve the distributional consequences of EU ETS.

7

Conclusion and limitations

This thesis estimated the effect of the European Allowances price on the Harmonized Index of Consumer Prices using Country and Time Fixed Effects. My results suggest that the EU ETS had a significant yet small regressive effect in the heat and power sectors in the time period between 2008 and 2018 in the 27 European countries considered. This finding is in line with the previous literature on this topic, even though I employed realized actual data instead of the most commonly used predictions and simulations. Moreover, I find that there are moderate differences in the estimated relationship of interest both across geographical regions and over the GDP per capita distribution. Specifically, I find that a one-euro increase in the price of the emission permits is associated with a larger than overall rise in the price index in the northern region of Europe, which includes Scandinavian and Baltic nations and the British-Irish Islands. However, the estimated coefficients for the other three regions are insignificant. Additionally, my findings suggest that the effect of the EUA price on the HICP is 0.132 points larger for the poorest five countries in the sample relative to the richest five. Interestingly, the effect for the richest five nations is above the estimated overall effect, while the effect for the countries in the middle of the national income distribution is insignificant. Altogether, these findings could serve as an incentive for policymakers to focus on redistributing the burden of this policy both within and between the participant countries. A possibility could be to have a European entity that collects more detailed data to better identify the groups most negatively affected by this policy and redistributes the revenues obtained through auctioning permits.

An important limitation of this thesis is the small sample used. The data set is a small panel composed of 27 countries and 11 time periods, which implies a high variance for the estimated coefficients and, thus, makes it hard to find statistical significance. This constraint was dictated by the lack of publicly available quality data for the variables in the model. A suggestion for future research is to collect data on more time periods - possibly use monthly data - to see if this change widely affects the conclusions. Another shortcoming of this thesis is the quite high imperfect multicollinearity in the model. Similarly to the small sample size, this issue causes the variance of the estimators to be large. On the other hand, not including certain control variables in the model would have likely caused Omitted Variable Bias.

Future research could broaden the scope of this thesis to investigate the relationship of interest in other sectors. I studied the distributional consequences of the EU ETS in the heat and power sectors and, thus, I only estimated the effect of the carbon price on direct energy costs. Using

the same methodology to analyse whether a rise in the EUA price is associated with an increase in the price of emission-intensive goods and service - that is in indirect energy costs - could be quite interesting. Previous literature has estimated the latter effect to be much smaller than that on direct energy cost, yet it has mostly preferred scenario analysis and predictions over historical data.

This thesis only focused on the effect of the EU ETS on the price index, yet a complete research on the distributional consequences of this policy should include an analysis of the eventual revenue recycling programmes. Several previous studies in this field have found that the use of the revenues collected from a cap-and-trade system can considerably change the distributional implications of this policy, some designs might even make it progressive. Thus, future research could expand this thesis with an examination on whether the estimated increase in the price index actually translated in a larger burden of the EU ETS for poorer people relative to their income, or if this cost was redistributed through fiscal policy. Another relevant aspect in the evaluation of the consequences of climate policies is the distribution of climate mitigation benefits. Previous literature has suggested that the environmental benefits brought about by carbon pricing policies are larger for the people and countries at the lower end of the income distribution. Future work could try to assess whether this holds in the setting of this thesis and whether the environmental gains are of comparable size to the economic costs.

8

References

Andersson, F. N. (2018). Estimates of the inflation effect of a global carbon price on consumer, investment, export, and import prices. Lund University Working Paper, 2018:22. Baltagi, B. (2008). Econometric analysis of panel data. John Wiley & Sons.

Beznoska, M., Cludius, J., & Steiner, V. (2012). The incidence of the european union emissions trading system and the role of revenue recycling: Empirical evidence from combined industry-and household-level data.

Boyce, J. K. (2018). Carbon pricing: Effectiveness and equity. Ecological Economics, 150, 52–61. Callan, T., Lyons, S., Scott, S., Tol, R. S., & Verde, S. (2009). The distributional implications of

a carbon tax in ireland. Energy Policy, 37 (2), 407–412.

Caney, S., & Hepburn, C. (2011). Carbon trading: Unethical, unjust and ineffective? Royal Institute of Philosophy Supplements, 69, 201–234.

Castillo-Martinez, L., & Reis, R. (2019). How do central banks control inflation? a guide for the perplexed. Unpublished manuscript.

Competition Authority Estonia. (2019). Electricity and gas markets in estonia 2018. Retrieved April 29, 2020, fromhttps://www.konkurentsiamet.ee

Council of European Energy Regulators. (2019). Monitoring report on the performance of european retail markets in 2018. Retrieved April 27, 2020, from https://www.ceer.eu

Dinan, T. M., & Rogers, D. L. (2002). Distributional effects of carbon allowance trading: How government decisions determine winners and losers. National Tax Journal, 199–221. Energy Authority Finland. (2018). National report. Retrieved April 27, 2020, from https://

Euribor. (2020). What is euribor? Retrieved April 29, 2020, from https : / / www . euribor -rates.eu/en/what-is-euribor/

European Commission. (2014a). Cyprus country report. Retrieved April 27, 2020, from https: //ec.europa.eu/energy/sites/ener/files/documents/2014 countryreports cyprus.pdf

European Commission. (2014b). Greece country report. Retrieved April 27, 2020, from https: //ec.europa.eu/energy/sites/ener/files/documents/2014 countryreports greece.pdf

European Commission. (2017). Energy union factsheet cyprus. Retrieved April 27, 2020, from

https://ec.europa.eu/commission/sites/beta- political/files/energy- union- factsheet-cyprus en.pdf

European Commission. (2020). Liquefied natural gas. Retrieved June 12, 2020, from https : //ec.europa.eu/energy/topics/oil-gas-and-coal/liquefied-natural-gas-lng en

European Commission Climate Action. (2015). Eu ets handbook. Retrieved June 5, 2020, from

https://ec.europa.eu/clima/sites/clima/files/docs/ets handbook en.pdf

European Commission Climate Action. (2016). The eu emission trading system factsheet. Re-trieved June 10, 2020, from https://ec.europa.eu/clima/sites/clima/files/factsheet ets en. pdf

European Commission Climate Action. (2020). Revision for phase 4 (2021-2030). Retrieved June 9, 2020, from https://ec.europa.eu/clima/policies/ets/revision en

European Environment Agency. (2019). Heating and cooling degree days. Retrieved June 12, 2020, from https://www.eea.europa.eu/data-and-maps/indicators/heating-degree-days-2/assessment

European Environment Agency. (2020). Energy intensity in europe. Retrieved June 12, 2020, from https://www.eea.europa.eu/data- and- maps/indicators/total- primary- energy-intensity-4

European Securities and Markets Authority. (2020). Mifid ii. Retrieved June 11, 2020, from

https://www.esma.europa.eu/policy-rules/mifid-ii-and-mifir

Eurostat. (2015). What we do. Retrieved June 12, 2020, from https://ec.europa.eu/eurostat/ about/overview/what-we-do

Eurostat. (2019). Natural gas market indicators. Retrieved April 27, 2020, fromhttps://ec.europa. eu/eurostat/statistics-explained/pdfscache/8894.pdf

Eurostat. (2020). Eurostat database. Retrieved April 26, 2020, from https : / / ec . europa . eu / eurostat/data/database

Farber, D. A. (2012). Pollution markets and social equity: Analyzing the fairness of cap and trade. Ecology LQ, 39, 1.

Hintermann, B. (2010). Allowance price drivers in the first phase of the eu ets. Journal of Environmental Economics and Management, 59 (1), 43–56.

International Energy Agency. (2019). The role of gas in today’s energy transitions. Retrieved June 12, 2020, fromhttps://www.iea.org/reports/the-role-of-gas-in-todays-energy-transitions

Kaswan, A. (2009). Greening the grid and climate justice. Environmental Law, 1143–1160. Klenert, D., & Mattauch, L. (2016). How to make a carbon tax reform progressive: The role of

subsistence consumption. Economics Letters, 138, 100–103. Kumar, K. N. R. (2020). Econometrics. CRC Press.

Mazighi, A. E. H. (2005). Henry hub and national balancing point prices: What will be the international gas price reference? OPEC Review, 29 (3), 219–230.

Metcalf, G. E. (2005). Tax reform and environmental taxation. National Bureau of Economic Research.

Mishkin, F., Matthews, K., & Giuliodori, M. (2013). The economics of money, banking and financial markets: European edition. Pearson.

NASA. (2019). Climate change and global warming. Retrieved June 12, 2020, from https : //climate.nasa.gov/

Our World in Data. (2019). Coal prices, 2001 to 2018. Retrieved April 26, 2020, from https: //ourworldindata.org/grapher/coal-prices

Rausch, S., Metcalf, G. E., & Reilly, J. M. (2011). Distributional impacts of carbon pricing: A general equilibrium approach with micro-data for households. Energy economics, 33, S20–S33.

Regulatory Office for Network Industries Slovakia. (2020a). List of electricity suppliers providing universal service. Retrieved April 29, 2020, from http://www.urso.gov.sk

Regulatory Office for Network Industries Slovakia. (2020b). List of gas suppliers providing universal service. Retrieved April 29, 2020, from http://www.urso.gov.sk

Sager, L. (2019). The global consumer incidence of carbon pricing: Evidence from trade. Centre for Climate Change Economics and Policy Working Paper, 352.

Stock, J. H., & Watson, M. W. (2015). Introduction to econometrics. Pearson.

Tietenberg, T., & Lewis, L. (2013). Environmental and natural resource economics. Upper Saddle River, New Jersey, Pearson.

United Nation Statistical Division. (1999). Methodology (m49). Retrieved June 12, 2020, from

https://unstats.un.org/unsd/methodology/m49/

U.S. Energy Information Administration. (2020a). Europe brent spot price fob (dollars per barrel). Retrieved April 26, 2020, from https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx? n=pet&s=rbrte&f=a

U.S. Energy Information Administration. (2020b). Henry hub natural gas spot price. Retrieved April 26, 2020, fromhttps://www.eia.gov/dnav/ng/hist/rngwhhdA.htm

Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press. Wooldridge, J. M. (2012). Introductory econometrics: A modern approach. Nelson Education.

9

Appendix

9.1 Countries included in the study

Country Region Income Classification Austria West Richest 5 Belgium West

Bulgaria East Poorest 5 Cyprus South Middle 5 Czechia East Middle 5 Denmark North Richest 5 Estonia North

Finland North France West Germany West

Greece South Poorest 5 Hungary East

Ireland North Richest 5 Italy South Middle 5 Latvia North Poorest 5 Lithuania North

Luxembourg West Richest 5 Malta South Middle 5 Netherlands West Richest 5 Poland East Poorest 5 Portugal South

Romania East Poorest 5 Slovakia East

Slovenia South

Spain South Middle 5 Sweden North

United Kingdom North

9.2 Multicollinearity Diagnostics

Table6shows the Variation Inflated Factor (VIF) values for the Ordinary Least Square model, that does not control for unobserved country and time effects.

OLS VIF EUA 0.08 4.66 (0.57) competition -0.00∗ 1.06 (-2.18) intensity 0.01∗∗ 2.35 (3.11) Crudeoil 0.24∗∗∗ 2.67 (9.77) Coal 0.45 6.89 (1.06) gas -3.60∗∗∗ 9.68 (-3.56) GDP 0.03 1.83 (0.21) Unemployment -0.22 1.32 (-1.83) Euribor 84.60 7.59 (0.66) cooling 0.02∗∗∗ 2.45 (3.70) heating 0.00 3.07 (0.28) crisis -11.97∗∗∗ 5.09 (-4.38) cons 80.52∗∗∗ (5.39) N 297

Table 7 displays the Variation Inflated Factor (VIF) values for the Country Fixed Effects model. Country FE VIF EUA 0.22∗ 21.08 (2.38) competition 0.01 1.32 (1.12) intensity -0.03 21.43 (-1.06) Crudeoil 0.20∗∗∗ 31.54 (5.77) Coal 0.74 88.28 (1.82) gas -3.85∗∗∗ 58.10 (-5.88) GDP 0.18 52.61 (0.62) Unemployment 0.71 6.35 (1.66) Euribor 140.04∗ 9.97 (2.08) cooling 0.01 3.37 (0.53) heating 0.00 21.11 (1.01) crisis -11.79∗∗ 8.26 (-3.43) cons 71.66∗ (2.49) N 297

Table8 shows the Variation Inflated Factor (VIF) values for the Time and Country Fixed Effects model.

Time & Country FE VIF

EUA 0.31∗ 26.07 (2.14) competition 0.01 1.32 (0.89) intensity -0.02 21.57 (-0.71) Crudeoil 0.08 718.22 (1.08) Coal 2.17∗∗ 3596.58 (2.94) gas -4.51 965.91 (-1.91) GDP 0.03 150.90 (0.11) Unemployment 0.71 6.41 (1.65) Euribor 1299.96∗ 4350.41 (2.21) cooling -0.01 3.49 (-0.76) heating 0.00 21.61 (0.33) crisis -70.56∗ 17046.66 (-2.53) 2009.Year 24.01 658.34 (1.99) 2010.Year 50.13∗ 2537.06 (2.20) 2011.Year 42.12 1867.78 (2.02) 2012.Year 49.98∗ 1952.75 (2.23) cons 87.58∗∗ (2.82) N 297