THE EU ETS AND FIRM FINANCIAL

THE EU ETS AND FIRM FINANCIAL

PERFORMANCE: EVIDENCE FROM THE

EUROPEAN ELECTRICITY SECTOR

Master Thesis

Copenhagen Business School

Master of Sciences in Applied Economics and Finance

Authors: Magnus Poulsen (102115) and Pontus Lofgren (124590)

Supervisor: Lisbeth la Cour, PhD

May 15, 2020

101 pages and 147,238 characters

Abstract

Prior studies on the economic consequences of the European Union Emissions

Trading System (EU ETS) on European electricity generating firms have so far re-

lied on data from the first phase of the scheme that ended in 2007. We address

this gap and find that EU ETS emission allowance price increases (decreases) have

a positive (negative) impact on stock return by employing multifactor models with

a balanced longitudinal dataset covering thirteen European electricity generating

firms during the third phase of the EU ETS. Moreover, we find that the positive re-

lationship is stronger for firms with carbon efficient electricity generation. Based

on the Arbitrage Pricing Theory and the Efficient Market Hypothesis, we argue

that price appreciations in EU ETS emission allowances positively affect firm per-

formance for European electricity generating firms in general and carbon efficient

generators in particular. A possible explanation is that inframarginal suppliers ob-

tain regulatory rent due to the pass-through of the marginal producer’s additional

emission compliance cost to consumers. This suggests that the EU ETS is success-

ful in financially incentivizing profit maximizing firms concerned with electricity

generation to decarbonize operations, which may be of interest to policymakers

considering more stringent emission compliance costs for other sectors.

2

Acknowledgements

First and foremost, we express our gratitude and appreciation to our supervisor

Lisbeth la Cour, Professor in Time Series Econometrics at the Department of Eco-

nomics, Copenhagen Business School, for excellent guidance throughout the pro-

cess of writing this thesis. We are also grateful to Claus Vorm, Deputy Head of

Multi Assets at Nordea Investment Management, for providing important insights

into how utility stocks perform in low interest environments. We were also for-

tunate to sit down with representatives from Ørsted to learn about the electricity

sector. Lastly, we thank Julian Karst, Information Service Officer at the European

Energy Exchange (EEX) for kindly providing the data required for our empirical

analysis.

3

Contents

1 Introduction 10

1.1 Background and Problem Discussion . . . . . . . . . . . . . . . . . 10

1.2 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.4 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Institutional Background and Literature Review 14

2.1 The European Electricity Market . . . . . . . . . . . . . . . . . . . 14

2.1.1 Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.1.2 Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1.3 The European Electricity Mix . . . . . . . . . . . . . . . . . 17

2.1.4 Price-Setting and the Merit Order Curve . . . . . . . . . . . 20

2.2 The European Union Emissions Trading System . . . . . . . . . . . 21

2.2.1 Evolvement of the EU ETS . . . . . . . . . . . . . . . . . . . 22

2.2.2 Market Stability Reserve . . . . . . . . . . . . . . . . . . . . 28

2.3 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.3.1 European Union Allowances and Firm Financial Performancein the Electricity Sector . . . . . . . . . . . . . . . . . . . . 29

2.3.2 Pricing of European Union Allowances . . . . . . . . . . . . 33

3 Theory and Hypotheses 35

4

CONTENTS

3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.1.1 Arbitrage Pricing Theory . . . . . . . . . . . . . . . . . . . . 35

3.1.2 Efficient Market Hypothesis . . . . . . . . . . . . . . . . . . 37

3.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Methodology 41

4.1 Research Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.2.1 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.2.2 Carbon Intensity . . . . . . . . . . . . . . . . . . . . . . . . 44

4.2.3 European Union Allowance Prices . . . . . . . . . . . . . . . 45

4.2.4 Market Returns . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.2.5 Control Variables: Oil and Natural Gas . . . . . . . . . . . . 46

4.2.6 Transformation of Data and Frequency . . . . . . . . . . . . 46

4.3 Econometric Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.3.1 OLS Estimation . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.3.2 Omitted Variable Bias . . . . . . . . . . . . . . . . . . . . . 49

4.3.3 Fixed Effects Regression . . . . . . . . . . . . . . . . . . . . 50

4.3.4 Random Effects Regression . . . . . . . . . . . . . . . . . . 52

4.3.5 OLS Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . 52

4.3.6 Autocorrelation . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.3.7 Stationarity . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.3.8 Econometric Methodology . . . . . . . . . . . . . . . . . . . 56

5 Results 60

5.1 Overview of the Data . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5.2 Autocorrelation and Stationarity . . . . . . . . . . . . . . . . . . . 67

5

CONTENTS

5.3 Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5.3.1 Hypothesis I . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5.3.2 Hypothesis II . . . . . . . . . . . . . . . . . . . . . . . . . . 77

5.3.3 OLS Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . 82

5.4 Summary of Findings . . . . . . . . . . . . . . . . . . . . . . . . . . 85

6 Discussion 87

6.1 EU Allowance Price Changes’ Impact on Financial Performance . . 87

6.2 Carbon Intensity and Financial Performance . . . . . . . . . . . . . 93

6.3 Practical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . 95

6.4 Suggested Further Research . . . . . . . . . . . . . . . . . . . . . . 96

7 Conclusion 98

A Appendix 106

A.0.1 Autocorrelation and Stationarity . . . . . . . . . . . . . . . 106

A.0.2 OLS Diagnostic . . . . . . . . . . . . . . . . . . . . . . . . . 111

6

List of Figures

2.1 Illustration of Electricity Grid. . . . . . . . . . . . . . . . . . . . . . 15

2.2 Overview of Electricity Markets. . . . . . . . . . . . . . . . . . . . . 16

2.3 The European Electricity Mix . . . . . . . . . . . . . . . . . . . . . 18

2.4 Illustration of the European Carbon Intensity . . . . . . . . . . . . 19

2.5 Illustration of the Merit Order Curve . . . . . . . . . . . . . . . . . 20

3.1 Illustration of the Merit Order Curve with emission compliance costs 39

4.1 Carbon Intensity for sample firms . . . . . . . . . . . . . . . . . . . 44

5.1 Price Development of the variables under analysis . . . . . . . . . . 61

5.2 Price development of the Portfolio of Electricity Generating Firms,the Market Portfolio and Centrica plc . . . . . . . . . . . . . . . . . 65

5.3 Plots of the logarithmic weekly return and the corresponding ACFplot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.4 OLS Diagnostic - Pooled regression . . . . . . . . . . . . . . . . . . 82

5.5 OLS Diagnostic - Equally weighted portfolio . . . . . . . . . . . . . 83

5.6 OLS Diagnostic - Hypothesis II - Polluter . . . . . . . . . . . . . . . 85

6.1 Development for 10-year Government Yields for 2019 . . . . . . . 92

A.1 Plots of the logarithmic weekly return and the corresponding ACFplot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

A.2 OLS Diagnostic: One-year subperiod . . . . . . . . . . . . . . . . . 117

7

LIST OF FIGURES

A.3 OLS Diagnostic: Two-year subperiod . . . . . . . . . . . . . . . . . 120

A.4 OLS Diagnostic: Firm-specific . . . . . . . . . . . . . . . . . . . . . 133

8

List of Tables

2.1 Overview of the Four Phases of the EU ETS . . . . . . . . . . . . . . 27

4.1 Overview of the companies under analysis . . . . . . . . . . . . . . 43

5.1 Descriptive statistics for variables under analysis . . . . . . . . . . . 63

5.2 Correlation matrix of returns . . . . . . . . . . . . . . . . . . . . . 64

5.3 Correlation matrix for Centrica plc . . . . . . . . . . . . . . . . . . 66

5.4 Summary of Augmented Dickey-Fuller test . . . . . . . . . . . . . . 69

5.5 Regression Results: Hypothesis I . . . . . . . . . . . . . . . . . . . 73

5.6 Regression Results: One-year subperiod . . . . . . . . . . . . . . . 75

5.7 Correlation matrix of returns for 2019 . . . . . . . . . . . . . . . . 76

5.8 Regression Results: Two-year subperiod . . . . . . . . . . . . . . . 77

5.9 Regression Results: Hypothesis II . . . . . . . . . . . . . . . . . . . 80

5.10 Regression Results: Firm-specific . . . . . . . . . . . . . . . . . . . 81

5.11 Extreme data points for Pooled regression . . . . . . . . . . . . . . 84

9

Chapter 1: Introduction

1.1 Background and Problem Discussion

The European Emission Trading System (EU ETS) was introduced in 2005 as a

policy instrument for the European Union to comply with the carbon emission re-

duction targets set out by the Kyoto protocol. Under the system, emitting entities

must surrender one European Union Allowance (EUA) per metric ton of green-

house gas emissions by year-end or face hefty fines. It is a cap-and-trade system

in which the annual number of allowances is decreasing, and firms can trade al-

lowances with one another (Chevallier, 2012). This sets a price on emissions that

serves to incentivize companies to invest in technologies that reduce emissions.

It is cost-efficient in the sense that firms with high abatement costs can purchase

emission rights from firms with cheaper ways to reduce emissions. Today, the EU

ETS is the world’s largest market for emissions and covers more than 13,500 instal-

lations, which are responsible for roughly half of the European Union’s emissions

(Hintermann et al., 2016).

The electricity sector is the largest polluter in the European Union. It is also

different from other sectors in a variety of interesting ways. Firstly, it is naturally

protected from foreign competition by means of infrastructure and regulation.

Secondly, it has a wide range of abatement options available that can substitute

10

CHAPTER 1. INTRODUCTION

carbon-intensive electricity generation. Lastly, it is currently the only sector that is

not allocated any emission allowances for free (Chevallier, 2012).

Perhaps surprising for the unversed reader, previous research has found that elec-

tricity generating firms profited from increased prices of emission allowances dur-

ing the first phase of the EU ETS that ran from 2005 to 2007, despite it constituting

an increased input cost. The phenomenon can be explained by the free allocation

method of emission allowances that was present in the first phase, combined with

a high pass-through rate of its opportunity cost to consumers (Sijm et al., 2006).

In effect, this resulted in windfall profits to the European electricity sector (Obern-

dorfer, 2009; Veith et al., 2009).

Much has changed in the last fifteen years. In terms of the regulatory nature of the

European Union Emissions Trading System, the electricity sector is now subject to

full auctioning of emission allowances, and the system itself has matured signif-

icantly (European Commission, 2015). In terms of the European electricity mix,

renewable energy production has penetrated the market, and carbon intensity has

roughly halved. As such, the results of the previous research may not hold today

and with this paper, we aim to address that gap.

1.2 Purpose

The purpose of the paper is to understand how price changes in European Union

Allowances affect financial performance for European electricity generating firms

during the third phase of the European Union Emissions Trading System. The

conclusions are relevant as it may provide important insights to European poli-

cymakers’ assessment of how to further develop the European Union Emissions

Trading System or for foreign policymakers considering implementing similar reg-

11


ulation. Further, it may be relevant to investors and electricity generating firms

seeking to maximize return on their investments.

1.3 Research Question

This paper seeks to deepen our understanding of how the European Union Emis-

sions Trading System’s greenhouse gas emission allowances affect firm financial

performance for European electricity generating firms during its third phase. We

define our overarching research question as follows:

How do price changes in European Union Allowances affect firm performance for

European electricity generating firms during the third phase of the EU ETS?

It will further nuance our understanding by investigating if and how these dynam-

ics change depending on the carbon intensity of the firm’s electricity production

by answering the following research question:

How does a potential impact of European Union Allowance price changes differ de-

pending on the carbon intensity of European electricity generating firms’ electricity

production during the third phase of EU ETS?

1.4 Thesis Structure

The remainder of the paper is structured as follows. Chapter II introduces the

reader to the electricity sector and the European Union Emissions Trading Sys-

tem to provide context for our overarching research question. To ensure sufficient

theoretical depth, it includes a literature review on the research of the interplay

between emission allowance price changes and firm performance for European

12


electricity generating firms and the price-setting forces on emission allowances.

Chapter III introduces the Arbitrage Pricing Theory and the Efficient Market Hy-

pothesis. Combined with previous research on the subject, these theories will lay

the foundation to our hypotheses. Chapter IV describes the data and the method-

ology. Chapter V will presents the empirical results discussed in Chapter VI, pre-

ceding the conclusion in Chapter VII.

13

Chapter 2: Institutional Background

and Literature Review

Chapter II begins by providing the reader with an introduction to the European

electricity sector including its infrastructure, markets, electricity mix, and price

setting mechanisms. Next, it presents an overview of the European Union Emis-

sions Trading System before providing a review of the literature covering its effect

the financial performance of European electricity generating firms as well as the

price setting forces on emission allowances.

2.1 The European Electricity Market

2.1.1 Infrastructure

The European Union has adopted a wide range of legislation aiming at achieving

the long-term goal of an integrated European energy market, in which electric-

ity is traded cross-borders competitively. Previously, each domestic energy market

was monopolistic or oligopolistic and consisted of one or a few vertically inte-

grated companies responsible for the generation, transportation, and distribution

of electricity. Today, regulation limiting the degree of vertical integration has

14

CHAPTER 2. INSTITUTIONAL BACKGROUND AND LITERATURE REVIEW

further divided the electrical sector into three participant groups: i) electricity

generators, ii) transmission system operators (TSO), and iii) distribution system

operators (DSO). The electricity grid is a network that connects electricity gen-

erators and consumers via transmission and distribution networks. Transmission

system operators are responsible for transporting electricity regionally over long

distances from power generators to distribution system operators that locally dis-

tribute electricity to households and industries. The European transmission grid

covers 300,000 km of power lines, including 355 cross-border lines. (Erbach,

2016).

Source: Erbach (2016)

Figure 2.1: Illustration of Electricity Grid.

2.1.2 Markets

Participants in the wholesale markets are electricity generators on the one side

and electricity suppliers and large industrial consumers on the other. They meet

on organized multilateral power exchanges such as Nordpool or over-the-counter.

Electricity is a unique commodity in the sense that it is essentially produced and

consumed simultaneously and that imbalances between production and consump-

tion in the grid can cause system collapse and blackouts. As such, the electricity

markets have been designed to accommodate this issue. The markets are ordered

15


sequentially starting years before the actual delivery and ending after the actual

delivery. This facilitates the balancing between supply and demand.

Source: Erbach (2016)

Figure 2.2: Overview of Electricity Markets.

In the forward and futures market, the two parties agree to the terms of the ex-

change years to weeks in advance of the transaction taking place. This allows gen-

erators to plan their output appropriately and for both parties to hedge against the

risk of fluctuating electricity prices. In the day-ahead market, electricity is traded

one day before delivery allowing electricity generators to adjust their capacity to

accommodate demand the following day. In the intra-day market, electricity is

traded on the day of delivery, and output can be fine-tuned. Lastly, in the balanc-

ing market, the Transmission System Operator will maintain system balance by

injections or take-offs of electricity (KU Leuven Energy Institute, 2015).

16


2.1.3 The European Electricity Mix

Electricity Generation Technologies

Electricity generation technologies are typically divided into three different cate-

gories: nuclear power, fossil-fueled power, and renewable electricity.

The process of generating electricity using nuclear power is thermal, in which

water is heated through nuclear fission to steam that drives a rotational turbine.

Nuclear power requires substantial initial investments, but the electricity is gener-

ated at low marginal cost and with no greenhouse gas emissions.

As with nuclear power generation, fossil-fuel power is generated through a ther-

mal process in which fuel is burned to create steam that drives a turbine. Fossil-

fuel generation can be divided into three subcategories, namely coal, natural gas,

and petroleum. Coal and natural gas fossil-fuel generation have efficiencies of

40% and 60%, respectively, meaning that a large portion of the energy is not being

transformed into electricity. These power plants are commonly operated using an

Integrated Gasification Combined Cycle procedure that allows for the use of both

fuel types as inputs. This allows the power plant to switch between fuel types de-

pending on the relative price of the input and the associated emission costs. Coal

emits, on average, approximately 2.5 times more carbon into the atmosphere than

natural gas for the same output of electricity.

Electricity generated from renewable energy sources are categorized into i) wind

power, ii) hydroelectric power, iii) biomass-based electricity, iv) photovoltaic elec-

tricity, and v) concentrating solar power.

Wind power generates electricity by exploiting wind to mechanically rotate a wind

turbine. Technological advancements have significantly increased efficiencies and

17


lowered the average cost of electricity output. Hydroelectric power exploits run-

ning water, typically from a reservoir, to drive a turbine that generates electricity.

It is a relatively stable source of electricity and can be constructed with very high

capacity. Hydroelectric power has been the primary source of renewable electric-

ity production in the European Union until recently when it ranks second to wind

power. Biomass-based electricity burns biomass, primarily wood, to create steam

that drives a turbine. Both photovoltaic and concentrating solar power use the

energy from the sun to create electricity (Rademaekers et al., 2011).

Development of the European Electricity Mix

Source: Jones (2020)

Figure 2.3: The European Electricity Mix

18


The European energy mix has undergone a significant shift towards less carbon-

intensive electricity generation. Since 2007, which was the last year of the first

phase of the EU ETS, electricity generation from coal has more than halved and

been entirely replaced by renewable energy sources, primarily wind power. In

2019, wind and solar power were responsible for a larger proportion of electric-

ity production than coal for the first time, and greenhouse gas emissions in the

electricity sector were 43% lower compared to the levels in 2007. Total electricity

generation has not changed considerably, with 3,312 TWh in 2005 and 3,227 TWh

in 2019.

Source: Jones (2020)

Figure 2.4: Illustration of the European Carbon Intensity

This transformation has been driven by multiple forces. An increased awareness

regarding global warming has increased demand for renewable energy sources,

with most electricity vendors offering “green energy” to its customers. Combined

with significant technological advancements in wind power that increase capacity

and efficiencies, investments in wind power increased substantially, often sup-

ported by government subsidies. In addition, the EU ETS’ price on emissions

has incentivized companies to invest in renewable energy production (Adabie and

19


Chamorro, 2008). Moreover, within the fossil-fuel generation, there has been a

large shift from coal to natural gas, driven by the increased cost of emissions. In

2019, for instance, a sharp increase in emission allowance prices partially drove

a significant fuel switch in thermal electricity generation from coal to natural gas

(Buck et al., 2020).

2.1.4 Price-Setting and the Merit Order Curve

The spot price of electricity in the wholesale market and short-term deployment of

power plants is typically determined in the day-ahead market on organized mul-

tilateral exchanges (Cludius et al., 2014). Electricity generators offer their bids

at, in theory, short-term marginal costs that consist of the production technology’s

cost of fuel and greenhouse gas emissions. By ranking the available sources of

electricity in ascending order of marginal cost together with the associated aggre-

gate output, we obtain the so-called merit order curve, which reflects a supply

curve. Under efficient market conditions, the price is determined by the power

plant that clears the market, given the demand at any given time.

Source: Cludius et al. (2014)

Figure 2.5: Illustration of the Merit Order Curve

For instance, in Figure 2.5, the market demands a certain amount of electricity,

20


which makes coal clear the market. Coal is then the price-setting technology and

the marginal producer. Because renewable energy sources and nuclear power

plants have lower marginal costs, they are infra-marginal producers with positive

contribution margins.

2.2 The European Union Emissions Trading System

The first indication of a European emission trading system appeared in 2000 when

the European Commission issued Green Paper on Greenhouse Gas Emissions Trading

within the European Union (Denny Ellerman et al., 2016). The EU ETS Directive

(Directive 2003/87/EC) was adopted in 2003 and subsequently launched in 2005

across all 27 member states as a policy instrument to ensure that the European

Union would meet its legally obligated emission reduction target of the 1997 Kyoto

protocol (Chevallier, 2012).

The EU ETS is a “cap and trade” system in which a cap is set on the total amount of

greenhouse gas emissions that can be emitted during a year. Each company must

surrender enough allowances to cover its installations’ emissions by year-end, or

hefty fines will be imposed (Chevallier, 2012). Because the cap is reduced over

time, the total emissions of the European Union are reduced.

Within the cap, companies can meet on the market and trade allowances. As

such, a price on greenhouse gas emissions is formed, which serves to incentivize

profit-maximizing companies to invest in green technologies that reduce emissions

(Hintermann et al., 2016). The companies that are able to reduce their greenhouse

emissions at a lower cost than the price of carbon will be financially incentivized

to do so. Under perfect market conditions in equilibrium, the marginal abatement

cost will equal the price of carbon and will be identical for all companies. As such,

21


investments will be concentrated at the most cost-effective abatement opportuni-

ties, which leads to a reduction in emissions at the lowest economic cost.

As of 2014, more than 13,500 energy-intensive installations responsible for ap-

proximately four percent of global greenhouse gas emissions are covered by the

system (Hintermann et al., 2016) making it the largest market for greenhouse gas

emissions (European Commission, 2015). Since its inception in 2005, the system

has been successfully implemented during three distinct phases, with the fourth

phase set to begin in 2021.

2.2.1 Evolvement of the EU ETS

Phase I: 2005 - 2007

The first phase of the EU ETS served as a “warm up phase” aimed at establishing

a price on carbon dioxide emissions, free trade in emission allowances, and the

infrastructure to monitor, report and verify emissions from installations covered

by the system (European Commission, 2013). In addition to power stations and

other combustion plants with an output above 20 MW, oil refineries, coke ovens,

iron and steel plants, cement clinkers, glass, lime, bricks, ceramics, pulp, as well as

the paper and board sectors were included in the system (European Commission,

2015). Together they covered 46% of the total CO2 emissions of EU countries with

the electricity sector being the largest emitter (Oberndorfer, 2009). Nine out of ten

allowances were allocated free-of-charge to limit the risk of adversely impacting

European competitiveness on international markets, which would result in carbon

spillover to countries outside the EU. A penalty charge to companies that failed

to sufficiently surrender allowances was set to e40 per ton of CO2. The price of

carbon dioxide reached e8 per ton in the first month (Chevallier, 2012).

22


Each covered country established its own “National Allocation Plan” that deter-

mined the amount of allowances that would be available for the country and how

they would be distributed between sectors (European Commission, 2015). The

National Allocation Plans had to comply with the criteria set out by the European

Commission in the ETS Directive to be approved. As such, the total EU cap con-

sisted of each member state’s respective national caps and would not be known

until all National Allocation Plans had been approved (Denny Ellerman et al.,

2016).

The EU had not yet gathered reliable emissions data, and the initial cap was set

based on estimates. In 2007, when the EU disclosed the emissions data compiled

from the monitoring activities, it became clear that the EU had vastly overallocated

allowances available in the market (Chevallier, 2012). At the time, an allowance

had limited temporal value as it could not be banked to the second phase that

would soon follow. This led to high volatility and spot prices plummeting dramat-

ically to a value of zero from e25-30 just months before (Aatola et al., 2013).

Phase II: 2008 - 2012

Phase II extended the EU ETS’ scope in sectors to include domestic aviation, in

geography to include Norway, Iceland, and Lichtenstein, and in emission types

by voluntary opt-in of nitrous oxide (N2O) (European Commission, 2015). The

system was linked to international carbon dioxide markets by accepting certain

Kyoto Protocol emission units under the Clean Development Mechanism (CDM)

and Joint Implementation (JI) alongside European Union Emission Allowances

(EUAs) (European Commission, 2015). These arrangements allow industrialized

countries to reduce their greenhouse gas emissions through international invest-

ments that reduce emissions in developing countries. The Kyoto credits that are

generated through CDM and JI projects are interchangeable with EUAs and can

23


thus be surrendered to offset greenhouse gas emissions subject to the regulation

(Denny Ellerman et al., 2016). The cap for these offsets during the second phase

was set to 1.3 billion (total cap in 2008 was 1.9 billion), but investments in large

hydroelectric power and forestry projects were not eligible. Although EU ETS was

the largest market for carbon dioxide and the primary driver for demand of Kyoto

credits, the use of offsets only constituted 10% of the cap despite being a perfect

substitute for EUAs (Denny Ellerman et al., 2016).

Trading volumes increased, and price volatility decreased during the second phase

due to a tighter emissions cap and because allowances were allowed to be banked

over the years (Aatola et al., 2013). Research finds that the more stringent emis-

sions cap significantly reduced emissions during the second phase (Martin & Wag-

ner, 2016).

Phase III: 2013 - 2020

The first commitment period of the Kyoto protocol ended in 2012, and it was

not followed by the Doha Amendment. In the absence of an international agree-

ment on greenhouse gas emission reduction, the third phase corresponds to the

objective of the EU’s Energy Climate Package introduced in 2008 with the goal of

reducing emissions by 20% by 2020 compared to 1990 (Chevallier, 2012).

The scope was extended in geography to include Croatia (that joined the EU in

2013), and in sectors by including the aluminium and petrochemical sectors and

in types by including N2O emissions from all installations and PFC from aluminium

production (European Commission, 2015).

A pan-EU cap for stationary installations declining by a linear factor of 1.74% com-

pared to the average total quantity of allowances issued annually in 2008 - 2012

was adopted. This corresponds to a decrease in absolute terms of 38,264,246 al-

24


lowances per year and would ensure a decrease in emissions from EU ETS covered

sectors by 21% compared to levels in 2005 (European Commission, 2015).

Although auctioning became the default allocation method in Phase III, a signif-

icant amount of allowances are still being issued for free. The power generation

sector, however, has been subject to full-auctioning since 2013 except for instal-

lations in certain countries covered by Article 10c of the EU ETS Directive. It

provided derogation from the default auctioning allocation method until 2019 to

support modernization investments that reduced the reliance on coal to generate

electricity. At the minimum, investments to diversify the energy mix and mod-

ernize their respective electricity sectors had to amount to the value of the free

allowances that were issued. Of ten eligible countries, eight states made use of the

exception: Bulgaria, Cyprus, Czech Republic, Estonia, Hungary, Lithuania, Poland

and Romania. (European Commission, 2015). The power generation sector was

considered suitable for full-auctioning due to its relatively low abatement costs to

reduce emissions and the weak exposure to international competition from firms

not covered by the EU ETS (Chevallier, 2012).

Phase IV: 2021 - 2030

The fourth phase of the EU ETS will begin from 2021 and run until 2030. To

increase the pace of the decrease in emissions, the EU will increase the annual

linear reduction rate of emissions to 2.2%, compared to 1.74% in phases III. The

purpose is to achieve a reduction in greenhouse gas emissions for sectors in the

EU ETS by 46% compared to the 2005 level. Also, the Market Stability Reserve

play a larger role in phase IV as the number of allowances put in the reserve

will temporarily double to 24% of the total number of allowances in circulation

from 2019 to 2023, after which the rate will return to 12% in 2024. The Market

Stability Reserve will be described in further detail in Section 2.2.2 but aims to

25


reduce the current overallocation of the total number of allowances in circulation.

The fourth phase will address the risk of carbon leakage by providing predictable,

robust, and a fair set of rules. The EU defines carbon leakage as an increase in

greenhouse gas emissions in one country as a product of strict climate policy in

another. The system of free allocation will be continue for another decade but will

focus on sectors at the highest risk of relocating their operations outside of the EU.

These sectors will continue receiving allowances for free, but free allocation will be

phased out to less exposed sectors by the end of phase IV. (European Commission,

2019b).

Additionally, as part of the fourth phase, there will be an increase in ”green” in-

vestments to modernize the system. The EU will set up several low-carbon fund-

ing mechanisms to help energy-intensive industrial sectors and the power sector

to meet the innovation and investment challenges facing the transition to a low

carbon economy. The two main new initiatives are an innovation fund and a mod-

ernization fund. The innovation fund will support the innovative technologies in

the industry and will be an extension of an existing fund and will correspond to a

market value of at least 450 million emission allowances. The modernization fund

will support investments in modernizing energy systems, energy efficiency, and fa-

cilitating the transition in carbon dependent regions in 10 lower-income member

states (European Commission, 2019b).

26


Tabl

e2.

1:O

verv

iew

ofth

eFo

urPh

ases

ofth

eEU

ETS

Key

Feat

ure

sPh

ase

IPh

ase

IIPh

ase

III

Phas

eIV

Geo

grap

hyEU

27EU

27-N

orw

ay,

Lich

tens

tein

and

Icel

and

EU27

-Nor

way

,Li

chte

nste

in,

Icel

and

and

Cro

atia

EU27

-Nor

way

,Li

chte

nste

in,

Icel

and

and

Cro

atia

Sect

ors

Pow

erst

atio

nsan

dot

her

com

bust

ion

plan

ts>

20M

W,

Oil

refin

erie

s,C

oke

oven

s,Ir

onan

dst

eelp

lant

s,C

emen

tcl

inke

r,G

lass

Lim

eB

rick

s,C

eram

ics,

Pulp

Pape

ran

dbo

ard.

Sam

eas

Phas

eI

plus

Avi

atio

nfr

om20

12

Sam

eas

Phas

eII

plus

Alu

min

ium

and

Petr

oche

mic

als

Sam

eas

Phas

eII

I

Gre

enho

use

gase

sC

O2

CO

2,N

2O

and

PFC

from

Alu

min

ium

proc

essi

ng

CO

2,N

2O

and

PFC

from

Alu

min

ium

proc

essi

ng

CO

2,N

2O

and

PFC

from

Alu

min

ium

proc

essi

ng

Cap

size

2,05

8M

T1,

858

MT

2,08

4M

Tde

crea

sing

by38

MT

per

year

(exc

ludi

ngM

SR)

Unk

now

n,bu

tth

eca

pw

illde

crea

seby

2.2%

per

year

(exc

ludi

ngM

SR)

Elig

ible

trad

ing

unit

sEU

AEU

A,C

ER,E

RU

EUA

,CER

,ER

UEU

A,C

ER,E

RU

The

tabl

ere

port

sth

eke

yfe

atur

esof

the

four

phas

esof

the

EUET

S.It

focu

ses

onG

eogr

aphy

,Sec

tors

,Gre

enho

use

gase

s,ca

psi

zean

del

igib

letr

adin

gun

its.

MT

isan

abbr

evia

tion

for

mill

ion

met

ric

tonn

es.

27


2.2.2 Market Stability Reserve

The EU ETS faced widespread criticism that the price of allowances was too low

to incentivize companies to reduce emissions (Perino & Willner, 2016). A sur-

plus of allowances, defined as the difference between the available amount of

allowances and the amount of allowances required for compliance in a given year,

had been built up during the second phase as a consequence of the financial crisis,

high imports of international credits and faster-than-expected transition towards

renewable electricity generation. The surplus was two billion allowances at the

start of phase III in 2013 and the cost of emitting a ton of CO2 was roughly equiv-

alent to a cup of coffee at e3. In the short-term, this risked undermining the

proper functioning of the emissions market and in the long-term, and it risked

adversely affecting the ability of the system to reduce emissions. As such, the EU

implemented a short-term and a long-term measure to mitigate these risks.

In 2015, the surplus was reduced to 1.78 billion allowances through what the

EU calls “back-loading”. Essentially, back-loading was the short-term measure of

postponing the auctioning of 900 million allowances to 2019 and 2020. It would

not change the overall amount of allowances available during the third phase but

served to redistribute the auctions in order to shift the short-term equilibrium to

achieve a higher price. 400 million, 300 million and 200 million allowances were

postponed during 2014, 2015, and 2016, respectively. As a result, the surplus was

reduced by an estimated 40% at the end of 2015.

As a long-term measure, the Market Stability Reserve took effect in January 2019.

It adjusts the number of allowances that are auctioned each year based on the

previous years’ surplus. In times of large surplus, allowances amounting to 12%

of the aggregate market surplus will be set aside, and their issuing shifted into

the future. As the surplus drops below 400 million, the allowances will be issued

28


to the market in intervals of 100 million per year (European Commission, 2013).

This process will continue until the surplus is depleted. The reserve operates com-

pletely based on a set of predefined rules and leaves no room for discretion to

the EU or member states. The aforementioned reintroduction of the back-loaded

allowances during 2014 to 2016 was cancelled and put in the reserve instead.

2.3 Literature Review

2.3.1 European Union Allowances and Firm Financial Perfor-

mance in the Electricity Sector

Even before the system was implemented in 2005, simulations suggested that the

EU ETS would benefit the electricity sector as a whole, as long as allowances were

distributed free of charge (Bode, 2006). A price on emissions constitutes an oppor-

tunity cost as an allowance can either be used as an input factor to cover emissions

or be sold to another firm that demands it, regardless of whether it was provided

for free or not. As such, the marginal production cost increases for companies

subject to the regulation, and profit-maximizing firms will attempt to transfer this

cost to its customers. In doing so, companies face the risk of losing market share to

competition not covered by the EU ETS. However, the electricity sector in the EU

has inherent entry barriers from foreign competition due to technical aspects that

limit the possible import penetration of electricity (Marin et al., 2018). As such,

electricity companies can transfer much of the opportunity cost that the price of

an allowance represents to its customers.

Sijm et al. (2006) combined spot and forward market prices of electricity in the

Netherlands and Germany with the price of EUA to investigate the pass-through

29


rate of the emission compliance costs to wholesale electricity prices. They find

that the pass-through rate, defined as the average increase in power price over a

certain period due to the increase in the price of an emission allowance, varied be-

tween 60 to 100% depending on the carbon intensity of the marginal production

technology. As an example, consider a typical coal power station that emits 0.95

ton of CO2 per MWh and compare it with a gas power station that emits 0.48 ton

of CO2 per MWh. The pass-through rate of the emission allowances on the whole-

sale electricity price is determined by which of the two power stations that are

currently setting the price. At a price of e20 per ton of emissions, the generation

cost per MWh for the gas plant increases by e9.6 and by e17 for the coal plant.

Consequently, the pass-through rate is higher when coal is the marginal producer.

Because allowances were allocated to electricity companies free of charge during

the first phase, electricity companies were able to reap large profits. Using numer-

ical models, Sijm et al. (2006) estimate that EU ETS induced windfall profits in

the Dutch power sector amounted to e300-600 million per year.

Because stock prices theoretically reflect firms’ future discounted cash flows, economists

often use stock-market data to estimate the impact on profits from policies (Marin

et al., 2018). Oberndorfer (2009) found that stock prices of EU electricity compa-

nies reacted positively to appreciations in EU ETS prices and negatively to depre-

ciation during the first phase of trading. Oberndorfer employed panel regressions

using an aggregated equally weighted portfolio of the most important electric-

ity stocks in the Eurozone as well as disaggregated pooled panel data regression.

The market returns, as well as oil, natural gas and electricity price returns, were

included as control variables to the regressions. The author stresses that this

is important not only because of their impact as input factors but also due to

their influence on the emission allowance price itself. The factor-beta coefficients

for emission allowances are positive at 0.02 and 0.01 for the aggregate equally

weighted portfolio and the pooled OLS, respectively, and significant with a p-value

30


below 1%. Oberndorfer argues that the positive influence is due to windfall prof-

its that occur as a result of the high pass-through rate of the opportunity cost of

the grandfathered emission allowances. Oberndorfer (2009) investigates the ef-

fect on a country-specific level and finds that the results hold true for all countries

investigated except Spain. Oberndorfer (2009) argues that Spain’s stringent price

regulation that limits the pass-through rate could be the reason for the inverse

effect. Zachmann and von Hirschhausen (2008) found that positive price shocks

in emission allowances disseminated more quickly to the final German wholesale

electricity price than negative price shocks and suggest that this price asymme-

try may be due to market power or ignorance amongst market agents due to the

immaturity of the market. This could result in increased profits but Oberndor-

fer (2009) did not find that such an asymmetry existed in stock price reactions,

indicating that market participants were either unaware of the asymmetric price

reactions in the wholesale electricity prices or that it was a German phenomenon.

Similar to Oberndorfer (2009), Veith et al. (2009) measure the EU ETS impact on

firm performance through investor expectations by employing a multifactor model

on a sample of 22 publicly traded electricity firms accounting for two-thirds of total

EU electricity generation. In line with Oberndorfer (2009), they find a positive re-

lationship between emission allowance futures prices and stock prices, indicating

financial market agents expect that electricity companies obtain higher earnings

by passing on the opportunity cost of the emission allowances to wholesale elec-

tricity prices. The authors fit the emission allowance return into their subperiod

pooled panel regressions and estimate that carbon prices positively yielded an es-

timated average stock market return by 0.8% during the first six months of the

trading system, suggesting that electricity generators may obtain regulatory rent.

As a robustness check, they again regress stock performance on overall stock per-

formance and emission allowance returns for companies that have carbon-neutral

electricity generation and thus fall outside of the system. Although these firms

31


sell the same common good as firms covered by the system, the empirical results

do not show a connection between emission allowance return and stock perfor-

mance. The authors present a lack of market power as a possible explanation, or

that investors in these entities do not consider emission price movements. In addi-

tion to regressing return on emission allowance returns and overall stock market

return, they control for commodity price impact by including the return on oil and

natural gas to a fixed effects regression model. The results regarding emission

allowance futures return remain robust with a factor-beta coefficient of 0.023 and

significant at 5%. Next, they further check for the robustness by controlling for

firm-specific characteristics in terms of fuel mix by including a dummy variable

equaling one if the company’s proportion of carbon-emitting electricity generation

is above the sample median. The binary variable is interacted with the return on

emission allowance prices to allow for both different intercept and slope. There is

no interaction effect when regressing emission allowance spot prices, but a slightly

negative coefficient when using emission allowance December 2008 futures. The

firms with a high share of fossil fuels in electricity generation lose approximately

half of the positive influence of emission allowance returns. According to the au-

thors, this indicates that investors did not consider the underlying fuel mix during

the first phase of the EU ETS that would end in December 2007 but that investors

expected future cash flows to be affected during the second phase. They argue

that investors likely anticipated and discounted adverse economic effects resulting

from a higher degree of auctioning that would be enforced during the subsequent

phase.

Bushnell et al. (2013) conducted an event study on how the large price decline in

emission allowances in April 2006 affected the stock prices of 552 European com-

panies subject to EU ETS in different sectors. They find that share prices within the

power sector decline, and that “clean” electricity companies’ share prices decline

further than their “dirty” counterparts. The authors argue that this implies that

32


the market understood that declining emission prices would reduce contribution

margins more severely for less carbon-intensive power generators. High emitting

electricity companies also experienced abnormal declines in share price, and the

authors suggest that this highlights a focus on revenue rather than costs among

investors.

To our knowledge, there has been no published research on the potential effects

of EUA price changes on firm performance among European electricity firms dur-

ing the second or third phase of the EU ETS. All the abovementioned studies lay

forward that their results may not hold true during later phases of the EU ETS,

primarily due to the end of grandfathering (Sijm et al., 2006; Oberndorfer, 2009;

Veith et al., 2009; Bushnell et al., 2013) and that the market becomes more effi-

cient as market agents become better informed (Oberndorfer, 2009; Bushnell et

al., 2013).

2.3.2 Pricing of European Union Allowances

The price of a ton of greenhouse gas emission in the EU ETS is determined by

the equilibrium between supply and demand in the market. Supply is primarily

determined by EU policy decision such as the size of the emissions cap, allocation

methods, linkages to international greenhouse gas emission markets and rules

about banking and borrowing of allowances. An example of the impact of supply

on EUA prices is the crash in 2006 when the markets realized that the EU had

oversupplied participants with allowances (Alberola et al., 2008). Hintermann

et al. (2016) review the literature of pricing during the second phase of the EU

ETS and find that demand is primarily determined by the amount of “business as

usual” emissions, which are mainly driven by the growth of the economy, its energy

efficiency and emission intensity. During times of economic growth, emissions

33


increase together with industrial production, which drives demand for allowances

resulting in price appreciations (Chevallier, 2012). Even unanticipated extreme

weather conditions drive demand for EUAs. For instance, during exceptionally

hot summer months or freezing winter months, households’ increased cooling and

heating have a short-term impact on business-as-usual emissions (Alberola et al.,

2008).

Moreover, in the Nordic region, low-marginal cost and renewable hydroelectric

power is responsible for baseload generation and constitutes more than half of

total electricity output in Sweden and Norway. As such, when water reservoir lev-

els are low, more carbon-intensive technologies must replace part of hydroelectric

power’s output leading to an increased demand for emission rights (Rickels et al.,

2012). Prices are also determined by the available abatement options and their

respective costs (Hintermann et al., 2016). As an example, coal and gas power

plants are commonly able to switch between the fuel types. Because natural gas

emits at least half as much as coal, a sector-wide switch between the two fuel types

can have an effect on the price of EUAs (Aatola et al., 2013).

34

Chapter 3: Theory and Hypotheses

The following chapter provides an explanation of the theory used to answer the

research question at hand. We will begin by explaining the Arbitrage Pricing The-

ory and then the Efficient Market Hypothesis and lastly turn to how these theories

are implemented to form testable hypotheses.

3.1 Theory

3.1.1 Arbitrage Pricing Theory

The Arbitrage Pricing Theory (APT) was developed by Ross (1976) as a multi-

factor asset pricing model that asserts a security’s expected return as a linear func-

tion of its relationship to various macroeconomic factors. The return of a risky

asset is determined by a set of common factors and an idiosyncratic risk compo-

nent:

E[ri] = α + βi,k · Fk + βi,k+1 · Fk+1 + εi (3.1)

where E[εi] = 0 by construction, whereas Cov[Fk, εi] = 0 and Cov[εi, εj] = 0 for

i 6= j by assumption. Multifactor models are tools that allow us to investigate the

ultimate source of risk and are useful to measure a risky asset’s exposure to certain

sources of uncertainty. As an example, Chen et al. (1986) identified the following

35

CHAPTER 3. THEORY AND HYPOTHESES

macroeconomic factors as significant in explaining security returns:

• Surprises in inflation

• Surprises in GDP as indicated by industrial production index

• Surprises in investor confidence due to changes in default premium in cor-

porate bonds

• Surprise shifts in the yield curve

The equation above forms a multidimensional security characteristic line (SCL)

that consists of a set of factors, whose beta-values are estimated using multivariate

regression analysis for a particular security (Bodie et al., 2011). Each beta-value

represents the security’s sensitivity to the respective factor’s systematic risk. As

such, an unexpected increase of one unit in a factor is associated with a change in

the expected return of the security equivalent to the beta-value (Berk & DeMarzo,

2017). The residual variance of the regression represents an idiosyncratic risk.

Although the arbitrage theory of capital asset pricing was developed as an alter-

native to the Capital Asset Pricing Model developed by Sharpe (1964), Lintner

(1965) and Treynor (1962), it does not assume that markets are perfectly effi-

cient Ross (1976). Hence, the market may occasionally misprice securities until

arbitrageurs exploit the anomaly, which adjusts the price to its fair value accord-

ing to the Law of One Price. As such, the APT does not assume that all investors

hold identical portfolios, namely the market portfolio identified through a mean-

variance model. Instead, investors hold highly diversified but different portfolios

that essentially have no idiosyncratic risk. In applications, however, it is assumed

that there is no idiosyncratic risk and that the equation above can explain the

expected return of a security.

36


3.1.2 Efficient Market Hypothesis

The Efficient Market Hypothesis was formalized by Fama (1970) and assumes

that security prices at any time fully reflect all available information. A market in

which prices always fully reflect all available information is efficient. There are

three levels of efficiency:

• The weak-form hypothesis states that security prices reflect all historical

trading data such as past returns, trading volumes and interest rates. This

implies that trend analysis is fruitless.

• The semi strong-form hypothesis asserts that all publicly available informa-

tion is reflected in the price of the security. Such information includes news

regarding product lines, management etc. in addition to historical trading

data.

• The strong-form version states that all information, including inside infor-

mation, is reflected in the stock price.

The dividend discount model states that the stock price is determined by the dis-

counted value of all future dividends (Berk & DeMarzo, 2017):

Pt =∞∑j=1

Dt+j

(1 + rt+j)(3.2)

Hence, any price change must be due to changes in expected future dividends or

the discount rate. Assuming markets are indeed efficient, firm performance can be

gauged through the stock price (Veith et al., 2009; Oberndorfer et al., 2012).

37


3.2 Hypotheses

This paper aims to find how changes in EUA prices affect firm performance of

European electricity generating firms. According to Arbitrage Pricing Theory, a

security’s expected return can be explained through its systematic exposure to

various macroeconomic factors, such as the return of EUA allowances. As such, a

positive effect should be revealed by a positive factor-beta coefficient on EUA price

changes through multivariate linear regression. A change in expected return, and

thereby stock price, reflects the expected future cash flows of the firm according

to the Efficient Market Hypothesis, and can, therefore, be used as a proxy for firm

financial performance.

Similar studies have already been conducted on the subject and argued for a pos-

itive effect on firm performance (Veith et al., 2009; Oberndorfer et al., 2012).

However, these studies were based on data from the first phase of the European

Union Emissions Trading System, and their results cannot be generalized to hold

during the currently active third phase of the system as there are several significant

differences. For instance, the emissions market has matured, and market agents

may have developed their understanding of the dynamics of the market. As such,

it is possible that the market has become more efficient and that participants have

changed their perceptions on how EUA price changes affect firm performance.

Moreover, multiple important policy changes may have changed the fundamental

functioning of the market. For the electricity sector, as an example, auctioning is

now the default method of emission allowance allocation and replaced free allo-

cation. Hence, one could suspect that the previously mentioned windfall profits

associated with grandfathering no longer occur. Additionally, renewable energy

sources have penetrated the European electricity market and are now responsible

for approximately one-third of electricity production output in the EU as compared

38


with a fifth just a decade ago (European Commission, 2019a). As such, the merit

curve may have been shifted in a way that changes the rent obtained by carbon

efficient inframarginal producers from its exposure to emission compliance costs.

Electricity generating technologies from the left: solar and wind power, nuclear power, thermalcoal power, and thermal natural gas power. Source: Cludius et al. (2014)

Figure 3.1: Illustration of the Merit Order Curve with emission compliance costs

We suspect that the implications of these changes apply different forces on the

relationship between EUA price changes and firm performance among electricity

generating firms. As an example, the penetration of renewable energy sources

during the last decade should arguably increase the positive effect, whereas the

end of grandfathering should have an opposite effect. These conflicting forces

notwithstanding, research shows that EUA price increases are passed on to con-

sumers in the bids of the marginal producer. As such, emission allowances do not

represent a significant regulatory burden for the marginal producer. Moreover, the

contribution margin of inframarginal producers increases with increased emission

allowance prices resulting in an overall increase in profitability that will be re-

flected in the stock prices of the respective companies. As an example, consider

the chart above, in which the dark area represents the costs of emissions for the

coal generating marginal producer. The arrow illustrates the pass-through of the

allowance price to the consumer, assuming 100% pass-through rate, resulting in

39


regulatory rents for the inframarginal producers. This leads to the first hypothesis

of the paper:

Hypothesis I: Emission allowance price increases (decreases) positively (negatively)

affected electricity generating stock returns during the sample period.

By the same logic as above, we suspect that inframarginal companies with a higher

share of low-margin renewable electricity generation obtained even larger contri-

bution margins than their more carbon-intensive counterparts. This leads to the

second hypothesis of the paper:

Hypothesis II: EU Emission allowance price increases (decreases) had a larger pos-

itive (negative) effect on carbon efficient electricity generating stock returns during

the sample period.

40

Chapter 4: Methodology

In Chapter IV, we present the data and methodology used to empirically test our

hypotheses. First, we present our research approach. Next, we introduce the data

used in our econometric analysis. Last, we present the econometric methodology.

4.1 Research Approach

This thesis is based on a deductive research approach in which we formulate a

set of hypotheses built upon existing theory and previous research. Based on the

hypotheses, we design a methodology that aims to test whether the hypotheses

hold true. Our empirical research is primarily based on the methodology of Veith

et al. (2009) and Oberndorfer (2009) that investigated the relationship between

price changes in EU emission allowances and financial performance of electricity-

generating firms during the first phase of the EU ETS. To increase reliability, fi-

nancial data is primarily retrieved from organized financial databases. Data on

firm-specific carbon intensity was obtained from a report published by PwC or

from reports published by the respective company. Price data for stocks, market

index, oil and natural gas were used to calculate weekly returns that were used

in the subsequent econometric analysis. To strengthen validity, we employ multi-

ple diagnostic tests to the data and econometric models. Additionally, robustness

is increased by controlling for commodity price impact and employing regression

41

CHAPTER 4. METHODOLOGY

models using aggregate and disaggregate stock return data as well as controlling

for firm-specific fixed effects.

4.2 Data

4.2.1 Firms

The sample selection was conducted in a series of steps. First, we identified the

largest publicly traded companies in the utility sector by market capitalization via

the Dow Jones STOXX Europe 600 Utilities index. It is the largest index of listed

utility companies in Europe. Next, we excluded companies that were not listed

in the index during the whole sample period of January 2013 to December 2019,

which left 26 companies. We then excluded companies that fell under the Water

Utility classification based on their Global Industry Classification Standard. Lastly,

we assessed whether the firm was involved in electricity generation throughout

the sample period via the information provided in their respective annual reports

for 2013 and 2018 or 2019. In the end, thirteen firms remained in the sample.

These firms are presented in Table 4.1. The stock return data of the companies

was retreived from Bloomberg.

42


Tabl

e4.

1:O

verv

iew

ofth

eco

mpa

nies

unde

ran

alys

is

Nam

eA

bbre

viat

ion

Tick

er(I

SIN

)C

oun

try

ofhe

adqu

arte

rC

arbo

nin

ten

sity

for

2015

(gC

O2/k

Wh)

SSE

PLC

SSE

GB

0007

9087

33U

nite

dK

ingd

om39

7En

desa

SAEL

EES

0130

6701

12Sp

ain

460*

Iber

drol

aSA

IBE

ES01

4458

0Y14

Spai

n21

7Ve

rbun

dA

GV

ERAT

0000

7464

09A

ustr

ia55

Nat

urgy

Ener

gyG

roup

SAN

TGES

0116

8703

14Sp

ain

431

Cen

tric

apl

cC

NA

GB

00B

033F

229

Uni

ted

Kin

gdom

145*

Engi

eSA

ENG

FR00

1020

8488

Fran

ce37

7El

ectr

icit

ede

Fran

ceSA

EDF

FR00

1024

2511

Fran

ce81

EDP

-Ene

rgia

sde

Port

ugal

EDP

PTED

P0A

M00

09Po

rtug

al52

0A

2ASp

AA

2AIT

0001

2334

17It

aly

436

Fort

umO

yjFO

RFI

0009

0071

32Fi

nlan

d21

Enel

SpA

ENE

IT00

0312

8367

Ital

y40

9R

WE

AG

RW

ED

E000

7037

129

Ger

man

y72

1

The

firm

-spe

cific

info

rmat

ion

has

been

colle

cted

thro

ugh

annu

alre

port

san

dB

loom

berg

term

inal

.Th

eca

rbon

inte

nsit

yw

asob

tain

edfr

omPw

C’s

Clim

ate

Cha

nge

and

Elec

tric

ity

repo

rt.

*End

esa

SAan

dC

entr

ica

plc

wer

eno

tin

clud

edin

the

Clim

ate

Cha

nge

and

Elec

tric

ity

repo

rtfr

omPw

C.

How

ever

,th

eca

rbon

inte

nsit

yfo

rEn

desa

SAan

dC

entr

ica

plc

was

obta

ined

thro

ugh

Ende

saSA

’sSu

stai

nabi

lity

Rep

ort

from

2015

and

aC

arbo

nD

iscl

osur

ePr

ojec

tfo

rC

entr

ica

from

2015

.So

urce

:En

desa

(201

5),P

wC

(201

6),C

entr

ica

(201

7)

43


4.2.2 Carbon Intensity

Carbon intensity is defined as the emission per unit of electricity output generated

and is usually measured in gCO2/kWh. We are concerned with the carbon inten-

sity of the firms’ portfolio of power plants located in Europe. We obtained the

carbon intensity in 2015 for the electricity-generating firms from the report Cli-

mate Change and Electricity published by PwC in 2016. The report did not include

the carbon intensity for Endesa SA and Centrica plc. Instead, carbon intensity for

Endesa SA was retreived from its 2015 annual sustainability report and the carbon

intensity for Centrica plc was obtained from a questionnaire to the organization

Carbon Disclosure Project.

Source: Endesa (2015), PwC (2016), Centrica (2017)

Figure 4.1: Carbon Intensity for sample firms

We create a binary variable that equals 1 if the company’s carbon intensity is above

the sample median and 0 otherwise. This allows us to divide our sample firms into

44


a carbon efficient and carbon intensive group.

Figure 4.1 shows the carbon intensity for the firms under analysis and the Eu-

ropean Union carbon intensity average. Notice that there is a large dispersion

between the firms under analysis regarding carbon intensity.

4.2.3 European Union Allowance Prices

Companies with installations subject to the EU ETS must surrender allowances

to cover their emissions by year-end. We assume that market participants are

risk-averse and that they hedge their exposure to EUA price fluctuations by taking

positions in futures December contracts. EUA futures with expiration in Decem-

ber have significantly higher volumes than emission allowances traded in the spot

market and should, therefore, reflect prices better. Moreover, futures’ prices are

less affected by short run demand and supply fluctuations and thus less noisy

(Oberndorfer, 2009). Similar studies have also been based on futures (Oberndor-

fer, 2009; Veith et al., 2009).

The futures settlement prices were received by request from the European Energy

Exchange (EEX) that kindly provides its data for academic purposes. Although

there are multiple exchanges in which EUA futures contracts are traded, prices de-

velop identically across markets (Mansanet-Bataller et al., 2007), consistent with

the Law of One Price.

4.2.4 Market Returns

The Dow Jones STOXX Europe 600 covers large, mid and small capitalization

companies across 17 European countries: Austria, Belgium, Denmark, Finland,

France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Norway, Poland,

45


Portugal, Spain, Sweden, Switzerland and the United Kingdom. It is weighted

based on free-float market capitalization, and its composition is reviewed on a

quarterly basis. The index represents a highly diversified portfolio and will be

used as a proxy for the overall market returns.

4.2.5 Control Variables: Oil and Natural Gas

In line with Arbitrage Pricing Theory and the idea of multifactor market models,

we add additional variables to the regression model that captures other potentially

influencing macroeconomic factors. We choose to follow Veith et al. (2009) and

Oberndorfer (2009) and include the price changes of oil and natural gas in our

regression equation. This will facilitate in comparing our results with their study.

Unlike Oberndorfer (2009), however, we choose not to include electricity prices,

which did not provide explanatory power in his study.

We use the price of one-month Brent crude oil contracts because it is extracted

from the North Sea and thus the primary source of oil in Europe. As for natural

gas prices, we use the one-month TTF Natural Gas contracts. The TTF Natural gas

is the largest market for natural gas in Europe. Both time series were retreived

from Bloomberg and Euro-denominated.

4.2.6 Transformation of Data and Frequency

Economic time series often exhibit growth that is approximately exponential, mean-

ing that in the long term the series grows by a certain rate per year on average.

This may have negative implications when performing linear regressions. To miti-

gate the risks associated with this, we calculate the natural logarithm of the time

series data to transform the series to exhibit linear growth (Stock & Watson, 2015).

46


This transformation also is often considered to stabilize the variance in the time

series data (Woolridge, 2013).

Using weekly data frequency, rather than daily, avoids issues with error-in-variables

problems with respect to irregularities (Scholes & Williams, 1977). Although

Oberndorfer (2009) used daily frequency, he argued that weekly observations are

preferable if the sample period is large enough because it reduces noise. We calcu-

late the weekly return using the closing price of the last trading day of the week.

Our sample period stretches from the first week of 2013 to the last week of 2020

with results in more than 360 observations for each series in the data set.

4.3 Econometric Theory

The following section provides an explanation of the econometric methods that

will be used to test the hypotheses presented in Section 3.2. First, the reader

is introduced to the Ordinary Least Squares (OLS) regression estimation. Next,

we discuss issues related to OLS estimation in terms of omitted variable bias and

present a Fixed Effects regression model to mitigate these issues. The inspiration

to this section is from Stock and Watson (2015), Woolridge (2013), and Enders

(2014).

4.3.1 OLS Estimation

Ordinary Least Squares (OLS) estimation a type of linear least squares method for

estimating the unknown parameters in a linear regression model. The multivariate

regression model can be expressed in matrix notation as follows, for consistency

47


vectors and matrices are denoted by bold type:

Y = Xβ + U (4.1)

Where:

• Y is the n× 1 dimensional matrix of n observations on the k + 1 regressors -

including the regressor for the intercept.

• X is the n×(k+1) dimensional matrix of n observations on the k+1 regressors

- including the regressor for the intercept.

• The k + 1 dimensional column vector Xi is the ith observation on the k + 1

regressors; that is, X′i = (1X1,i . . . Xk,i), where X′i denotes the transpose of

Xi.

• U is the n× 1 dimensional vector of the n error term.

• β is the (k + 1) × 1 dimensional vector of the k + 1 unknown regression

coefficients.

For the OLS estimation to provide the best linear unbiased estimators, the follow-

ing assumptions must hold.

1. E (ui|Xi) = 0, ui has conditional mean of zero.

2. (Xi, Yi) , i = 1, . . . , n, are independently and identically distributed (i.i.d)

draws from their joint distribution.

3. Large outliers are unlikely: Xi and ui have nonzero finite fourth moments.

4. X has full column rank i.e. there is no perfect multicollinearity.

48


The principle of OLS estimation is to minimize the sum of squared prediction

errors over all n observations. The sum of squared predicted errors can be written

as:n∑i=1

(Yi − b0 − b1X1,i − . . .− bkXk,i)2 (4.2)

Where Yi is the observed value, which the OLS estimation predicts, b0, b1, . . . , bk are

estimates of β0, β1, . . . , βk, which minimize the sum of squared prediction errors

and X1,i and Xk,i are the regressors included in the OLS estimation. The estimates

of the coefficients that minimize the sum of squared prediction errors are called

OLS estimators and are denoted as β in matrix notation. The difference between

Yi and the predicted Yi is the OLS residual and is denoted U in matrix notation.

The OLS estimators can be estimated by a closed form solution. These are ob-

tained by taking the derivative and setting the derivatives of the sum of squared

prediction errors with respect to each coefficient in the vector to zero and solving

for the estimator β. The solution to the equation system yields the OLS estimator

β. The closed form solution can be written as follows:

β = (X ′X)−1X ′Y (4.3)

A challenge in using multivariate linear regression models to explain a real-world

phenomena is to include all variables that have explanatory power on the depen-

dent variable. As such, there is typically an issue related to omitted variable bias.

4.3.2 Omitted Variable Bias

Omitted variable bias occurs when a regressor is correlated with a variable not in-

cluded in the analysis, and that variable influences the dependent variable. There

49


are two general conditions, which must hold true for an omitted variable bias to

arise: i) the omitted variable is a determinant of the dependent variable and ii)

the omitted variable is correlated with the regressors included in the analysis. If

there is an omitted variable bias in the model, the first least squares assumption

is is violated. The reason for this is that the error term ui is correlated with one

of the regressors and therefore, assumption one does not hold true. This leads to

the OLS estimators being biased and inconsistent. This can be explained mathe-

matically, let the correlation between Xi and ui be corr (Xi, ui) = ρXu. Assume

the second and third assumptions of the OLS holds true, but the first is violated,

because ρXu 6= 0. Then:

β1→β1 + ρXuσuσX

(4.4)

Even in large samples the omitted variable bias can an issue, because β does not

converge in probability to the true value β. The term ρXuσuσX

is the bias of β, which

is also present in large samples. The size of the bias depends on the correlation

between the regressor and error term, |ρXu| (Stock & Watson, 2015).

To mitigate the issues of omitted variable bias related to firm-specific character-

istics, we investigate the effect of a EUA price change by using the fixed effects

regression model. The fixed effects regression model is explained in the next sec-

tion.

4.3.3 Fixed Effects Regression

The fixed effects regression model is used to control for omitted variables in panel

data, when the omitted variables vary across entities but do not change over time.

The general panel regression model can be written as follows:

Yi,t = β0 + β1Xi,t + β2Zi + ui,t (4.5)

50


Where Zi is the unobserved variable that varies across entities i = 1, . . . , n, but

does not change over time. The goal is to estimate β1, i.e. to isolate the effect on

Yi given a change in Xi holding Zi constant. The model can be rewritten by letting

ai = β0 + β2Zi, and we obtain the generalized fixed effects regression model:

Yi,t = βXi,t + αi + Ui,t (4.6)

With i = 1, . . . , n and t = 1, . . . , T . This model allows the model to have specific

intercepts αi, where each αi can be understood as the fixed effect of entity i.

The fixed effects model’s entity-specific intercepts can be captured by using binary

variables, which in our case refers to the specific companies in our sample. Let

D1i be a binary variable which equals 1, when i = 1 and equals zero otherwise

and let D2i equal 1 when i = 2 and equal zero otherwise and so on. To avoid the

dummy variable trap, one of the binary variables should be excluded. The fixed

effects regression model can be written as:

Yi,t = β0 + β1Xi,t + γ2D2i + γ3D3i + . . .+ γnDni + ui,t (4.7)

This formula is equivalent to the generalized fixed effects regression model, but

explicitly shows the company-specific intercepts.

The first and second assumptions for the fixed effect regression model are slightly

different from the assumptions for the multivariate regression model, but the third

and the fourth are similar.

The assumptions for the fixed effects regression model are:

1. ui,t has conditional mean zero: E (uit|Xi,1, Xi,2, . . . , Xi,T , ai) = 0

2. (Xi,1, Xi,2, . . . , Xi,T , ui,1, ui,2, . . . , ui,T ) , i = 1, . . . , n are i.i.d. draws from

51


their joint distribution.

3. Large outliers are unlikely: Xi and ui have nonzero finite fourth moments.

4. There is no perfect multicollinearity.

4.3.4 Random Effects Regression

In contrast to fixed effects models, random effect models require that the unob-

served variables are uncorrelated with all the explanatory variables. If ai repre-

sents the unobserved variables, then Cov (xi,t,j, ai) = 0 must hold. In our case,

this his highly unlikely, as stock prices and EUA allowance prices are affected by

complex relationships between a magnitude of factors. Recall in Section 2.3.2,

the price of EU ETS allowances is determined by a variety of forces, for instance

weather conditions, which affects business-as-usual emissions that drive demand

for allowances. As such, we cannot plausibly argue that the unobserved variables

are uncorrelated with all explanatory variables, and we choose not to employ a

random effects model.

4.3.5 OLS Diagnostics

The first assumption for multivariate OLS estimation states that the error terms

should have a conditional mean of zero. This assumption implies that at any given

value of Xi, the errors should have a mean of zero. This can be investigated by

plotting the error terms over the fitted values of the model, which allows us to

visually assess if the estimated model’s residuals have a mean of zero indicating

white noise. Additionally, this plot can further provide indication if the error terms

are homo- or heteroscedastic. For homoscedasticity, the residuals should not show

indication of a pattern in the plot.

52


As stated earlier, the second assumption is that variables are independently and

identically distributed across entities for i = 1, . . . , n relating to how the sample is

drawn. We can use the Normal QQ plot to investigate the residuals to see if they

are normally distributed. If the residuals are normally distributed, they should

follow the diagonal line in the QQ-plot.

The third assumptions that large outliers are unlikely is important because the es-

timated coefficients are sensitive towards large outliers (Stock & Watson, 2015).

According to James et al. (2017) it is possible to identify outliers by examining

the Residuals vs leverage plot. To investigate the assumption for perfect multi-

collinearity, a correlation matrix for the variables is drawn.

The assumptions for fixed effects regression are quite similar to the multivariate

regression. The first assumption fixed effects is identical the first assumption in

multiple regression - the error term has to have a conditional mean of zero. This

can be violated if the errors are correlated with the values of the dependent vari-

ables. Further, for fixed effects Xi,t is allowed to be correlated over time within

the entity, i.e. Xi,t can be autocorrelated. If the error terms exhibit autocorrela-

tion, heteroscedasticity and autocorrelation consistent (HAC) standard errors can

be used to take this into account. The third and fourth assumptions are the same

as in the multivariate regression model. (Stock & Watson, 2015).

4.3.6 Autocorrelation

Time series data can exhibit autocorrelation, referring to that the value observed in

one period is correlated with the value in another. If present in the series, the least-

squares estimator might not be the best linear unbiased estimator. To examine if

the data exhibits autocorrelation, one can plot the autocorrelation function (ACF

plot) for the series. The plot displays the autocorrelation function’s correlation

53


coefficients between lags in time. Significant spikes indicate that autocorrelation

is present in the series (Stock & Watson, 2015).

A formal test for autocorrelation is the Durbin Watson test. It tests for the auto-

correlation in lag 1. If the residual is given by εt = ρεt−1 + ut, then the Durbin

Watson statistics tests the null hypothesis of ρ = 0, which means no first-order au-

tocorrelation and the alternative hypothesis ρ 6= 0 meaning first-order correlation

exists.

The Durbin Watson statistic is defined as:

DW =

∑Nt=2(εt − εt−1)2∑N

t=1 ε2t

(4.8)

Where ε is the residuals of our model, if the p-value reported by the Durbin Watson

test is significant it rejects the null hypothesis indicating that first order autocor-

relation among the residuals is present (Watson, 1951).

Again, if the ACF plot and/or the Durbin Watson test indicate that the errors are

autocorrelated, then the standard errors should be modelled using the heteroskedasticity-

and autocorrelation-consistent (HAC) standard errors (Stock & Watson, 2015).

4.3.7 Stationarity

Stock and Watson (2015) define time series data to be stationary if its probability

distribution does not change over time. This means that the joint distribution of

(Ys+1, Ys+2, . . . , Ys+T ) does not depend on s regardless of the value of T. If this is the

case Yt is said to be nonstationary. Stationarity requires that the future is like the

past in a probabilistic sense. If data is not stationary, it imposes several problems,

and the assumptions above might not hold true (Stock & Watson, 2015).

54


To test for stationarity, an informal and important method is to plot the time series

and the corresponding ACF plot for a visual interpretation. This allows us to see if

there is a stochastic trend present in the data. As stated in Section 4.2.6, the data

is transformed into its natural logarithmic return to stabilize the variance in the

data.

A formal process to detect stationarity is to conduct an augmented Dickey-Fuller

test to test for a unit root. The idea of the test is to test whether previous lags of

the Yt has an effect on Yt. Therefore the null hypothesis is H0 : γ = 0 versus the

one-sided alternative H1 : γ < 0 in the regression:

∆yt = β0 + γyt−1 +

p∑i=2

βi∆yt−i+1 + εt (4.9)

Where ∆yt is the change in the variable, β0, γ and βi are the estimated intercept,

the coefficient for the first lag and the coefficients for additionally lags running

from i = 0 to as many lags included in the model, respectively. By rejecting the

null hypothesis, the data is considered to be stationary. If a time trend element

tt is included in the regression, the alternative hypothesis changes to Yt being

stationary around a deterministic time trend, and the regression becomes:

∆yt = β0 + tt+ γyt−1 +

p∑i=2

βi∆yt−i+1 + εt (4.10)

The resulting t-statistic needs to be compared with the reported Dickey-Fuller crit-

ical values to determine whether to accept or reject the null hypothesis. The crit-

ical values vary depending on the equation under investigation (Stock & Watson,

2015). Enders (2014) argues that the selection of lag length can be difficult. If

an insufficient amount of lags is selected then the residuals do not behave like

white-noise processes while too many lags will reduce the power of the test in

order to reject the null of a unit root. When adding a lag, the number of pa-

55


rameters increases and the number of usable observations decreases. Stock and

Watson (2015) suggest that the lag length p can be estimated by using the Akaike

information criteria (AIC).

When testing our times series data, we will first investigate the visualization of

the data in order to determine whether or not to include a time trend in the Aug-

mented Dickey-Fuller test. When we are using real return data, the expectation

is that we should not find our data to be stationary around a deterministic trend

(Enders, 2014). A drift term will be included in the model because, on average,

the stock market is yielding a positive return (Martin & Wagner, 2016).

4.3.8 Econometric Methodology

The econometric analysis of this research will, to a large extent be based on Veith

et al. (2009) and Oberndorfer (2009) that studied how EUA prices affected firm

performance for electricity-generating companies during the first phase of the EU

ETS. This will allow us to compare our results in the subsequent discussion.

First, the returns of the electricity companies included in our sample will be re-

gressed on the return of the market portfolio and return on EU ETS allowances

during the whole sample period. The model to be estimated:

ri,t = β0 + β1rMkt,t + β2rEUA,t + εi,t (4.11)

Where ri,t represents the returns of each firm, i runs from from 1 to 13 over the

firms and t indicates the time, which runs from the first week in 2013 to the last

week in 2019. rMkt,t represents the returns of the market portfolio and rEUA,t

represents the returns of the emission allowances. β0, β1 and β2 are the estimated

coeffiecients and εi,t is the error term for each firm at each point in time.

56


A positive and significant β2 coefficient would indicate a positive relationship be-

tween price changes in EUA and firm performance. Next, we extend the model

to control for commodity price impact by including control variables for the re-

turns on one-month Euro-denominated futures contracts for Brent crude oil and

TTF natural gas. Oil and natural gas are frequently used as input factors among

conventional electricity producers and an change in price in these factors could

have an impact on expected future cash flows that could be reflected in the share

price. Moreover, recall Section 2.3.2 that explains that the price of fossil fuels

drive prices on emissions. For instance, a decrease in natural gas prices relative to

coal will lead to lower emissions and lower demand for allowances, resulting in a

lower equilibrium price for allowances. The model to be estimated:

ri,t = β0 + β1rMkt,t + β2rEUA,t + β3rOil,t + β4rGas,t + εi,t (4.12)

Where the variables from Equation 4.11 are the same, and the additional variables

rOil,t and rGas,t represents return for one-month Euro-denominated Brent crude oil

and TTF natural gas futures running from the first week in 2013 to the last week

in 2019.

Following Oberndorfer (2009), we will run these regressions both using pooled

panel data and the aggregate returns in an equally weighted portfolio. The model

is slightly different when running the regression with the equally weighted port-

folio, and the model becomes:

rPf,t = β0 + β1rMkt,t + β2rEUA,t + εt (4.13)

Where rPf,t is the return for the equally weighted portfolio, t runs from the first

week of 2013 to the last week of 2019. The rest of the variables are the same as

in the pooled regression model explained above.

57


Including the control variables oil and natural gas the model to be estimated is:

rPf,t = β0 + β1rMkt,t + β2rEUA,t + β3rOil,t + β4rGas,t + εt (4.14)

Next, we follow Veith et al. (2009) and control for fixed effects which allows us

to partial out firm-specific effects. For instance, firms are involved in electricity

generation to varying extents and subject to different local regulations that could

have an effect, for example, by limiting the pass-through rate. A dummy variable

equaling 1 for firm i, or 0 otherwise, will be added to the model. The model to be

estimated:

ri,t =∑i

ΘiDFIRMi + β1rMkt,t + β2rEUA,t + β3rGas,t + β4rOil,t + εi,t (4.15)

We then turn to investigating how the effect has changed over time. For instance,

policy actions such as the Market Stability Reserve that was announced in 2018

could possibly change the relationships that are observed. Potential differences

could provide additional insights that can contribute to the subsequence discus-

sion. We run the extended model that includes oil and natural gas rolling seven

times over subperiods of 12 months, using the pooled data in order to achieve a

sufficiently large set of observations.

Lastly, we will investigate if the carbon intensity of the underlying generation mix

has an effect on financial performance. Recall from Section 4.2.2 that we will

create a dummy variable representing whether the firm has a higher or lower

carbon intensity than the median. This dummy variable will be interacted with

the emission allowance return factor through the following equation:

ri,t = β0 + γ0Polluteri + β1rMkt,t + β2rEUA,t + γ1 (rEUA,t ∗ Polluteri) + εi,t (4.16)

58


This will allow for different intercepts and coefficients in terms of emission al-

lowance returns. Consider firm i that has a lower than median carbon intensity.

Its binary variable Polluter will equal 1 in the regression. γ0 + β0 will represent the

intercept for the firm and the interaction term γ1 (rEUA,t ∗ Polluteri) coefficient

represents the interaction effect. In the case that firm i has a higher than median

carbon intensity, then both γ0 and γ1 will fall out of the equation.

We consider heteroskedasticity and autocorrelation by consistently calculating and

reporting heteroscedasticity and autocorrelation consistent standard errors (HAC).

59

Chapter 5: Results

Chapter V will walk the reader through the empirical analysis of the paper. The

chapter is structured as follows. First, the reader is presented with an overview

of the data. Second, the data will be tested for autocorrelation and stationarity.

Third, the results from the regressions relating to the first and then the second

hypotheses will be presented. Lastly, diagnostics plots will be provided and dis-

cussed.

5.1 Overview of the Data

Figure 5.1 shows the indexed price development of the aggregate equally weighted

portfolio of the sample firms, the market portfolio, oil, and natural gas during the

third phase of the EU ETS. The aggregate price development of the sample stocks

appears to move in the same direction as the market portfolio that is proxied

through the Dow Jones STOXX Europe 600 index through most of the period.

However, during the period 2018 and 2019, the electricity-generating companies

outperform the market. Simultaneously, the price of European Union Allowances

exhibit a strong positive trend with a high degree of volatility. This could poten-

tially be a consequence of the Market Stability Reserve that was announced in

2018 and implemented in 2019. Recall from Chapter II, the Market Stability Re-

serve limited the supply of allowances available in the market in efforts to increase

60

CHAPTER 5. RESULTS

The figure illustrates the indexed price development of the equally weighted portfolio (Portfolio ofelectricity generating firms), European Union Allowances (EUA), Dow Jones STOXX Europe 600(Market Portfolio), one-month Euro-denominated TTF natural gas contracts (Gas) and one-montheuro-denominated Brent crude oil (Oil). The figure is based on weekly data from January 1st 2013to December 31st 2019.

Figure 5.1: Price Development of the variables under analysis

the price of emission allowances. Prior to this, the emission allowance price de-

veloped similarly to oil and natural gas. Both oil and natural gas have seen their

prices decrease significantly. The price of oil has nearly halved, and natural gas

has dropped to a quarter of its starting value.

Table 5.1 presents descriptive statistics for the data used in the analysis. The

equally weighted portfolio of our sample firms has outperformed the market dur-

ing the period with an annualised mean return of 9.6% compared with 6.0% for

the market portfolio. The average return is positive for all companies in the sam-

ple except Centrica plc (CNA). The stock with the highest return is A2A SpA (A2A)

with an average annual return of 22.6% during the sample period. The price for an

61

CHAPTER 5. RESULTS

emission allowance has increased with an annualised mean of almost 20% over the

sample period. It has a high standard deviation of more than 40%. As illustrated

in Table 5.1, prices of an emission allowance varied significantly with a range

between approximately e3 to e30. The price of one-month Euro-denominated

futures contracts for Brent crude oil and TTF natural gas decreased on average by

5% and 10% per annum, respectively, during the sample period.

Next, calculate the Pearson’s correlation coefficients between the series to identify

possible multicollinearity issues between regressors and to potential irregularities.

The correlation matrix in Table 5.2 does not raise any concerns in terms of perfect

multicollinearity between the series. The variables with the highest correlation

coefficients have been colored - the correlation coefficients between 0.6 and 0.7

have been colored light gray and above 0.7 dark gray. The sample firms are all

positively correlated with the returns of emission allowances. The correlations

vary between 0.014 (Iberdrola SA) to 0.288 (Verbund AG). Moreover, as one can

expect, the electricity generating companies’ returns are positively correlated with

the return of the market portfolio. Once again, Centrica plc stands out with its

relatively low correlation to its peers with a correlation of 0.43 to the equally

weighted portfolio of sample firms.

Because of the low correlation and the abnormal negative average return for Cen-

trica plc, we suspect that the stock could potentially be different in some relevant

aspect from the rest of the population. In order to investigate Centrica plc, we plot

the price development of the equally weighted portfolio of electricity generating

firms, the market portfolio and Centrica plc for the total sample period.

62

CHAPTER 5. RESULTS

Table 5.1: Descriptive statistics for variables under analysis

Dependentvariables

Mean Median Max Min Std. Dev Obs

SSE 0.058 0.103 0.080 - 0.140 0.194 364ELE 0.197 0.262 0.139 - 0.083 0.196 364IBE 0.163 0.194 0.118 - 0.095 0.186 364VER 0.150 0.209 0.100 - 0.116 0.260 364NTG 0.125 0.123 0.139 - 0.093 0.210 364CNA - 0.132 - 0.122 0.124 - 0.194 0.271 364ENG 0.053 - 0.018 0.131 - 0.084 0.219 364EDF 0.015 0.022 0.140 - 0.141 0.305 364EDP 0.132 0.140 0.106 - 0.115 0.225 364A2A 0.226 0.248 0.113 - 0.117 0.252 364FOR 0.131 0.158 0.091 - 0.154 0.216 364ENE 0.156 0.300 0.099 - 0.102 0.216 364RWE 0.013 0.237 0.161 - 0.194 0.342 364Portfolio 0.096 0.209 0.091 - 0.074 0.172 364Independent variablesMarket return 0.060 0.144 0.054 - 0.071 0.150 364EUA price 10.18 6.850 29.58 3.230 7.429 365EUA return 0.198 0.224 0.196 - 0.428 0.481 364Oil price 58.94 55.81 89.03 26.43 15.00 365Oil return - 0.048 0.078 0.135 - 0.154 0.295 364Gas price 19.40 19.30 29.35 9.430 4.835 365Gas return - 0.108 - 0.201 0.281 - 0.232 0.392 364

This table reports the mean, median, maximum (Max), minimum (Min), the standarddeviation (Std. Dev.) and the number of observations (Obs) for each firm, the portfolioof electricity generating firms (Portfolio), European Union Allowances (EUA), Dow JonesSTOXX Europe 600 (Market), one-month Euro-denominated Brent crude oil (Oil) futurescontracts, and one-month Euro-denominated TTF natural gas futures contracts (Gas) fromJanuary 1st 2013 to December 31st 2019. The Mean, median and standard deviationhave been transformed into annualised figures. Maximum and minimum are presented inweekly values.

63

CHAPTER 5. RESULTS

Tabl

e5.

2:C

orre

lati

onm

atri

xof

retu

rns

Port

folio

SSE

ELE

IBE

VER

NTG

CN

AEN

GED

FED

PA

2AFO

REN

ER

WE

Oil

Gas

MK

TEU

A

Port

folio

1SS

E0.

538

1EL

E0.

687

0.35

91

IBE

0.75

30.

393

0.72

21

VER

0.70

40.

249

0.34

90.

416

1N

TG0.

746

0.33

60.

601

0.71

40.

393

1C

NA

0.43

20.

586

0.27

40.

269

0.17

10.

340

1EN

G0.

798

0.45

90.

530

0.62

40.

463

0.56

50.

371

1ED

F0.

668

0.32

10.

352

0.44

00.

395

0.42

20.

322

0.52

81

EDP

0.60

90.

367

0.51

50.

579

0.34

40.

545

0.25

50.

488

0.37

81

A2A

0.53

00.

260

0.46

00.

543

0.31

50.

496

0.23

00.

485

0.26

90.

420

1FO

R0.

690

0.36

30.

363

0.41

80.

469

0.44

30.

278

0.52

10.

428

0.35

30.

332

1EN

E0.

681

0.35

40.

620

0.70

80.

344

0.59

40.

283

0.65

00.

364

0.49

70.

631

0.38

01

RW

E0.

771

0.31

30.

421

0.48

30.

483

0.47

40.

257

0.57

20.

430

0.40

70.

319

0.47

40.

482

1O

il0.

322

0.28

00.

116

0.14

70.

219

0.28

40.

296

0.24

80.

246

0.22

00.

135

0.29

00.

173

0.19

31

Gas

0.10

30.

047

0.05

80.

027

0.09

90.

068

-0.0

010.

059

0.01

40.

044

0.08

20.

139

0.07

40.

075

0.10

91

MK

T0.

686

0.41

10.

513

0.62

20.

374

0.59

90.

411

0.65

40.

485

0.53

50.

484

0.49

60.

630

0.43

90.

347

0.07

21

EUA

0.18

90.

076

0.05

70.

014

0.28

80.

025

0.05

40.

094

0.16

10.

033

0.11

20.

243

0.04

40.

082

0.10

10.

184

0.05

71

This

tabl

ere

port

sPe

arso

n’s

corr

elat

ion

coef

ficie

nts

for

the

seri

esin

the

data

set.

The

vari

able

sar

e:Po

rtfo

lio:

the

aggr

egat

eeq

ually

wei

ghte

dre

turn

sof

the

sam

ple

ofel

ectr

icit

yge

nera

ting

firm

s;in

divi

dual

com

pani

es(s

eeTa

ble

5.1

for

full

nam

es);

MK

T:D

owJo

nes

STO

XX

Euro

pe60

0;EU

A:D

ecem

ber

futu

res

cont

ract

for

Euro

pean

Uni

onA

llow

ance

s;O

il:on

e-m

onth

Euro

-den

omin

ated

futu

res

cont

ract

for

Bre

ntcr

ude

oil;

and

gas:

one-

mon

thEu

ro-d

enom

inat

edfu

ture

sco

ntra

ctfo

rT

TFna

tura

lgas

.C

orre

lati

onco

effic

ient

sbe

twee

n0.

6an

d0.

7ar

eco

lore

din

light

gray

,cor

rela

tion

coef

ficie

nts

abov

e0.

7ar

eco

lore

din

dark

gray

.

64

CHAPTER 5. RESULTS

The figure illustrates the indexed price development of the equally weighted portfolio (Portfolioof Electricity Generating Firms), Dow Jones STOXX Europe 600 (Market portfolio) and Centricaplc. The figure is based on daily data from January 1st to December 31st 2019. In addition, it isindicated, when the United Kingdom voted for Brexit the 23rd of June 2016.

Figure 5.2: Price development of the Portfolio of Electricity Generating Firms, the Market Portfolioand Centrica plc

Figure 5.2 clearly shows the abnormal price development of Centrica plc during

the total period. Centrica plc yielded a negative return of -69.5% while the market

portfolio yielded a positive return of 40.6%. Centrica plc appears to have followed

similar price movements as its peers and the market portfolio until the middle of

2016, when it diverged and moved in the opposite direction to the average of the

sample firms. This coincides with the Brexit referendum vote on the 23rd of June

2016, which could potentially have an effect on how the market values Centrica

plc’s shares.

The market could discount certain risks associated with Brexit for Centrica plc, for

instance, the potential loss of cross-border electricity trading between the United

65

CHAPTER 5. RESULTS

Table 5.3: Correlation matrix for Centrica plc

Before Brexit After Brexit Total period

Portf. CNA EUA Portf. CNA EUA Portf. CNA EUA

Portfolio 1 1 1CNA 0.551 1 0.341 1 0.432 1EUA 0.192 0.132 1 0.18 -0.015 1 0.189 0.054 1

This table displays Pearson’s correlation coefficients for the equally weighted portfolio ofelectricity generating firms (Portf.), Centrica plc (CNA), and European Union Allowances(EUA) before Brexit, after Brexit and for the total period.

Kingdom and the European Union. It could also potentially affect the way that

market participants discount the effect of price changes in emission allowances if

they believe that Centrica plc’s British installations will no longer be subject to the

emission compliance cost after the United Kingdom has left the European Union.

The correlation matrix in Table 5.3 displays a large decrease in the correlation

between the returns of Centrica plc and emission allowances from 0.13 to -0.015

before and after Brexit. Whether this is a consequence of Brexit or merely a co-

incidence remains unclear and is outside the scope of this research. However, we

decide to perform a sensitivity analysis by performing the first sets of regressions

with and without Centrica plc in the sample. In the case of material differences

between the regressions, we must consider taking further steps to accommodate

this potential issue.

66

CHAPTER 5. RESULTS

5.2 Autocorrelation and Stationarity

As described in Section 4.3.6 a common method to investigate if times series are

stationary is to visualize the data. Figure 5.3 illustrates ten different plots. The

left hand side displays the natural logarithmic weekly return for the total sample

period for the aggregate equally weighted returns of the sample of electricity gen-

erating firms, the emission allowances, the market portfolio as well as one-month

Euro-denominated futures for Brent oil and TTF natural gas. Firm-specific plots

can be found in the appendix.

Overall, the plots on the left-hand side in Figure 5.3 indicate that the series are

stationary. The first plot on the left-hand side is the return for the equally weighted

portfolio of electricity generating firms and indicates a relatively constant mean

with a few spikes over the sample period. The emission allowances returns show

large spikes in the beginning of the sample period but seems to stabilize over time.

Natural gas stands out with a large variation at the end of the period. In general,

the plots indicate that the series are stationary.

We formally test for stationarity by testing for unit root in each series with the

Augmented Dickey-Fuller test. Because none of the plots of the returns indicate

that a trend is present we do not include a trend term. The test’s null hypothesis

is that the time series has a unit root. As such, the series is stationary according

to the test if it returns a γ t-statistic that is significant compared to the reported

Dickey-Fuller critical values.

Table 5.4 returns the results from the Augmented Dickey-Fuller test for each series.

One lag was selected by the Akaike Information Criterion and used in the test. The

Augmented Dickey-Fuller returns γ t-statistics that are significant when compared

to the reported critical values.

67

CHAPTER 5. RESULTS

Figure 5.3: Plots of the logarithmic weekly return and the corresponding ACF plot

68

CHAPTER 5. RESULTS

Table 5.4: Summary of Augmented Dickey-Fuller test

Dependent variables Intercept γ Lags Obs

SSE 0.31 -15.221*** 1 364ELE 0.002*** -15.385*** 1 364IBE 0.006*** -15.852*** 1 364VER 0.07* -13.884*** 1 364NTG 0.08* -14.937*** 1 364CNA 0.20 -13.923*** 1 364ENG 0.42 -14.833*** 1 364EDF 0.90 -14.063*** 1 364EDP 0.08* -15.489*** 1 364A2A 0.01** -14.304*** 1 364FOR 0.07* -14.746*** 1 364ENE 0.03** -14.732*** 1 364RWE 0.85 -13.255*** 1 364Portfolio 0.07* -15.419*** 1 364

Independent variables

Market 0.26 -14.618*** 1 364Oil 0.70 -13.321*** 1 364Gas 0.39 -16.162*** 1 364EUA 0.20 -14.249*** 1 364

Significance code: *p<0.1; **p<0.05; ***p<0.01

Critical values1% Level -3.445% Level -2.8710% Level -2.57

This table reports the p-values for the intercepts (Intercept), the t-statistic for the γ-coefficient (γ), number of lags selected (Lags) and number of observations (Obs). Theresults are a summary for each sample firm under analysis with the abbreviation from Ta-ble 4.1, the equally weighted portfolio (Portfolio), the market proxy - Dow Jones STOXXEurope 600 (Market), the one-month Brent crude oil (Oil), TTF natural gas contracts(Gas) and the European Union Allowances (EUA). The number of lags is selected by theinformation criteria AIC.

The autocorrelation function (ACF) plots on the right hand side of Figure 5.3

display the correlation coefficients between a time series and its lags. There are a

few significant spikes on the lags of the series for the natural logarithmic return of

the equally weighted portfolio, emission allowances, and natural gas. As such,

the autocorrelation function indicates that these series exhibit some degree of

autocorrelation. This will be taken into account by calculating heterscedastic and

69

CHAPTER 5. RESULTS

autocorrelation consistent standard errors.

5.3 Regression Results

The empirical analysis is structured as follows. First, we test Hypothesis I by

employing multivariate regression models on the data for the whole period using

disaggregate pooled data and aggregate data for the return of the stocks as well

as controlling for firm-specific fixed effects. Next, we control for commodity price

by including the returns of one-month Euro-denominated futures contracts for

Brent crude oil and TTF natural gas. To investigate if the empirical results are

consistent over time we run the pooled and aggregated regressions rolling over

twelve month subperiods. Last, we test for Hypothesis II by including a binary

variable representing the firms’ carbon intensities relative to the sample median. It

is interacted with the return on emission allowances to allow for different intercept

and slope in regard to emission price changes.

5.3.1 Hypothesis I

In this section, we present the empirical results related to the first hypothesis,

namely, if there is a positive effect from emission price increases on the stocks of

our sample firms. We begin by regressing the firms’ stock return with the returns

of the market portfolio and emission allowances as described with Equation 4.11

with the pooled panel data. The regression is run including and excluding Centrica

plc that appeared to behave differently to its peers to see if it affects the regression

results significantly. The results of the regressions are presented in Table 5.5.

The pooled regression that excludes Centrica plc (Excl. CNA) has a slightly higher

adjusted R2 and minor differences in the explanatory variables’ coefficients. Not

70

CHAPTER 5. RESULTS

surprisingly, the factor-beta for the market portfolio is positive but below one,

which is typical for defensive large utility stocks. In line with the first hypothesis,

the factor coefficients for the return on emission allowances are positive at 0.036

and 0.038 with and without Centrica plc, respectively, and significant at 1% level.

We extend the model to take into account a potential commodity price impact by

including control variables for the returns on oil and natural gas. The model that

is estimated is Equation 4.12. The addition of oil and natural gas returns in the

model has a minor positive contribution to the adjusted R2. The change in coeffi-

cients for the return on emission allowance prices and market portfolio is slightly

negative and inconsequential to the conclusions from the previous regression. Oil

is significant at 1% using the full sample and 5% excluding Centrica plc. Because

natural gas is not significant, we test whether natural gas is jointly significant to-

gether with oil by performing an F-test. It indicates that natural gas and oil are

jointly significant with a p-value of 0.04. As such, we decide to keep natural gas

in the equation. Moreover, since we cannot find any significant differences in the

results between the regressions that include the full sample or the sample that

excludes Centrica plc in the pooled regression we decide to proceed by using the

full sample of companies in the regressions that will follow.

Next, we run the same regressions but using the equally weighted portfolio. The

base model that includes the return on the market and the return on emission al-

lowances estimates a lower market factor-beta of 0.774 and a higher EUA factor-

beta of 0.054. As with the pooled regression, the addition of the control variables

for oil and natural gas produces very similar results with an emission allowance

coefficient of 0.050. Again, natural gas is not significant, and we perform an F-

test to test for joint significance with oil. Contrary to the pooled regression, it

fails to reject the null hypothesis. Recall from Table 5.2, the correlation between

the return on oil and natural gas is relatively low at approximately 0.1, indicating

71

CHAPTER 5. RESULTS

that we do not have an issue with suppression. A possible explanation to the loss

of significance is the reduction in sample size caused of the aggregation of the

firm-level observations. The regressions that use the equally weighted portfolio

have a substantially higher goodness-of-fit with an adjusted R2 of almost 0.5 com-

pared to approximately 0.25. This is likely due to the reduced noise caused by the

aggregation of the idiosyncratic movements of the individual stocks.

The fixed effect regressions generate identical coefficients as the pooled full-sample

regressions on three decimal points. This indicates that there are no fixed effects

in our sample, i.e. that there are no firm-specific effects to partial out. The aver-

age intercept is 0.001 and not significant at 5% for both fixed effects regressions,

which indeed is very close to the intercept of the corresponding pooled regres-

sions. The maximum absolute value of the individual intercepts is 0.004, but it is

not significant. To formally test if there are any individual fixed effects, we com-

pute F-tests on the effects based on the comparison between the fixed and pooled

models with the null hypothesis that there do not exist individual fixed effects. It

returns a p-value of 0.13, and we cannot reject the null hypothesis. We calculate

the cluster-robust variance-covariance matrix, which generates larger standard er-

rors for all estimated coefficients except natural gas.

72

CHAPTER 5. RESULTS

Tabl

e5.

5:R

egre

ssio

nR

esul

ts:

Hyp

othe

sis

I

Reg

ress

ions

:

Pool

edPo

rtfo

lioFi

xed

Effe

cts

Pool

edPo

rtfo

lioFi

xed

Effe

cts

Sam

ple

Full

Excl

.C

NA

Full

Full

Full

Excl

.C

NA

Full

Full

Mar

ket

0.79

0∗∗∗

0.79

4∗∗∗

0.77

4∗∗∗

0.79

0∗∗∗

0.76

5∗∗∗

0.77

6∗∗∗

0.74

1∗∗∗

0.76

5∗∗∗

(0.0

21)

(0.0

22)

(0.0

45)

(0.0

39)

(0.0

22)

(0.0

23)

(0.0

45)

(0.0

44)

EUA

0.03

6∗∗∗

0.03

8∗∗∗

0.05

4∗∗∗

0.03

6∗∗∗

0.03

4∗∗∗

0.03

5∗∗∗

0.05

0∗∗∗

0.03

4∗∗∗

(0.0

07)

(0.0

07)

(0.0

14)

(0.0

13)

(0.0

07)

(0.0

07)

(0.0

14)

(0.0

12)

Oil

0.03

6∗∗∗

0.02

5∗∗

0.04

7∗0.

036∗

(0.0

11)

(0.0

11)

(0.0

24)

(0.0

18)

Gas

0.00

50.

008

0.01

00.

005

(0.0

08)

(0.0

08)

(0.0

21)

(0.0

06)

Inte

rcep

t0.

001∗∗

0.00

1∗∗∗

0.00

10.

001

0.00

1∗∗

0.00

1∗∗∗

0.00

10.

001∗

(0.0

004)

(0.0

004)

(0.0

01)

(0.0

004)

(0.0

004)

(0.0

004)

(0.0

01)

(0.0

004)

Obs

erva

tion

s4,

732

4,36

836

44,

732

4,73

24,

368

364

4,73

2R2

0.25

00.

259

0.49

40.

250

0.25

10.

260

0.50

00.

252

Adj

uste

dR2

0.24

90.

258

0.49

10.

248

0.25

10.

259

0.49

50.

249

Res

idua

lStd

.Er

ror

0.02

90.

029

0.01

70.

029

0.02

90.

017

FSt

atis

tic

786.

258∗∗∗

761.

575∗∗∗

175.

986∗∗∗

787.

188∗∗∗

396.

643∗∗∗

382.

719∗∗∗

89.7

84∗∗∗

397.

115∗∗∗

Dur

bin

Wat

son

p-va

lue

0.36

10.

337

Not

e:∗ p<

0.1;∗∗

p<0.

05;∗∗∗

p<0.

01

This

tabl

ere

port

sco

effic

ient

san

dst

anda

rder

rors

(in

pare

nthe

ses)

for

Equa

tion

s4.

11,4

.12,

4.13

,4.1

4,an

d4.

15es

tim

ated

aslin

ear

regr

essi

onw

ith

hete

rosc

edas

tici

tyan

dau

toco

rrel

atio

nco

nsis

tent

(HA

C)

stan

dard

erro

rs.

The

depe

nden

tva

riab

leis

the

disa

ggre

gate

stoc

kre

turn

inth

epo

oled

regr

essi

ons

and

the

aggr

egat

ere

turn

inth

eeq

ually

wei

ghte

dpo

rtfo

liore

gres

sion

.Th

ein

depe

nden

tva

riab

les

are:

Mar

ket:

Dow

Jone

sST

OX

XEu

rope

600;

EUA

:Dec

embe

rfu

ture

sco

ntra

ctfo

rEu

rope

anU

nion

Allo

wan

ces;

Oil:

one-

mon

thEu

ro-d

enom

inat

edfu

ture

sco

ntra

ctfo

rB

rent

crud

eoi

l;an

dG

as:

one-

mon

thEu

ro-d

enom

inat

edfu

ture

sco

ntra

ctfo

rT

TFna

tura

lga

s.R

etur

nsar

eca

lcul

ated

onw

eekl

ypr

ice

seri

estr

ansf

orm

edto

its

natu

rall

ogar

ithm

.

73

CHAPTER 5. RESULTS

Subperiod regressions

Next, we test if the regression results are consistent over time by dividing the sam-

ple into annual subperiods beginning with the first week in January and ending

with the last week in December

Table 5.6 presents the regression results for the one-year subperiods. Emission al-

lowance returns appear to have a positive and significant effect on the stock prices

during 2013, 2017 and 2018 at 0.048, 0.084, and 0.053, respectively. The re-

maining years do not produce significant factor coefficients for emission allowance

returns. 2019 stands out as a different year with a factor-beta for the market port-

folio that is exceptionally low at 0.281 compared with other years’ consistently

positive factor betas ranging between 0.766 to 0.843. Moreover, the 2019 regres-

sion’s R2 is considerably lower at 0.043 compared to the range 0.148 to 0.375.

The low explanatory power suggests that something that is not included in the

model is having a strong impact on the sample during the period.

74

CHAPTER 5. RESULTS

Tabl

e5.

6:R

egre

ssio

nR

esul

ts:

One

-yea

rsu

bper

iod

Dep

ende

ntva

riab

le:

Firm

stoc

kre

turn

2013

2014

2015

2016

2017

2018

2019

Mar

ket

0.84

3∗∗∗

0.80

2∗∗∗

0.76

6∗∗∗

0.84

2∗∗∗

0.82

7∗∗∗

0.82

3∗∗∗

0.28

1∗∗∗

(0.0

60)

(0.0

51)

(0.0

49)

(0.0

59)

(0.1

03)

(0.0

51)

(0.0

91)

EUA

0.04

8∗∗∗

0.01

2−

0.00

20.

024

0.08

4∗∗∗

0.05

3∗∗

−0.

010

(0.0

13)

(0.0

12)

(0.0

37)

(0.0

21)

(0.0

20)

(0.0

22)

(0.0

20)

Oil

−0.

041

0.10

9∗∗∗

0.00

30.

025

0.04

80.

062∗∗

0.06

7∗

(0.0

50)

(0.0

41)

(0.0

28)

(0.0

20)

(0.0

32)

(0.0

30)

(0.0

35)

Gas

−0.

152∗∗

0.03

8∗∗

0.07

5∗∗

−0.

025

−0.

037

0.07

0∗∗∗

−0.

004

(0.0

72)

(0.0

18)

(0.0

36)

(0.0

25)

(0.0

29)

(0.0

24)

(0.0

11)

Inte

rcep

t0.

001

0.00

4∗∗∗

−0.

002∗

0.00

01−

0.00

030.

004∗∗∗

0.00

2∗

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

Obs

erva

tion

s66

367

667

668

967

667

667

6R2

0.23

00.

375

0.33

80.

320

0.15

30.

325

0.04

9A

djus

ted

R2

0.22

60.

371

0.33

50.

316

0.14

80.

321

0.04

3R

esid

ualS

td.

Erro

r0.

029

0.02

50.

030

0.03

30.

029

0.02

80.

027

FSt

atis

tic

49.2

18∗∗∗

100.

613∗∗∗

85.8

20∗∗∗

80.4

12∗∗∗

30.3

10∗∗∗

80.8

34∗∗∗

8.60

0∗∗∗

Not

e:∗ p<

0.1;∗∗

p<0.

05;∗∗∗

p<0.

01

This

tabl

ere

port

sco

effic

ient

san

dst

anda

rder

rors

(in

pare

nthe

ses)

for

Equa

tion

s4.

11an

d4.

12es

tim

ated

aslin

ear

regr

essi

onw

ith

het-

eros

ceda

stic

ity

and

auto

corr

elat

ion

cons

iste

nt(H

AC

)st

anda

rder

rors

.Th

ede

pend

ent

vari

able

isth

edi

sagg

rega

test

ock

retu

rn.

The

inde

pen-

dent

vari

able

sar

e:M

arke

t:D

owJo

nes

STO

XX

Euro

pe60

0;EU

A:D

ecem

ber

futu

res

cont

ract

for

Euro

pean

Uni

onA

llow

ance

s;O

il:on

e-m

onth

Euro

-den

omin

ated

futu

res

cont

ract

for

Bre

ntcr

ude

oil;

and

gas:

one-

mon

thEu

ro-d

enom

inat

edfu

ture

sco

ntra

ctfo

rT

TFna

tura

lgas

.R

etur

nsar

eca

lcul

ated

onw

eekl

ypr

ice

seri

estr

ansf

orm

edto

its

natu

rall

ogar

ithm

.

75

CHAPTER 5. RESULTS

To get a better understanding of how 2019 is different from other years, we com-

pute the correlation matrix for the equally weighted portfolio, the Dow Jones

STOXX Europe 600 index, oil, natural gas, and emission allowance price changes

during 2019.

Table 5.7: Correlation matrix of returns for 2019

Portfolio Market Oil Gas EUA

Portfolio 1Market 0.208 1

Oil 0.232 0.501 1Gas -0.068 -0.027 -0.149 1EUA 0.150 0.188 0.061 0.169 1

This table reports the Pearson’s correlation coefficients for the subperiod 2019. The cor-relation coefficients are calculated based on the weekly natural logarithmic returns. Thevariables are the portfolio of electricity generating firms (Portfolio), the market proxy -Dow Jones STOXX Europe 600 (Market), the one-month Brent crude oil (Oil), TTF natu-ral gas contracts (Gas), and the European Union Allowances (EUA).

The correlation between the aggregate returns of the sample companies and Dow

Jones STOXX Europe 600 index is low at 0.208 during 2019 as compared with

0.686 during the full sample period. Figure 5.1 shows a relatively idle price de-

velopment for the market portfolio and a strong price increase for the equally

weighted portfolio. The correlation between the equally weighted portfolio and

oil is substantially stronger at 0.501 in 2019 compared with 0.322 during the full

sample period. However, it is not strong enough to raise concerns regarding sup-

pression.

To include more observations in the regressions, we extend the subperiods to cover

two years but leave 2013 by itself. The last subperiod (2018 - 2019) continues to

be notably different from the previous subperiods. The market coefficient for the

subperiod 2018 to 2019 is 0.609 and significant, which is lower than the other

subperiods that yield coefficients between 0.785 and 0.843. Moreover, the R2 is

once again lower than in the previous subperiods. As such, the reduced sample

76

CHAPTER 5. RESULTS

size of the subperiod 2019 regression does not explain the unexpected results.

Table 5.8: Regression Results: Two-year subperiod

Dependent variable:

Firm stock return2013 2014 - 2015 2016 - 2017 2018 - 2019

Market 0.843∗∗∗ 0.785∗∗∗ 0.826∗∗∗ 0.609∗∗∗

(0.060) (0.035) (0.047) (0.046)

EUA 0.048∗∗∗ 0.012 0.050∗∗∗ 0.028∗

(0.013) (0.012) (0.015) (0.015)

Oil −0.041 0.036 0.034∗ 0.054∗∗

(0.050) (0.022) (0.017) (0.022)

Gas −0.152∗∗ 0.046∗∗∗ −0.031∗ 0.016(0.072) (0.017) (0.018) (0.011)

Intercept 0.001 0.0004 0.0001 0.003∗∗∗

(0.001) (0.001) (0.001) (0.001)

Observations 663 1,352 1,365 1,352R2 0.230 0.347 0.254 0.169Adjusted R2 0.226 0.345 0.252 0.167Residual Std. Error 0.029 0.028 0.031 0.028F Statistic 49.218∗∗∗ 179.168∗∗∗ 116.015∗∗∗ 68.689∗∗∗

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

This table reports coefficients and standard errors (in parentheses) for Equations 4.11 and4.12 estimated as linear regression with heteroscedasticity and autocorrelation consistent(HAC) standard errors. The dependent variable is the disaggregate stock return. Theindependent variables are: Market: Dow Jones STOXX Europe 600; EUA: December fu-tures contract for European Union Allowances; Oil: one-month Euro-denominated futurescontract for Brent crude oil; and gas: one-month Euro-denominated futures contract forTTF natural gas. Returns are calculated on weekly price series transformed to its naturallogarithm.

5.3.2 Hypothesis II

In this section, we present the empirical results related to the second hypothesis,

namely, if a potential impact from price increases in European Union Allowances

77

CHAPTER 5. RESULTS

is stronger on firm stock return for firms with a lower carbon-intensive electricity

generation.

Recall that we create a binary variable “Polluter” that equals 1 if firm i’s carbon

intensity is above the sample median. To test the second hypothesis, we add the

binary variable to Equation 4.16 that controls for commodity price impact and

interact it with the return of emission allowances. This allows for carbon intensive

companies to have a different intercept and slope in regard to emission allowance

return. The results of this regression is presented in Table 5.9.

As hypothesized, the results indicate that there indeed is an interaction effect

between the degree of carbon intensity in electricity production and price changes

in emission allowances. While companies that have a lower than median carbon

intensity have a factor coefficient for emission allowance price changes of 0.051

while their carbon intensive counterparts have a lower 0.051 - 0.038 = 0.013

factor coefficient when taking the interaction effect into account. Interestingly, the

binary variable representing carbon insntive firms is positive, with a coefficient of

0.02, suggesting that carbon intensive companies, ceteris paribus, experienced a

higher average return during the sample period.

The addition of the binary variable for polluters and its interaction term with

emission allowance returns only slightly increases goodness-of-fit measured by the

adjusted R2. The factor betas for the market, emission allowances, natural gas, and

the interaction term Polluter * EUA are all individually significant at 1% level. The

binary variable polluter is significant at 5%, and oil is not significant at 10%. All

reported standard errors are heteroscedastic and autocorrelation consistent.

Next, we investigate the firm specific effects by running individual regressions

for each firm in the sample. Thirteen regressions are presented in Table 5.10.

Each regression has 364 observations, and the adjusted R2 varies between 0.189

78

CHAPTER 5. RESULTS

(CNA) to 0.425 (ENG). The Durbin Watson test for RWE indicates issues with

autocorrelation in its first lag. Three companies have a significant factor coefficient

for emission allowances: Verbund AG (0.138), Electricite de France SA (0.028),

and Fortum Oyj (0.089).

79

CHAPTER 5. RESULTS

Table 5.9: Regression Results: Hypothesis II

Dependent variable:

Firm stock return

(1) (2)

Market 0.765∗∗∗ 0.765∗∗∗

(0.022) (0.022)

EUA 0.034∗∗∗ 0.051∗∗∗

(0.007) (0.009)

Oil 0.005 0.005(0.008) (0.008)

Gas 0.036∗∗∗ 0.036∗∗∗

(0.011) (0.011)

Polluter 0.002∗∗

(0.001)

Polluter * EUA −0.038∗∗∗

(0.013)

Intercept 0.001∗∗ 0.0002(0.0004) (0.001)

Observations 4,732 4,732R2 0.251 0.253Adjusted R2 0.251 0.252Residual Std. Error 0.029 (df = 4727) 0.029 (df = 4725)F Statistic 396.643∗∗∗ (df = 4727) 267.038∗∗∗ (df = 4725)

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

This table reports coefficients and standard errors (in parentheses) for Equations 4.12 and4.16 estimated as linear regression with heteroscedasticity and autocorrelation consistent(HAC) standard errors. The dependent variable is the disaggregate stock return. Theindependent variables are: Market: Dow Jones STOXX Europe 600; EUA: December fu-tures contract for European Union Allowances; Oil: one-month Euro-denominated futurescontract for Brent crude oil; and gas: one-month Euro-denominated futures contract forTTF natural gas. Returns are calculated on weekly price series transformed to its naturallogarithm.

80

CHAPTER 5. RESULTS

Tabl

e5.

10:

Reg

ress

ion

Res

ults

:Fi

rm-s

peci

fic

Dep

ende

ntva

riab

le:

Firm

stoc

kre

turn

SSE

ELE

IBE

VER

NTG

CN

AEN

GED

FED

PA

2AFO

REN

ER

WE

Mar

ket

0.45

8∗∗∗

0.69

8∗∗∗

0.80

2∗∗∗

0.57

0∗∗∗

0.79

6∗∗∗

0.63

3∗∗∗

0.93

9∗∗∗

0.91

6∗∗∗

0.78

2∗∗∗

0.82

8∗∗∗

0.63

5∗∗∗

0.93

1∗∗∗

0.95

5∗∗∗

(0.0

63)

(0.0

72)

(0.0

59)

(0.0

79)

(0.0

70)

(0.0

90)

(0.0

61)

(0.0

85)

(0.0

68)

(0.0

77)

(0.0

80)

(0.0

68)

(0.1

24)

EUA

0.01

70.

009

−0.

007

0.13

8∗∗∗

−0.

011

0.01

40.

026

0.08

7∗∗∗

−0.

001

0.04

00.

089∗∗∗

0.00

10.

036

(0.0

23)

(0.0

18)

(0.0

18)

(0.0

26)

(0.0

22)

(0.0

25)

(0.0

20)

(0.0

28)

(0.0

22)

(0.0

29)

(0.0

22)

(0.0

23)

(0.0

32)

Oil

−0.

001

0.01

2−

0.00

40.

014

0.01

2−

0.03

40.

0001

−0.

040

0.00

10.

025

0.03

30.

019

0.02

8(0

.026

)(0

.019

)(0

.019

)(0

.031

)(0

.029

)(0

.046

)(0

.026

)(0

.034

)(0

.029

)(0

.025

)(0

.023

)(0

.022

)(0

.042

)

Gas

0.10

0∗∗∗

−0.

050

−0.

048∗

0.06

60.

062∗∗

0.16

3∗∗∗

0.01

40.

083∗

0.03

0−

0.04

20.

081∗∗

−0.

041

0.04

5(0

.032

)(0

.030

)(0

.026

)(0

.045

)(0

.029

)(0

.049

)(0

.034

)(0

.050

)(0

.035

)(0

.039

)(0

.039

)(0

.033

)(0

.055

)

Inte

rcep

t0.

001

0.00

3∗∗

0.00

2∗∗

0.00

20.

002

−0.

003∗

−0.

0001

−0.

001

0.00

20.

003∗∗

0.00

20.

002

−0.

001

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

01)

(0.0

02)

Obs

erva

tion

s36

436

436

436

436

436

436

436

436

436

436

436

436

4R2

0.19

20.

268

0.39

20.

216

0.36

60.

198

0.43

10.

261

0.28

80.

244

0.30

80.

401

0.19

9A

djus

ted

R2

0.18

30.

260

0.38

50.

207

0.35

90.

189

0.42

50.

253

0.28

00.

235

0.30

00.

394

0.19

0R

esid

ualS

td.

Erro

r0.

024

0.02

30.

020

0.03

20.

023

0.03

40.

023

0.03

70.

026

0.03

10.

025

0.02

30.

043

FSt

atis

tic

21.3

17∗∗∗

32.9

02∗∗∗

57.9

04∗∗∗

24.7

20∗∗∗

51.8

99∗∗∗

22.2

07∗∗∗

67.9

86∗∗∗

31.6

79∗∗∗

36.3

03∗∗∗

28.9

53∗∗∗

39.9

78∗∗∗

59.9

60∗∗∗

22.2

43∗∗∗

Dur

bin

Wat

son

p-va

lue

0.88

40.

579

0.74

80.

986

0.32

50.

296

0.37

40.

781

0.46

60.

617

0.92

40.

866

0.04

1

Not

e:∗ p<

0.1;∗∗

p<0.

05;∗∗∗

p<0.

01

This

tabl

ere

port

sco

effic

ient

san

dst

anda

rder

rors

(in

pare

nthe

ses)

for

Equa

tion

4.12

esti

mat

edas

linea

rre

gres

sion

wit

hhe

tero

sced

asti

city

and

auto

corr

elat

ion

cons

iste

nt(H

AC

)st

anda

rder

rors

.Th

ede

pend

ent

vari

able

isth

est

ock

retu

rnfo

rth

esp

ecifi

cco

mpa

ny.

The

inde

pend

ent

vari

able

sar

e:M

arke

t:D

owJo

nes

STO

XX

Euro

pe60

0;EU

A:

Dec

embe

rfu

ture

sco

ntra

ctfo

rEu

rope

anU

nion

Allo

wan

ces;

Oil:

one-

mon

thEu

ro-d

enom

inat

edfu

ture

sco

ntra

ctfo

rB

rent

crud

eoi

l;an

dga

s:on

e-m

onth

Euro

-den

omin

ated

futu

res

cont

ract

for

TTF

natu

ralg

as.

Ret

urns

are

calc

ulat

edon

wee

kly

pric

ese

ries

tran

sfor

med

toit

sna

tura

llog

arit

hm.

81

CHAPTER 5. RESULTS

5.3.3 OLS Diagnostics

In this section, we will visually assess if the OLS assumptions hold true for the

models used to test hypotheses I and II.

Model Diagnostics for Hypothesis I

The figure illustrates four plots of the regressions’ residuals for the pooled regression residuals,and these are used to examine if the OLS assumptions hold.

Figure 5.4: OLS Diagnostic - Pooled regression

We plot the diagnostics plots of the extended models that include oil and natu-

ral gas as control variables for both the pooled and aggregated portfolio data to

investigate whether the OLS assumptions hold. The linearity assumption can be

checked by investigating the first plot Residuals vs Fitted. They do not display

any fitted pattern, suggesting that we can assume a linear relationship between

our predictors and the outcome variables. In the Normal Q-Q plot, the residuals

82

CHAPTER 5. RESULTS

The figure illustrates four plots for the regressions’ residuals for the equally weighted portfolio,and these are used to examine if the OLS assumptions hold.

Figure 5.5: OLS Diagnostic - Equally weighted portfolio

should follow the diagonal line to indicate normality of the residuals. In our case,

the residuals approximately follow this line with some divergence in both ends, in

particular for the pooled regression. The Scale-Location plot is used to visually

see if the variances in the residual errors are constant, in other words, it can be

used to identify a heteroscedasticity issue. The horizontal red line indicates that

the variability does not change depending on the size of the fitted values. This sug-

gests that we do not have an issue with heteroscedasticity. The last plot, Residuals

vs Leverage, helps us find influential outliers by plotting the standardized residu-

als and high leverage points. The plot also provides the three most extreme data

points that are particularly worth checking for validity. No observations lie outside

of Cook’s distance, and the plot does not raise concerns regarding outliers. How-

ever, we decide to investigate the extreme data points to confirm the validity of

the data.

83

CHAPTER 5. RESULTS

Table 5.11: Extreme data points for Pooled regression

Observation Company Date rFirm rEUA rMkt rOil rGas

1107 VER 2013-04-19 -9.8% -42.8% -2.6% -3.2% -2.5%2463 ENG 2018-05-11 -0.4% 11.5% 0.7% 3.0% 4.1%4564 ENE 2016-10-07 -13.5% 13.7% -0.6% 6.2% 17.3%

This table reports extreme data points as suggested by the diagnostics plot in Figure 5.5.rFirm is the return series for each firm. rEUA, rMkt, rOil and rGas are the return series foreach independent variable.

Table 5.11 displays the observations that the Residuals vs Leverage plot suggests

could be influential outliers for the pooled regression. As we can see, they corre-

spond to weeks with significant returns but are indeed valid. All things considered,

the models do not seem to violate any of the OLS assumptions.

The OLS diagnostics for the one-year and two-year subperiod regressions can be

found in the appendix. The OLS diagnostic indicates similar results as the total

sample period, and do not seem to violate the OLS assumptions.

Model Diagnostics for Hypothesis II

We plot the diagnostics plots of the model that include a dummy variable for

carbon intensive firms, oil, and natural gas to investigate whether the OLS as-

sumptions hold. The linearity assumption can be controlled by examining the first

plot Residuals vs Fitted. As previous diagnostics, they do not display any fitted

pattern, suggesting a linear relationship between our predictors and outcome vari-

ables. The residuals, in the Normal Q-Q plot, approximately follow the diagonal

line with some divergence in both ends. The Scale-Location plot indicates that

the variability does not change depending on the size of the fitted values. This

suggests that we do not have an issue with heteroscedasticity. The Residuals vs

Leverage indicates that no observations lie outside of Cook’s distance, and the

plot does not raise concerns regarding outliers. The models do not seem to violate

84

CHAPTER 5. RESULTS

The figure illustrates four plots for the regressions’ residuals for the equally weighted portfolio,and these are used to examine if the OLS assumptions hold.

Figure 5.6: OLS Diagnostic - Hypothesis II - Polluter

any of the OLS assumptions. The OLS diagnostic for the firm-specific regression

results can be found in the appendix, and the diagnostic plots do not indicate a

violation of the OLS assumptions.

5.4 Summary of Findings

Our empirical results suggest that European Union Allowance price changes have

a positive impact on the return on the stocks of our sample of thirteen European

electricity-generating firms during the third phase of the EU ETS. In addition to

controlling for the overall return on the stock market, we have controlled for com-

modity price impact by including the return one-month Euro-denominated futures

85

CHAPTER 5. RESULTS

for Brent crude oil and TTF natural gas. The empirical results are consistent across

multiple regression models that use the pooled panel data, aggregated panel data

as well as controlling for firm-specific fixed effects. However, subperiod regres-

sions return ambiguous results with significant positive effects in 2013, 2017 and

2018 but not on 2014, 2015, 2016, and 2019.

Additionally, we have controlled for the relative degree of carbon intensity and

found that carbon intensive firms lose a large share of the positive effect that price

increases in emission allowances have on stock performance. On a firm-specific

level, three companies’ stock returns exhibit being positively influenced by price

increases in emission allowances. These are Verbund AG from Austria, Electricite

de France SA from France, and Fortum Oyj from Finland.

86

Chapter 6: Discussion

In Chapter VI, we interpret the empirical results and discuss and how the findings

relate to previous research, economic theory, and the deduced hypotheses. The

purpose of the chapter ultimately to provide answers to the research questions and

their implications as well to present our suggestions regarding further research.

6.1 EU Allowance Price Changes’ Impact on Finan-

cial Performance

We performed regressions with weekly frequency beginning with the first week of

2013 and ending with the last week of 2019. In addition to the overall stock mar-

ket, we controlled for potential commodity price impact by including price changes

in oil and natural gas futures, in line with the underlying concept of a multifac-

tor model. Moreover, three different econometric methods were employed. The

central finding of the econometric analysis is that the coefficients on the emission

allowances are consistently positive and significant when looking at the whole pe-

riod. In line with our first hypothesis, an increase in emission allowance price is

associated with an appreciation of the stock price of our sample of European elec-

tricity generating firms during the third phase of the EU ETS. As such, investors

appear to expect that future cash flows will increase when emissions are more

87

CHAPTER 6. DISCUSSION

expensive and decrease when they are less costly.

Specifically, if we consider the coefficient from the OLS regression using the pooled

panel data and controlling for oil and natural gas to gauge the magnitude of the

impact, a one percent increase in emission allowance price is associated with a

positive 0.034% return on the stock price. The price of an emission allowance

increased from approximately e6 to e26 during the third phase of the EU ETS,

which corresponds to a 333% increase. If we fit the return into the estimated

regression, the increase in emission prices yielded a positive impact on the stock

market return for the considered companies by an average of 11.3% from the

beginning of 2013 to the end of 2019.

Our methodology was designed based on the studies by Oberndorfer (2009) and

Veith et al. (2009) that investigated the impact of emission allowance prices on

stock performance for European electricity generating firms during the first phase

of the EU ETS that ran through 2005 - 2007. Recall from the literature review,

these studies found a positive relationship between the variables, in line with our

results, and argued that this is primarily due to windfall profits in the electricity

sector. Windfall profits occurred due to the combination of grandfathering, i.e.

free allocation of emission allowances, and a high pass-through rate of the oppor-

tunity cost of the allowances to consumers. The implementation of auctioning as

the default method of allocation for the electricity sector on the onset of the third

phase marked the end of these windfall profits. Consequently, we expected that

the extent of the positive relationship would be weaker than during the first phase

of the EU ETS. Nevertheless, our results indicate that the positive impact of emis-

sion allowances on the stock value of European electricity generating companies

has increased.

The average price of an emission allowance December futures contract during the

third phase of the EU ETS was lower than during the first phase. As such, we can

88


exclude a stronger effect due to a larger nominal base as a reason. Instead, one

possible explanation is that the market has become more efficient as market par-

ticipants have had time to improve their understanding of the interplay between

emission allowance price changes and financial performance. The first phase of

the EU ETS was a pilot stage that, among other things, aimed to build the founda-

tion of the system by trial-and-error and pave the way for its future development.

As a result, there was much uncertainty regarding several components of the sys-

tem, perhaps most clearly illustrated by the plummet in emission allowance spot

price during the end of the first phase. For the investor, such uncertainty would

present challenges in discounting a conceivable effect into the stock price, which

in turn would be reflected in a weak connection.

During the third phase, however, the EU ETS was well-established, and several

mechanisms had been put in place to increase transparency and alignment be-

tween member states. For instance, the single EU-wide emissions cap that replaced

the National Allocation Plans and the implementation of the Market Stability Re-

serve have likely contributed to better aligned expectations and stability among

market participants. Moreover, conditions for the fourth phase that will begin in

2021 and end in 2030 have been disclosed with the most significant difference

likely being the larger linear rate at which the emissions cap decreases. These

things considered, investors are better able to discount the impact of emissions al-

lowance futures price changes to the stock price of European electricity-generating

companies.

Oberndorfer (2009) states that it would be interesting to investigate if his results

hold after auctioning has become the default method of allocation, and argues that

the effect will depend on the pass-through rate and the carbon intensity of the in-

framarginal producers relative to the intensity of the marginal producer. Over the

last decade, there has been a dramatic penetration of renewable energy sources

89


into the European electricity mix, partially driven by technological advancements

in wind power, increased demand for “green energy”, and the increased marginal

generation cost for emitting power plants due to auctioning. This has pushed

the merit curve further to the right, with a larger cluster of carbon-neutral infra-

marginal producers. Because conventional thermal generation remains the typical

marginal producer, the emission allowance cost increases the wholesale electricity

price on the market. As such, when the price of an emission allowance increases,

the larger cluster of inframarginal producers gain higher contribution margins that

are reflected in the overall financial performance of the European electricity sector.

Another structural change to the European electricity sector that may explain part

of the stronger connection is the European Union’s efforts to reduce vertical in-

tegration by further dividing the electricity sector into its three core activities:

generation, transmission, and distribution. This split may have made our sample

of electricity generating companies less diversified and more focused towards elec-

tricity generation than the samples of Oberndorfer (2009) and Veith et al. (2009).

Because generation is the only activity that is subject to the emissions regulation,

the European electricity companies of today may be more exposed to the emission

allowance price changes than they were a decade ago.

Although emission allowance returns exhibit significant explanatory power dur-

ing the full observation period, the results for the subperiods do not consistently

display the corresponding correlation. There is a temporary lack of economic sig-

nificance during 2014 to 2016 and 2019. When we extend the periods to allow

for more observations, 2014 and 2015 as well 2018 and 2019 are not significant.

During 2014 to 2016, prices for emission allowances were considered too low to

have an impact on business-as-usual allowances, and the system faced widespread

criticism (Perino & Willner, 2016). With prices running as low as e3.2 for De-

cember futures contracts, it is possible that investors did not predict an economic

90


impact on the cash flows of electricity-generating firms. In efforts to increase

the emission allowance prices, the European Commission took action to limit the

large oversupply of the total number of allowances in circulation. In 2014, 2015,

and 2016 the European Commission decided to back-load 400, 300 and 200 mil-

lion allowances, respectively, and in 2018 it decided on implementing the Market

Stability Reserve in the subsequent year to ensure that the price of emission al-

lowances would increase. The prices of emission allowance December futures be-

gan to increase dramatically in the second half of 2017, more than one and a half

year before the implementation of the system, indicating that investors expected

that such a system would succeed in increasing emission allowance prices. With

trust amongst market participants that prices would indeed stabilize at a higher

level, investors may have regained its expectations that price changes in emission

allowances would have an economic impact on European electricity-generating

firms.

Nevertheless, emission price returns lost its economic significance on firm stock

performance in 2019. However, the year stands out in the regressions in several

ways. For instance, the market beta from 2013 - 2018 is consistently between

0.766 and 0.843 as is expected by low-risk utility stocks that are less exposed to

economic cycles (Grout & Zalewska, 2006) while it is low at 0.281 during 2019.

Moreover, the adjusted R2 for 2019 is substantially lower at 0.043 compared with

an average of 0.296 during the previous subperiods regressions. Combined, this

indicates that the sample companies’ shareholder return was influenced by some-

thing that was not included in the model.

In 2019, central banks provided an increased stimulus to the financial markets

by increasing liquidity and lowering interest rates (Bell, 2020). Figure 6.1 dis-

plays the decrease in 10-year government bond interest rates in the United States,

Germany, the United Kingdom, Spain, and Italy during 2019. German interest

91


rates are negative, and the remaining European countries’ interest rates reach be-

low 1% during the year. With such unfavourable interest rates, many investors

chose to seek returns from other securities. Investors regard utility stocks as a

relatively defensive investment in the stock market because of the stable nature of

cash flows and low exposure to the business cycle that allows utility companies to

consistently offer large dividends to shareholders. In low-interest environments,

investors may seek the return from utility companies rather than bonds (Huston,

2015). As such, the outperformance of the sample companies over the Dow Jones

Euro STOXX 600 index may be a result of this and obscure the economic impact of

EUA allowances and overall market performance on the stock return of electricity

generating companies.

The figure illustrates the development in interest rates for selected economies for 2019. The figureis based on daily observations for the United States (US), Germany (GR), the United Kingdom(UK), Spain (SP) and Italian (IT) 10 year government bonds. Source: Bloomberg

Figure 6.1: Development for 10-year Government Yields for 2019

During the year, the price of natural gas roughly halved driven by an exceptionally

mild winter that resulted in oversupplies in Europe. Recall from Chapter II that

92


natural gas and coal are substitutes in thermal power generation and that coal is at

least roughly twice as carbon-intensive. The decrease in natural gas prices and the

simultaneous increase in emission prices led to a significant fuel switch from coal

to natural gas during the year (Qin, 2019). In turn, this contributed to a decrease

in power generation from coal by 25% and a fall in emissions from the electricity

sector by 12%, which is likely to be the largest annual fall ever (Buck et al., 2020).

The significant decrease in the carbon intensity of the generation mix during 2019

made the electricity sector much less exposed to emission price changes, possibly

resulting in the loss of economic significance in the regression during that year.

In conclusion, we confirm our first hypothesis, emission allowance price increases

have, on average, positively impacted stock return for our sample of thirteen Eu-

ropean electricity generating firms during the sample period. As such, emission

allowance returns served to expose the firms to systematic risk with an unexpected

increase in the factor resulting in an increase in the stock value. On the premise

of the Efficient Market Hypothesis, these empirical results indicate that investors

discount an expectation that price increases in emission allowances will, on aver-

age, positively impact the electricity generating firms’ future cash flows. As such,

financial performance for European electricity generating firms appear to be pos-

itively affected by price increases in European Union Allowances during the third

phase of the EU ETS.

6.2 Carbon Intensity and Financial Performance

We hypothesized that the return on the stocks of our sample of European electricity-

generating firms with a low carbon-intensive electricity generation was more pos-

itively (negatively) affected by price increases (decreases) of emission allowances

during the third phase of the EU ETS. We formed a binary variable representing

93


firms with a higher than sample median carbon-intensive electricity generation in

Europe. It was interacted with the return on emission allowances to allow for

different intercept and slope regarding the emission allowance factor. The factor

coefficient for the interaction term was significantly negative, indicating that the

electricity generators’ portfolios of power plants affect the relationship between

price changes in emission allowances and stock return.

The empirical results suggest that investors predict that a less carbon-intensive

generation portfolio provides an increased positive effect on future cash flows from

price increases in emission allowances. Intuitively, the reason is that the contri-

bution margins for carbon efficient electricity generators increase when carbon-

intensive price-setting marginal producers must increase their bids to cover the

additional emission compliance costs. The results are consistent with the findings

of Vieth et al. (2009) that found that half of the positive effect from emission al-

lowance price increases were lost for companies with a higher than median share

of carbon emitting production. Verbund AG, Electricite de France SA and Fortum

Oyj are the three companies that individually exhibit a positive and significant

interaction effect between emission price changes and stock performance. Not

surprisingly, these companies are the least carbon-intensive firms in our sample of

thirteen European electricity-generating firms. Their primary sources of electricity

generation are hydroelectric, nuclear, and a combination of both for Verbund AG,

Electricite de France SA and Fortum Oyj, respectively. These technologies gener-

ate electricity without emissions and at relatively low marginal costs and should

benefit from increased costs of emission compliance.

Conclusively, the empirical results confirm the second hypothesis, emission al-

lowance price increases had a larger positive impact on stock return for our sam-

ple of electricity generating firms with a carbon efficient portfolio of power plants

relative to their carbon-intensive peers during the sample period. Hence, carbon

94


efficient firms are more exposed to systematic risk in terms of emission allowance

return. An unexpected increase in the factor will lead to a larger increase in stock

return for the carbon efficient producer than for the carbon intensive generator.

On the premise of the Efficient Market Hypothesis, this indicates that investors

discount an expectation that price increases in emission allowances will have a

larger positive impact on carbon efficient firms’ future cash flows. As such, we

conclude that financial performance is more positively affected by price increases

in emission allowances for carbon efficient electricity generating firms, than their

carbon intensive counterparts. This suggests that the EU ETS is successful in finan-

cially incentivizing profit maximizing firms concerned with electricity generation

to decarbonize their portfolio of power plants.

6.3 Practical Implications

To our knowledge, this paper is the only empirical contribution to the question of

how emission pricing under the European Emission Trading System’s third phase

affects financial performance in the electricity sector. The results indicate that

financial market agents expect emission price changes to have consequences on

future cash flows, reflected in the valuation of the sample of electricity-generating

firms. On the premise of the Arbitrage Pricing Theory, emission allowance return

is a significant factor-beta that offers exposure to systematic risk. Investors can use

this empirical result to find an additional return on their investments and to hedge

positions given expectations of the price developments of emission allowances.

The electricity sector is currently the only sector that is subject to full auctioning,

whereas other sectors receive allowances for free to varying degrees. The results

may offer important insights to policymakers considering enforcing auctioning as

the default method of allocation to other sectors. The electricity sector is naturally

95


protected from international competition by means of infrastructure, regulation,

and limitations in the distance that electricity can be efficiently transmitted. The

results may not apply to other sectors that are more significantly exposed to inter-

national competition, for instance, the manufacturing sector. Implementing full

auctioning on the European manufacturing sector may adversely affect its com-

petitiveness in an international context. Moreover, the ultimate goal of reducing

emissions would likely not be achieved as emissions would simply be transferred if

production is moved to countries that do not impose a cost on emissions. However,

the results may be interesting for other sectors, such as the cement and domestic

aviation sectors. In Europe, cement emits more greenhouse gases than the Belgian

economy and aviation is accountable for 3% of EU greenhouse gas emissions. Ce-

ment is typically produced close to its geographic end-market, and inter-European

aviation is for obvious reasons protected from international competition. As such,

full actioning may be viable for these sectors, and the evidence from the electricity

sector may provide useful insights to European policymakers in the development

of the European Emission Trading System.

6.4 Suggested Further Research

The empirical results of this paper are based on a sample of thirteen publicly

traded electricity companies. As such, the results are not necessarily generalizable

to all European electricity-generating companies, for instance, privately owned

firms. The sample firms have operations outside of Europe that are not subject

to the EU ETS, which has not been accounted for in the econometric analysis. If

researchers could obtain data directly from electricity-generating firms, it could

allow for more specific and in-depth analysis of the interplay between emission

pricing and profitability.

96


On the same topic, a qualitative approach, for instance, using surveys and inter-

views, could provide important insights into how corporate managers in the Euro-

pean electricity sector consider emission allowance in their decision-making pro-

cess. As an example, it would be interesting to learn to what extent are electricity-

generating companies are actively hedging against price fluctuations in emission

allowances.

Next year, the fourth phase of the system will come into effect. The linear re-

duction rate will increase, leading to a more stringent cap. Combined with the

Market Stability Reserve continuing to limit oversupply, many expect that emis-

sion prices will continue to increase. A similar study that investigates the effects

of emission pricing’s impact on firm performance in the European electricity sector

during phase four would be interesting.

The electricity sector is currently the only sector that is subject to full auctioning.

This is due to the relatively low abatement costs and its natural protection from

foreign competition and emission spillover to regions outside of the European

Union. Sectors with higher abatement costs and exposure to international trade,

for instance, the steel sector, risk adversely losing international competitiveness

if they would have to fully compensate for their emissions. However, the cement

and inter-European air travel sectors are naturally protected in similar ways as

the electricity sector, and it would be particularly interesting with studies on how

price changes of emission allowance affect firm performance in these sectors.

97

Chapter 7: Conclusion

The European Union Emissions Trading System (EU ETS) is a policy instrument

that aims to incentivize firms to decarbonize operations by imposing emission

compliance costs through tradeable emission allowances. The electricity sector

is the largest emitter subject to the scheme, and the regulatory impact on the

sector plays a vital role in the overall success of the system.

This paper empirically investigates how price changes in EU ETS emission al-

lowances affect financial performance among European electricity generating firms

and compares the impact depending on the carbon intensity of the firm’s electricity

generation. Based on the Arbitrage Pricing Theory, we test if European electricity

stocks are exposed to systematic risk from unexpected emission price changes. We

utilize the financial stock markets to proxy financial performance on the premise

of the Efficient Market Hypothesis. Using a balanced longitudinal dataset with

weekly frequency on the stock performance of thirteen listed European electricity

generating firms from the beginning of 2013 to the end of 2019, the study empir-

ically tests for a potential impact of price changes on EU ETS emission allowance

December futures contracts on stock return. For robustness, multiple econometric

models are employed using disaggregated pooled returns and aggregated equally

weighted returns as well as controlling for firm-specific fixed effects. Further, we

control for the overall stock market performance by including the returns of the

Dow Jones Stoxx Europe 600 index and commodity price impact by including Euro

98

CHAPTER 7. CONCLUSION

denominated one-month futures contracts for Brent Oil and TTF Natural Gas. A

binary variable representing the firms’ relative carbon intensity to the median in

2015 is included and interacted to isolate its potential effect on the relationship

between firm stock return and EU ETS emission price return.

In line with Oberndorfer (2009) and Veith et al. (2009), the econometric models

find a positive and significant relationship between EU ETS emission price changes

and stock return for the sample of European electricity generating firms during the

sample period. However, the results are not consistent across subperiods. Further,

in line with Veith et al. (2009) the models find that the positive impact is larger

for firms with carbon efficient electricity generation. The empirical results indicate

that European electricity generating firms in general, and carbon efficient firms in

particular, are exposed to systematic risk from EU ETS emission allowances. Fi-

nancial market agents discount a positive impact of an increased EU ETS emission

allowance price on future cash flows of electricity generating firms. The impact

is expected to be larger for firms with carbon efficient operations. As such, we

conclude that there is a positive relationship between EU ETS emission allowance

prices on the financial performance of European electricity generating firms and

that the positive relationship is stronger for firms with carbon efficient operations.

The positive relationship may partially be explained by the high pass-through rate

of the additional emission compliance cost to the bids of the marginal power plant

(Sijm et al., 2006) that raises the wholesale electricity price in equilibrium. In

turn, this leads to increased regulatory rent for the inframarginal suppliers in the

market. Firms with carbon efficient portfolios of power plants have lower emission

compliance costs and therefore reap larger regulatory rents relative to their carbon

intensive peers.

This study contributes important findings to a limited and outdated set of aca-

demic studies and suggests that the EU ETS is indeed successful in financially

99

CHAPTER 7. CONCLUSION

incentivizing profit maximizing firms concerned with electricity generation to de-

carbonize operations. Although the results are not necessarily directly generaliz-

able to other sectors, the results may be of interest for policymakers considering

more stringent emission compliance in other sectors, primarily the cement and

domestic air travel sectors, that are naturally protected from foreign competition.

100

References

Watson, D. (1951). Testing for Serial Correlation in Least Squares Regression, II.

Biometrika., 38.

Treynor, J. L. (1962). Toward a theory of market value of risky assets (Doctoral

dissertation No. 2).

Sharpe, W. F. (1964). A Theory of Market Equilibrium under Conditions of Risk.

The Journal of Finance, 19(3), 425–442.

Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky Invest-

ments in Stock Portfolios and Capital Budgets. The Review of Economics and

Statistics, 47, 13–37.

Fama, E. F. (1970). Efficient Capital Markets : A Review of Theory and Empirical

Work. The Journal of Finance, 25(2), 383–417.

Ross, S. A. (1976). The arbitrage theory of capital asset pricing. Journal of Eco-

nomic Theory, 13(3), 341–360.

Scholes, M., & Williams, J. (1977). Estimating beta. Journal of Financial Economics,

5, 309–327.

Chen, N.-f., Roll, R., & Ross, S. A. (1986). Economic Forces and the Stock Market.

The University of Chicago Press, 59(3), 383–403.

Bode, S. (2006). Multi-period emissions trading in the electricity sector-winners

and losers. Energy Policy, 34(6), 680–691.

101

REFERENCES

Grout, P., & Zalewska, A. (2006). The impact of regulation on market risk. Journal

of Financial Economics, 80(1), 149–184.

Sijm, J., Neuhoff, K., & Chen, Y. (2006). CO2 cost pass-through and windfall profits

in the power sector. Climate Policy, 6(1), 49–72.

Mansanet-Bataller, M., Pardo, A., & Valor, E. (2007). CO2 prices, energy and weather.

Energy Journal, 28(3), 73–92.

Alberola, E., Chevallier, J., & Cheze, B. (2008). Price drivers and structural breaks

in European carbon prices 2005-2007. Energy Policy, 36(2), 787–797.

Zachmann, G., & von Hirschhausen, C. (2008). First evidence of asymmetric cost

pass-through of EU emissions allowances: Examining wholesale electricity

prices in Germany. Economics Letters, 99(3), 465–469.

Oberndorfer, U. (2009). EU Emission Allowances and the stock market: Evidence

from the electricity industry. Ecological Economics, 68(4), 1116–1126.

Veith, S., Werner, J. R., & Zimmermann, J. (2009). Capital market response to

emission rights returns: Evidence from the European power sector. Energy

Economics, 31(4), 605–613.

Bodie, Z., Kane, A., & Marcus, A. J. (2011). Investments (9th ed.). New York, Mc-

Graw Hill.

Rademaekers, K., van de Laan, J., Boeve, S., Lise, W., & Kirchsteiger, C. (2011).

Investment needs for future adaptation measures in EU nuclear power plants

and other electricity generation technologies due to effects of climate change

(tech. rep. March). European Union.

Chevallier, J. (2012). Econometric analysis of carbon markets: The European Union

emissions trading scheme and the clean development mechanism (1st). Springer

Netherlands.

Oberndorfer, U., Alexeeva-Talebi, V., & Loschel, A. (2012). Understanding the

Competitiveness Implications of Future Phases of EU ETS on the Industrial

Sectors. SSRN Electronic Journal, (10).

102

REFERENCES

Rickels, W., Gorlich, D., Oberst, G., & Peterson, S. (2012). Carbon Price Dynamics –

Evidence from Phase II of the European Emission Trading Scheme, (1804).

Aatola, P., Ollikainen, M., & Toppinen, A. (2013). Price determination in the EU

ETS market: Theory and econometric analysis with market fundamentals.

Energy Economics, 36, 380–395.

Bushnell, J. B., Chong, H., & Mansur, E. T. (2013). Profiting from regulation: Evi-

dence from the European carbon market. American Economic Journal: Eco-

nomic Policy, 5(4), 78–106.

European Commission. (2013). COMMISSION DECISION of 5 September 2013

concerning national implementation measures for the transitional free al-

location of greenhouse gas emission allowances in accordance with Article

11(3) of Directive 2003/87/EC of the European Parliament and of the C.

Woolridge, J. (2013). Introductory Econometrics (5th). Mason, South Western.

Cludius, J., Hermann, H., Matthes, F. C., & Graichen, V. (2014). The merit order ef-

fect of wind and photovoltaic electricity generation in Germany 2008-2016

estimation and distributional implications. Energy Economics, 44(2014),

302–313.

Enders, W. (2014). Applied Econometric Time Series (4th). Alabama, John Wiley &

Sons Inc.

Endesa. (2015). 2015 Sustainability Report (tech. rep.).

European Commission. (2015). EU ETS Handbook (1st). European Union.

Huston, J. L. (2015). The Declaration of Dependence: Dividends in the Twenty-First

Century (1st). Archway Publishing.

KU Leuven Energy Institute. (2015). The current electricity market design in Europe

(tech. rep.). KU Leuven Energy Institute.

Stock, J. H., & Watson, M. W. (2015). Introduction to Econometrics (Third). Harlow,

United Kingdom, Pearson Education Limited.

103

REFERENCES

Denny Ellerman, A., Marcantonini, C., & Zaklan, A. (2016). The european union

emissions trading system: Ten years and counting. Review of Environmental

Economics and Policy, 10(1), 89–107.

Erbach, G. (2016). Understanding electricity markets in the EU. European Parli-

mentary Research Service, (November), 10.

Hintermann, B., Peterson, S., & Rickels, W. (2016). Price and market behavior in

phase II of the EU ETS: A review of the literature. Review of Environmental

Economics and Policy, 10(1), 108–128.

Martin, I., & Wagner, C. (2016). What is the Expected Return on a Stock. Retrieved

May 9, 2020, from https://www.aqr.com/About-Us/AQR-Insight-Award/

2017/What-is-the-Expected-Return-on-a-Stock

Perino, G., & Willner, M. (2016). Procrastinating reform: The impact of the market

stability reserve on the EU ETS. Journal of Environmental Economics and

Management, 80, 37–52.

PwC. (2016). Climate Change and Electricity (tech. rep.). PwC.

Berk, J., & DeMarzo, P. (2017). Corporate Finance (4th). Harlow, Pearson.

Centrica. (2017). Carbon Disclosure Project Centrica (tech. rep.).

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2017). An Introduction to Sta-

tistical Learning: with Applications in R (7th). London, Springer.

Marin, G., Marino, M., & Pellegrin, C. (2018). The Impact of the European Emis-

sion Trading Scheme on Multiple Measures of Economic Performance. En-

vironmental and Resource Economics, 71(2), 551–582.

European Commission. (2019a). EU Energy in Figures (1st). Luxembourg.

European Commission. (2019b). Report on the functioning of the European carbon

market (tech. rep.). European Commission.

Qin, B. (2019). EU power: Year in Review (tech. rep.). BloombergNEF.

Bell, M. (2020). Review of markets over 2019 (tech. rep.). J.P.Morgan Asset Man-

agement.

104

https://www.aqr.com/About-Us/AQR-Insight-Award/2017/What-is-the-Expected-Return-on-a-Stock

https://www.aqr.com/About-Us/AQR-Insight-Award/2017/What-is-the-Expected-Return-on-a-Stock

REFERENCES

Buck, M., Redl, C., Hein, F., & Jones, D. (2020). The European Power Sector in 2019

(tech. rep.).

Jones, D. (2020). The Global Electricity Review 2020 (tech. rep.). Ember.

105

Appendix A: Appendix

A.0.1 Autocorrelation and Stationarity

106

APPENDIX A. APPENDIX

107


108


109


Figure A.1: Plots of the logarithmic weekly return and the corresponding ACF plot

110


A.0.2 OLS Diagnostic

Regression Results: One-year subperiod

OLS Diagnostic for Regression Results: One-year subperiod - 2013

111



112



113



114



115



116



Figure A.2: OLS Diagnostic: One-year subperiod

117


Regression Results: Two-year subperiod

OLS Diagnostic for Regression Results: Two-year subperiod - 2014 - 2015

118



119



Figure A.3: OLS Diagnostic: Two-year subperiod

120


Regression Results: Firm-Specific

OLS Diagnostic for Regression Results: Firm-specific - SSE

121


OLS Diagnostic for Regression Results: Firm-specific - ELE

122


OLS Diagnostic for Regression Results: Firm-specific - IBE

123


OLS Diagnostic for Regression Results: Firm-specific - VER

124


OLS Diagnostic for Regression Results: Firm-specific - NTG

125


OLS Diagnostic for Regression Results: Firm-specific - CNA

126


OLS Diagnostic for Regression Results: Firm-specific - ENG

127


OLS Diagnostic for Regression Results: Firm-specific - EDF

128


OLS Diagnostic for Regression Results: Firm-specific - EDP

129


OLS Diagnostic for Regression Results: Firm-specific - A2A

130


OLS Diagnostic for Regression Results: Firm-specific - FOR

131


OLS Diagnostic for Regression Results: Firm-specific - ENE

132


OLS Diagnostic for Regression Results: Firm-specific - RWE

Figure A.4: OLS Diagnostic: Firm-specific

133

THE EU ETS AND FIRM FINANCIAL

Documents