• No results found

The effect of the EU ETS on industrial performance

N/A
N/A
Protected

Academic year: 2021

Share "The effect of the EU ETS on industrial performance"

Copied!
51
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

THE EFFECT OF THE EU ETS ON

INDUSTRIAL PERFORMANCE

Bachelor Thesis Economics

BSc Future Planet Studies: Economics

Author: Pim Krom

Student number: 10547843 Supervisor: E. Jakucionyte

(2)

Abstract

The European Union Emission Trading System (ETS) is the largest international market-based climate policy and is implemented to help the EU in achieving its targets to reduce greenhouse gas emissions. One of the major concerns from the industries about this cap-and-trade system is the loss of competitiveness due to uneven carbon constraints on global markets. The resulting carbon leakage could lead to

ineffectiveness of the scheme and job loss in the energy-intensive sectors. This paper empirically investigates the impact of the ETS on industrial performance by estimating the effects on the production volume and the employment in German industrial sectors that are deemed to be exposed at risk of carbon leakage: the basic metals industry and non-metallic mineral products industry. With a simple time series model the impact of the policy is assessed using years before the launch of the ETS as the

counterfactual period. A second analysis estimates the effect on the production volume using an extended time series and a third analysis compares the industries with two less energy-intensive industries. According to the results of the different models, the effect of the EU ETS on the production volume was predominantly negative, suggesting that the concern from the industries about carbon leakage can be justified. The effect on the employment does not significantly differ from zero and no significant difference is observed with the industries that are less energy-intensive.

Statement of Originality

This document is written by Pim Krom who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Table of Contents

1. Introduction ... 4

2. Literature Review ... 7

3. Data ... 10

4. Methodology ... 15

4.1 Effect on industries deemed to be exposed at risk of carbon leakage (2000q1-2016q2) ... 16

4.2 Extended time series (1994q1-2016q2) ... 18

4.3 Differences across 4 industries (2000q1-2016q2) ... 19

5. Results ... 21

5.1 Effect on industries deemed to be exposed at risk of carbon leakage (2000q1-2016q2) ... 21

5.2 Extended time series (1994q1-2016q2) ... 26

5.3 Differences across 4 industries (2000q1-2016q2) ... 29

6. Conclusion ... 31 7. Discussion ... 32 Bibliography ... 33 Appendix A ... 35 Appendix B ... 39 Appendix C ... 43

(4)

1. Introduction

The current economy is driven by fossil fuels, resulting in wide scale emissions of greenhouse gases. Global warming, a negative external effect of burning fossil fuels, is the largest global public goods problem humanity has ever faced (Weitzman, 2014). Without great reduction of these emissions devastating environmental impacts of the resulting climate change will occur (Avi-yonah & Uhlmann, 2009). International concerns about the dangerous human interference with the climate system has manifested in the foundation of the United Nations Framework Convention on Climate Change (UNFCCC) in 1994. To commit the parties of this convention, internationally binding emission reduction targets are set in an agreement, the Kyoto Protocol. According to the agreement, 37 industrialized countries and the European Community are committed to reduce their greenhouse gas emissions to an average of 5% against the emission levels in 1990 during the first commitment period from 2008 to 2012 (United Nations, 1998). In the second period from 2013 to 2020, the target increased to a reduction of 18% below 1990 levels.

In order to meet the targets of the Kyoto Protocol, the European Union (EU) has implemented a union wide market-based mechanism, the EU Emission Trading System (EU ETS). It is the largest in its kind, covering around 45% of total greenhouse gas emissions in the EU (European Commission, 2017b). 12000 installations in 30 different countries are obliged to participate (Chan, Li, & Zhang, 2013). The EU ETS forms the centrepiece of the EU’s climate policy to reduce its greenhouse gas emissions below 20% of 1990 levels by 2020. With the launch of the program at the beginning of 2005, the European Commission (EC) has set a cap on total greenhouse gases that can be emitted by installations covered by the system (ibid.). This capped amount of emissions is divided into European Union Allowances (EUA’s). Each EUA gives the right to emit one tonne of CO2, which is the most important greenhouse gas. The allowances are partly allocated by the national governments and partly traded at a national auction. The initial allocation of allowances is based on historic data on energy usage. At the end of each year, every participating installation has to surrender an EUA for each tonne of CO2 it emitted. The ETS allows firms to trade in EUA’s, so in case of a deficit they can buy from firms with a surplus of EUA’s or they can trade at the national auction. In this way the ETS has successfully put a price on CO2, internalizing the negative external effect. Due to the additional costs of EUA’s the ETS has created a financial incentive for firms to reduce their emissions. This market-based instrument allows the market to select the most efficient and most innovative reduction techniques and therefore lowers the costs of emission reduction (Avi-yonah & Uhlmann, 2009).

The ETS is divided into certain phases, which are differentiated by the stringency of the rules. Phase I was launched in 2005, followed by phase 2 which started in 2008, simultaneously with the first commitment period of the Kyoto Protocol. The third phase coincides with the second commitment period and is since 2013 in operation. The Commission has already presented a legislative proposal for the fourth phase, which will start in 2020. Phase 1 was a 3-year pilot phase to establish a price for CO2 and obtain reliable data for the allocation in later phases. Almost all allowances were allocated for free during this phase in order to create a gradual transition for the participating installations. In phase 2 the total cap of emissions decreased, the penalties for non-compliance increased and the proportion of free allocation was reduced. The third phase introduced new constraints, such as a yearly decreasing total amount of allowances which makes it possible to reduce the total greenhouse gas emission to 20% below the levels of 1990 by the year 2020 (European Commission, 2017b).

(5)

The EU ETS has proven that putting a price on CO2 can be achieved and the trade in it can work (European Commission, 2017b). Hence, it is seen as an inspiring development for other countries with plans for implementing a climate policy. However, there are some critics who emphasize the weaknesses of the ETS. For example, the trade system has led to the emergence of new forms of derivatives, called carbon derivatives (Chester & Rosewarne, 2011). Chester & Rosewarne (2011) state in their paper that the use of such financial instruments to stabilize the carbon market price is paradoxical, given its destabilizing effects as is occurred during the recent global financial crisis. Avi-Yonah & Uhlmann (2009) praise the

implementation of emission trading systems after decades of inaction, but argue that a carbon tax would be a better approach to reduce greenhouse gas emissions. One of the main concerns for policymakers regarding a cap-and-trade system is the loss of competitiveness due to uneven emission constraints. Energy-intensive firms operating on a global market, especially with homogeneous goods, could experience decreased competitiveness as a result of additional carbon costs (Reinaud, 2008). Their competitors in countries with less stringent climate policies do not face the additional costs of EUA’s and therefore have a cost benefit. The loss of industrial competitiveness could result in carbon leakage. Carbon leakage is the phenomenon that businesses transfer production to other countries with laxer carbon constraints for reasons of costs (European Commission, 2017a). Competitiveness driven carbon leakage can occur via two routes: the loss of market shares to unconstrained competitors and a relocation of production to countries with less stringent climate policies (Reinaud, 2008). Two adverse effects of carbon leakage are the increase in production in non-constrained countries weakening the effectiveness of the policy and the reduction of domestic market shares resulting in job loss and profit reduction (Branger, Quirion, & Chevallier, 2016). Not all firms under the ETS are susceptible to carbon leakage. According to a report from Carbon Trust (2004) there are three factors that determine the potential exposure of an industry to the ETS: the energy intensity, the ability to pass through additional carbon costs and the abatement opportunities. To prevent adverse effects of the ETS, the industries that are deemed to be exposed to a significant risk of carbon leakage receive a higher share of free allowances than other industries (European Union, 2014). The EC has decided which industries could face such a risk in the Carbon Leakage List, that is included in Directive 2014/746/EU (ibid.).

Despite of the mitigating measures from the EC for industries that are susceptible to carbon leakage, there are still complaints about the stringency of the policy. Industries claim that the free allocation of allowances is too low and concerned companies have even brought legal proceedings in several countries, challenging the decision of the EC on the proportion of free allocation (Trounson, 2016). Nevertheless, the European court did not accept the arguments of the objectors and even decided to tighten up the

allocation of free allowances. Because of the policy importance the effects of the ETS on industrial performance have to be monitored constantly. When the ETS does harm the industrial performance, some measures have to be undertaken, such as international agreements, border-tax agreements or output-indexed allocation (Grubb & Neuhoff, 2006). To judge whether the concerns from the industry can be justified, it is interesting to investigate what the impact of the policy has been so far.

In this study, the effect of the EU ETS on the industrial performance is researched. Industrial performance will be measured by means of two variables: production volume and employment. To estimate the effect of the ETS, an empirical analysis will be performed using time series regressions on data from different German industrial sectors. By comparing the years under the ETS with the counterfactual period before the launch of the program the effect can be estimated. The analysis consists of three parts, differing in the time frame and the amount of sectors treated. Analysis 1 investigates the effect of the ETS on production

(6)

volume and employment in the manufacture of basic metals (hereafter basic metals industry) and the manufacture of other non-metallic mineral products (hereafter non-metallic mineral products industry), which are both deemed to be the most susceptible industries to adverse effects of the ETS. The time frame is set at the second quarter of 2001 to the second quarter of 2016. The second analysis evaluates the impacts of the policy on only the production volume in the two industries treated in analysis 1, but covering an extended time series beginning at the third quarter of 1995. Analysis 3 investigates the differences between four sectors: the basic metals industry, the non-metallic mineral products industry, the manufacture of paper and paper products (hereafter paper industry) and the manufacture of wood and of products of wood and cork, except furniture together with the manufacture of articles of straw and plaiting materials (hereafter wood industry). The latter two sectors are both less energy-intensive than the industries from analyses 1 and 2, which allows to investigate whether energy-intensity determines the effect of the ETS on industrial performance. The models used in the last analysis contain less control variables due to absence of data. All data that is used is retrieved form the Eurostat database.

This analysis differs from previous studies in three ways. First, the previous researches use firm level data from different countries, whereas this study analyses the effects on German industries aggregated in sectors classified by the two digit NACE-industry code (rev. 2). Second, other analyses use

non-participating firms as the control group. In this analysis the years prior to the launch of the ETS are used as the counterfactual period to estimate the policy effect. At last, this time series analysis covers a longer period of the trading system, being the first study that researches production volume and employment during the third phase. Therefore, this study provides additional insights into the effects of the EU ETS on industrial performance.

The results from this study differ due to the different models and time series that are used. The first analysis suggests a positive effect of the ETS on the production volume in the basic metals industry, but a negative effect in the non-metallic mineral products industry. On employment there are no significant effects observed for both industries. In the second analysis using an extended time frame, the positive effect on the production volume in the basic metals industry is turned into a negative effect, except for the first phase of the policy. The extended time series model of the non-metallic mineral products industry still exhibits a significant negative effect of the ETS on the production volume. At last, the results of the third analysis show that the more energy-intensive industries are not significantly more severely impacted than less energy-intensive industries. The research partly supports the concern of the industries regarding adverse effects of the ETS.

The remainder of this paper is divided in six sections. Section 2 summarizes the previous literature on this topic, followed by section 3 that explains the data used in the analyses. In the fourth section the

methodology is specified and section 5 contains the results. Section 6 concludes this research and in section 7 the limitations of the analysis and areas for improvement are discussed.

(7)

2. Literature Review

The effect of the Emission Trading System of the EU on the industrial performance has been studied before by various researchers. The studies differ in, for example, the choices in variables, sample selection, time series and control groups. The majority of this literature consists of analyses carried out before the launch of the program to model the impacts of the ETS. These ex ante assessments are hard to perform accurately, as key assumptions need to be made, such as the price of allowances which is dependent on economic development. Ex post assessments on the impact of the EU ETS on economic performance are important for policymakers in their decision whether to engage in similar climate policy or not. Literature on the ex-post evaluation of the EU ETS is still scant, but is rapidly growing. Martin, Muûls, & Wagner (2016) conclude that the main challenge of ex post assessments is to distinguish between the different impacts of the EU ETS on the outcomes of interest. None of the studies is perfect, but contributes to more insight into the reality.

Demailly & Quirion (2006, 2008) did two separate ex ante studies on the effect of the allocation method of the EU ETS on CO2 abatement, competitiveness and carbon leakage. First, the European cement industry was studied on the effect on EBITDA and firm production levels using a trade model, CEMSIM-GEO. This is a model for homogeneous products with high transportation costs. They tested for two different approaches on allocation of allowances; grandfathering and output-based allocation. The results for firms in the case of grandfathering were a significant production loss and CO2 leakage. Cement will benefit from a significant increase of EBITDA under grandfathering. In the case of output-based allocation of 90%, neither the production level nor the EBITDA is significantly affected by the ETS.

In their study on the iron and steel industry, Demailly & Quirion (2008) addressed the same two

dimensions of competitiveness: production and profitability. This time the emphasis was not to evaluate the difference between allocation methods, but to construct a simple and transparent partial equilibrium model by making key assumptions on marginal abatement cost curve, price elasticity of demand, price elasticity of trade, pass-through rates and allocation updating rules. They concluded that for the European iron and steel sector competitiveness losses are small and that the results are robust. The researchers argued that the opposition on more stringent ETS in later phases is not justified by this study.

The first empirical ex post assessment of the competitiveness and employment impacts of the EU ETS is done by Anger & Oberndorfer (2008). Their goal was to measure the impact of relative allocation of EU emission allowances on competitiveness and employment of participating German firms. The relative allocation was measured by the allocation factor, which is calculated as the quotient of allowances allocated to verified emissions. Thus, an allocation factor larger than 1 indicates over-allocation and a factor smaller than 1 means that a firm is under-allocated. The dependent variables to measure

employment were the annual changes in number of employees. To measure competitiveness, Anger and Oberndorfer used yearly changes in revenue as an indicator of the ‘ability to sell’. They used the method of two-stage least squares estimation because of their choice for the allocation factor as an independent variable. This variable may be endogenous due to the fact that its calculation is based on verified

emissions from 2005, which may be determined on the basis of employment and revenue development in that year. Therefore, instruments including more strongly correlating sectoral variables besides the economic variables are used to estimate the allocation factor.

(8)

The main results of their regression is that they do not find empirical evidence for a significant impact of the relative allocation on revenue and employment of German firms. The researchers suggest that this result could be due to the low burden of the policy on participating firms in the beginning stadium of the ETS. The firm-level data used was from 2004 till 2005, so competitiveness effects of the regulation could occur at a later phase. It is clearly that at the time of this publication, much more research was needed to gain ex post insights into the effects of the EU ETS.

A paper by Commins et al. (2009) studied the impact of energy taxes and the EU ETS on European firms from 1996 to 2007. They estimated the effect on technological progress, employment, investment and return to capital. In fact, Commins et al. tried to test the following hypotheses: the pollution heaven hypothesis against a combination of the Porter hypothesis and the Factor Endowment hypothesis. The pollution heaven hypothesis implies that, when facing stringent environmental policy, firms try to reduce costs by locating in countries with less stringent policy, ‘pollution heavens’. The Factor Endowment hypothesis emphasizes the availability of clean technology as factor inputs, which could improve the production. With the Porter hypothesis Commins et al. mean the incentive for firms to innovate as a result of environmental regulation. The pollution heaven hypothesis, also known as ‘carbon leakage’, suggests the deterioration of companies’ performances, whereas the Factor Endowment hypothesis in combination with the Porter hypothesis implies that environmental policy could improve firm

performance in countries with the availability of clean technology.

The result for return to capital was, although insignificant, a positive effect of the ETS. Higher taxes induce an increase in productivity, which supports the Porter hypothesis. The effect on employment was

significantly negative, favouring the pollution heaven hypothesis. Investment and return to capital increased as a result of a higher tax on energy. This suggests that innovation has led to substitution of labour with capital. Since this study only covers the first two years of the EU ETS, the results can only be interpreted as indicative.

Another study on the impact of the EU ETS is done by Abrell et al. (2011). They used panel data from more than 2000 European firms on their emissions and performance from 2005 to 2008 to analyse the

behavioural change of firms between phase I (2005 - 2006) and phase II (2007 – 2008). The major part of this research is about investigating whether firms have reduced their emissions as a result of the

transition from phase 1 to phase 2 or because of changes in the economic environment. In addition, they analysed the effect of initial allocation on CO2 reduction effort of regulated firms. In the last part of their study, Abrell et al. investigated the treatment effect of the EU ETS on companies’ performances. This study did not use the price of allowances (EUA’s) as an explanatory variable, since carbon spot prices in the different phases are incomparable for phase I and II. This is because the price during the first phase was only a short-term signal, as EUA’s were only to be used in the years 2005-2007 and price was a long-term signal during the second phase EUA’s, because they were bankable at least until 2020. Participating installations received the allowances for free during the first trading period and this allocation was too generous. At the end of phase 1 it turned out that the free allocation exceeded the demand. For this reason the price fell to 0 just before the beginning of the second phase, after which the price rose again due to a new allocation with more constraints. This price fluctuation is disturbing in explaining the effect of the policy on companies’ performances.

Abrell et al. studied the effect of the ETS on the value added, the profit margin and employment of participating firms for a panel of European firms. The effect is measured by evaluating the difference with

(9)

firms that are not treated by the ETS. To reduce the bias due to the selection of the non-participating firms, the researchers used a propensity score matching to assign a non-participating firm to each participating firm. The results were not significant: being subject to the ETS did not result in change of value added, profit margin and employment. At the 10 percent significance level there were some interesting results, because there was a difference in the development of profit margins between two subsamples; there was a difference in effects between firms that initially received more allowances than needed (over-allocated) and firms that received less than needed (under-allocated). Over-allocated firms saw their profit margins increase during the period studied and under-allocated firms faced decreasing profit margins. However, the overall conclusion of the analysis was that participating firms did not significantly suffer from a loss of competitiveness with respect to non-participating firms.

Chan, Li & Zhang (2013) used participating and nonparticipating firms of similar sizes within the same industry to create control and treatment groups. They matched firm financial data and emission records to estimate both the effect of emission trading as the effect of initial allocation on firm competitiveness. The fact that the firms from the same industry are chosen as the counterfactual avoids the bias omitted variables including time-invariant industry specific differences. Chan, Li and Zhang studied the effect of the ETS on firm unit material cost, employment and turnover based on a panel of 5873 firms in the electric power, cement and iron and steel industry from 10 different countries during 2005-2009. The results of the impact of the ETS differed across the industries. For the power sector, unit material costs rose as a result of emission trading by 5 percent during 2005-2007 and 8 percent during 2008-2009. This could be explained by the higher costs due to purchasing allowances or by substituting low-cost fuels with expensive but less polluting alternatives. The turnover of the power sector increased on average by 30 percent in phase II, which can reflect the pass-through of compliance costs in the energy price, resulting in higher revenue. On the cement and iron and steel industry, the ETS seems to have no statistically

significant effect on the three variables. The researchers concluded that concerns about carbon leakage in these industries are not supported by this study.

Another study on German firms is done by Petrick & Ulrich (2014). Germany is the largest economy in the EU ETS and a leading exporter of manufactured goods, which makes the country a suitable place to evaluate the effects of the program on the manufacturing sector. An estimation is done on the impact of the ETS on employment, gross output and exports of participating firms. This study also makes use of a propensity score matching to pair treated and control firms. The time span of the research covers phase I and the first half of phase II.

The results prove that the EU ETS had led to a reduction of CO2 emission by 20% by participating firms compared with nonparticipating firms. Moreover, the researchers found that the ETS had no negative effect on gross output, employment or exports during the time period 2005 to 2010. Therefore, Petrich and Ulrich question the legitimacy of the compensation received by industries from the EU Commission for adverse effects on competitiveness.

(10)

3. Data

The dataset that is used in this study consists of macro-economic and industrial data from Germany. All data is obtained from the database of Eurostat, which is the statistical office of the European Union. Eurostat’s key role is to ‘supply the Commission and other European Institutions with data so they can define, implement and analyse Community policies’ (Eurostat). Eurostat does not collect data, but harmonizes it to make it comparable with other countries. The database is comprehensive and contains reliable data of statistical authorities of the Member States. The availability of data from the years before the launch of the ETS is important, because it allows us to compare the years under the policy with a counterfactual period to determine the effect of the program on industrial performance. From the Eurostat database, information is extracted on production volume, employment, producer price, gross domestic product and the real effective exchange rate.

In contrast to previous studies using firm specific panel data, in this research a simple time series model is used with data on industry level. In similar studies, data on participating and non-participating firms is used with the latter group as a counterfactual. Each participating firm is matched to a non-participating firm to estimate the difference of the effect of the policy. The matching procedure of these data could lead to a bias, because non-participating firms have other characteristics than participating firms (Chan et al., 2013). This is likely, since the exclusion of industrial firms is based on a given threshold in, for example, the size of a firm (European Commission, 2017b). In order to reduce the possibility of a difference in the effect due to another characteristic than being subject to the policy, Petrick & Ulrich (2014), Chan et al. (2013) and Abrell et al. (2011) performed a propensity score matching procedure. The propensity score is the probability that a firm is participating in the ETS, given an observable characteristic (Petrick & Ulrich, 2014). The non-participating firm with a score closest to the score of the participating firm is than used in the control group to estimate the effect of the policy. Nonetheless, a bias could be present due to the absence of similar firms. To avoid a possible bias this study uses aggregated industry level data with the years before the launch of the ETS as the counterfactual period.

Germany has been chosen to be the country of interest for a couple of reasons. Firstly, the data of Germany is available for a large time frame on all variables used in this research. The selection of this country allows us to analyse a more comprehensive model, because for other countries data is lacking. Secondly, Germany is the most important country in the ETS, since 24% of all allowances are allocated to German firms (Anger & Oberndorfer, 2008). This means that the analysis on this country can provide us with insights into the effects of the ETS on the entire European industry.

The study evaluates the effect of the EU ETS on the economic performance of several particular industrial sectors. In the first analysis, the effect on the industries of basic metals and non-metallic mineral products is investigated. These sectors are classified by the two-digit code of the NACE industrial classification system (revision 2), that is used in the European Union. Firms operating in these sectors are among the first to experience changes in their economic performance. Of these industrial divisions1, 15 out of 24

1 two-digit NACE rev. 2 code. Manufacture of other non-metallic mineral products is classified with the

(11)

classes2 and 9 out of 16 classes are included in the Carbon Leakage List3, for the non-metallic mineral products industry and basic metals industry respectively. The sectors are relevant in the assessment of this effect, because of two reasons: trade exposure and energy intensity.

Industries that are exposed to international trade, experience more competition when they are subject to the ETS compared with sectors that are less exposed to international trade. For example, the power sector is the most heavily affected by the carbon ETS, but firms in this sector are able to pass through the carbon costs to the consumers in the electricity price (Laing, Sato, Grubb, & Comberti, 2013). This ability to pass through carbon costs is one of the main factors determining the potential impact of the ETS on companies’ competitiveness (CE Delft & Oeko-Institut, 2015). Firms in the power sector have a higher ability to pass through the carbon cost, because of the price inelastic demand and the low trade intensity of electricity (Sato et al., 2007). The sectors of our interest are more trade-exposed and therefore the demand for products is more price elastic (ibid.). Passing through the costs of buying EUA’s or using cleaner but more expensive fuels resulting in a price increase will lead to market share loss. Because adverse impacts on the firm performances in the power sector are less likely than in the sectors in our analysis, the power sector is not addressed by this study. The industry of basic metals and non-metallic mineral products are energy intensive and large contributors to greenhouse gas emission. According to Quirion & Hourcade (2004) the cement industry is one of the most impacted by the ETS due to its second highest CO2 emission/turnover ratio among twelve EUR-15 industries, just after power production. The iron and steel sector is one of the most CO2-intensive industries as well and is relatively open to international trade (Quirion & Hourcade, 2004). Both industries are taken up in the sectors of interest: iron and steel is part of the basic metals industry and cement is classified under non-metallic mineral products industry.

In the third analysis two sectors are added in order to compare the effect with less energy intensive industries. For this reason the paper industry and wood industry are chosen. In figure 1 it can be observed that these industries are less energy-intensive than the industries of basic metals and non-metallic

mineral products. Besides that, the products manufactured in these sectors are relatively homogeneous goods, making them appropriate for a comparison with the products manufactured in the sectors treated in the first two analyses.

2 Subdivisions indicated with a four-digit NACE rev. 2 code

3 The carbon leakage list consists of sectors which are considered by the European Commission to be

(12)

Figure 1. Development of energy intensity in German manufacturing branches, 2000-2013 (Schlomann, Eichhammer, Reuter, Frölich, &

Tariq, 2015).

Concerns of policymakers consist of competitiveness loss and environmental effectiveness loss (Reinaud, 2008). To investigate whether such a trend is present due to the ETS, the effect of the program on employment and production volume is analysed.

Employment

In this analysis the effect of participation in the ETS on the number of employees of industrial firms is analysed. A decline in this number could be the result of relocating by firms out of the EU to avoid compliance costs or the result of decreasing profitability, forcing firms to fire employees. A possible reduction in industrial employment is a major political issue regarding the ETS. It is therefore that most studies investigate the effect on this economic indicator. The data on employment in the industries is retrieved from Eurostat with a quarterly frequency from the first quarter of 2000 till the second quarter of 2016. Thus, the dataset contains 66 observations for employment. The unit of measure of the data is an index value with 2010 as the reference year.

Production volume

The effect of the ETS on this variable is analysed, because it can show whether the German industrial production has declined in volume. This could suggest that polluting production moves to countries without a similar policy. The availability of quarterly indexed data on production volume for the sector of interest was better than for employment, allowing us to cover the first quarter of 1994 to the second quarter of 2016, which means 90 observations.

To estimate the impact of the EU ETS on the indicators of industrial performance, dummy variables are included. For years before the ETS, the counterfactual period, the value of the dummy is 0. For years during the course of the scheme, the treatment period, the dummy value equals 1. To differentiate the three different phases of the policy, additional regressions are run including dummies for each phase separately. These dummies are the variables of interest in this analysis.

(13)

In addition, a variable is included in order to capture the effect of the financial crisis. This is a dummy that equals 1 during the quarters of the trough in the business cycle and 0 otherwise.

To control for variance in the dependent variable that is not caused by the EU ETS, control variables are added. The GDP, proxies for demand, producer price index, lagged values of dependent variables and the real effective exchange rate are added. In the remainder of this section, the control variables will be further explained.

GDP

The Gross Domestic Product is included, because it is the most important and broadest indicator of economic output and growth. The variable will capture the variance due to the fluctuations in economic activity. Quarterly indices (2010=100) on German GDP are retrieved from Eurostat for the period starting at the first quarter of 1994 till the second quarter of 2016. Previous studies have not included the GDP in their models, but these studies estimate the effect by comparing data with a control group that consists of firms that are subject to the same economic environment. Variation in the economic activity is therefore not relevant. Since this research investigates the effect of the implementation of the ETS with the years before its launch as the counterfactual period, changes in the economic environment have to be taken into account.

Production volume of fabricated metal products

This variable is added, because the manufacture of fabricated metal products (except machinery and equipment) industry is an important downstream sector for the basic metals industry. This variable is used as a proxy for the demand for basic metals, because an increase in the production of fabricated metal products will increase the demand for basic metals (Boutabba & Lardic, 2016). The indexed

production values (2010=100) of fabricated metal products for the period 1994 quarter 1 to 2016 quarter 2 are obtained from the Eurostat database.

Production volume in the construction sector

For the same reason as for the latter variable, the production volume in the construction sector is included in the analysis of the non-metallic mineral products industry (Branger et al., 2016). The construction is the most important downstream sector and therefore an increase in this sector will positively affect the production and employment in the non-metallic mineral products industry. The indexed production values (2010=100) of the construction sector for the period 1994 quarter 1 to 2016 quarter 2 are used.

Producer price index

The producer price, also called output price, measures the trading price from the producers view of point (Eurostat, 2017a). This variable is an indicator of international competitiveness because variation in this variable explains variation in demand and therefore the output and employment. The producer price is heavily influenced by input prices and therefore captures variation in for example oil prices (CE Delft & Oeko-Institut, 2015; Durand & Giorno, 1997). Additional carbon costs due to ETS compliance are input prices as well, but since the firms are operating as price takers on an international market, the resulting price increase would be negligible. Previous studies do not use this parameter, as these researches compare results with a control group that is subject to the same fluctuations in input prices. Because this study analyses the changes in employment and production volume after the implementation of a policy, with the years before as the counterfactual period, the variation in output prices have to be taken into

(14)

account.

Since the producer price index is influenced by demand for products manufactured in the industry of interest, it could correlate with the proxies for demand. In the regression analysis this can lead to the problem of multicollinearity and this must be taken into account.

Lagged values of dependent variables

In time series data, the value of the dependent value in one period is typically correlated with its value in the previous period. This is called serial correlation. Therefore, to capture the variation in the dependent variable caused by serial correlation, lagged values are added. The optimal amount of lagged values have to be determined in order to construct the most appropriate model.

Real effective exchange rate index

The real effective exchange rate is included to control for the effect of cost competitiveness on the industrial production. The real effective exchange rate measures the weighted average of a country’s currency value relative to an index of other major currencies, adjusted for the effects of inflation. The weights are determined by comparing the relative trade of a country’s currency against each country in the index (Pilbeam, 2013). The REER is deflated by the price index against a panel of 42 countries: the 28 EU countries, Australia, Canada, United States, Japan, Norway, New Zealand, Mexico, Switzerland, Turkey, Russia, China, Brazil, South Korea and Hong Kong (Eurostat, 2017b). A rise in the REER means a loss of competitiveness.

Meaning of abbreviations:

PV = Production volume E = Employment

ETS = Dummy variable for the total years under EU ETS. Equals 1 for the first quarter of 2005 till the second quarter of 2016 and 0 otherwise ETS1 = Dummy variable for the first phase of the EU ETS. Equals 1 for the

first quarter of 2005 till the fourth quarter of 2007 and 0 otherwise ETS2 = Dummy variable for the second phase of the EU ETS. Equals 1 for the

first quarter of 2008 till the fourth quarter of 2012 and 0 otherwise ETS3 = Dummy variable for the third phase of the EU ETS. Equals 1 for the

first quarter of 2013 till the second quarter of 2016.

CRISIS = Dummy variable. Equals 1 for the first quarter of 2008 till the second quarter of 2009

GDP = Gross Domestic Product

PVF = Production volume of fabricated metal products PVC = Production volume in the construction sector

PP = Producer price

(15)

4. Methodology

In this study, three different time series analyses are performed to estimate the effect of the ETS on industrial performance using the ordinary least squares (OLS) method. In the first model, data between the first quarter of 2000 and the second quarter of 2016 is used to investigate the effect of the EU ETS on the employment and production volume in the basic metals industry and the non-metallic mineral products industry. The second analysis is performed to extend the time series, in order to have a more representative counterfactual period. Only the impact of the ETS on production volume is analysed because data on employment was not available before the year 2000. The data that is used is from the first quarter of 1994 till the second quarter of 2016. The last analysis uses data from 2000 quarter 1 till 2016 quarter 2, to compare the effect of the ETS on employment and production volume across different sectors.

First some general steps are described to test the data on characteristics required for a valid estimation of the effect of the ETS. Thereafter, the specific models for each analysis are given.

As is usual with time series models, the data has to be tested for stationarity. A problem with estimation while using non-stationary time series is that the mean, the variance and the autocorrelation is inconstant over time which can result in invalid estimates from an OLS. Boutabba & Lardic (2016) use the Augmented Dickey Fuller (ADF) test to investigate whether the data has a unit root or is non-stationary. In this

research as well, the null hypothesis whether a unit root is present in the time series data is tested against the alternative hypothesis which implies that the data is stationary.

Before the ADF test can be performed, the optimal amount of lagged values for each variable has to be determined. This is done by selecting the lowest Schwarz’ Bayesian Information Criterion (SBIC) for all variables. Before applying the ADF test, the presence of a trend or a drift is checked in a time series plot, because these properties of the time series data will affect the outcome of the test.

In case of non-stationarity, the level data on the variables is transformed into first difference values. This means that the data consists of differences between current values and the values one time period before. For example, PPt is transformed into DPPt = PPt – PPt-1. In addition, some of the explanatory variables are subject to seasonal fluctuation. When a trend can be distinguished in each separate year, the data exhibit seasonal fluctuation. To take account of these seasonal effects the variables have to be adjusted. If seasonal effects are present, an independent variable can be adjusted for seasonal effects and non-stationarity at the same time by using the difference between the current value and the value in the same quarter a year earlier. For example, GDP is transformed into GDP_SDt = GDPt – GDPt-4.

Again for the transformed variables, the stationarity has to be tested. When stationarity is still not achieved, the next step is taking the second difference of the differenced data. So, the difference between the differenced value at time t and the differenced value one quarter earlier, e.g. GDP_SD is transformed into DGDP_SDt = GDP_SDt - GDP_SDt-1.

The major drawback of differencing is the loss of data points, because with each time a variable is differenced a data point is lost, which makes the time series shorter. Therefore, a new time frame has to be set, beginning at the quarter of the first differenced value of the variable that is transformed most often.

After a certain amount of transformations, all variables are stationary and ready for use in a regression. But as is conventional with time series analyses, serial correlation has to be taken into account. This can

(16)

be done by adding lagged values of the dependent variable to the model. In this analysis, a model is used with the amount of lagged variables that exhibits the highest proportion of explained variance, measured by the adjusted R2. This model is determined by repeatedly running regressions of the dependent variable on all stationary explanatory variables with each time another amount of lagged values of the dependent variable. From all regressions the adjusted R2 is determined to pick the regression with the highest goodness of fit.

The seasonal effect on the dependent variables is addressed by adding seasonal dummies to the model. The fact that quarterly data is used makes it convenient to take account of quarterly seasonal effects. Thus, the seasonal dummies Q1, Q2 and Q3 are added to the model.

At last, the time series models are tested for heteroscedasticity to determine whether robust standard errors have to be applied. Therefore, the Breusch-Pagan test is performed with the null hypothesis of homoscedasticity.

The remainder of this section discusses the methods and models that are used for the three analyses.

4.1 Effect on industries deemed to be exposed at risk of carbon

leakage (2000q1-2016q2)

Basic metals industry

For the data used in this analysis, the results of the ADF test and the selections of the optimal amount of lagged dependent values can be found in appendix A.

The models used to estimate the effect of the ETS on the performance of the basic metals industry including stationary variables and the optimal amount of lags of the dependent variables are:

𝑃𝑉𝑡 = 𝛽0+ 𝛽1𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛽2𝑃𝑉𝐹𝑆𝐷𝑡+ 𝛽3𝐷𝑃𝑃𝑡+ 𝛽4𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛽5𝐸𝑇𝑆𝑡+ 𝛽6𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽7𝑄1𝑡+ 𝛽8𝑄2𝑡+

𝛽9𝑄3𝑡+ 𝛽10𝑃𝑉𝑡−1+ 𝛽11𝑃𝑉𝑡−2+ 𝛽12𝑃𝑉𝑡−3+ 𝛽13𝑃𝑉𝑡−4+ 𝜀𝑡 (1)

𝐸𝑡 = 𝛼0+ 𝛼1𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛼2𝑃𝑉𝐹𝑆𝐷𝑡+ 𝛼3𝐷𝑃𝑃𝑡+ 𝛼4𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛼5𝐸𝑇𝑆𝑡+ 𝛼6𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛼7𝑄1𝑡+ 𝛼8𝑄2𝑡+

𝛼9𝑄3𝑡+ 𝛼10𝐸𝑡−1+ 𝛼11𝐸𝑡−2+ 𝛼12𝐸𝑡−3+ 𝛼13𝐸𝑡−4+ 𝛼14𝐸𝑡−5+ 𝜀𝑡 (2)

𝑋𝑆𝐷𝑡 indicates the seasonal differenced value of variable X with respect to the same quarter a year before, at

time t.

The goal of the regressions is to investigate whether the ETS has a significant effect on the two dependent variables. This effect is estimated with the coefficients of the dummy variable ETS, 𝛽5 and 𝛼5 for model 1 and

model 2 respectively.

The results of the Breusch-Pagan test for heteroscedasticity are: 𝜒2 (13) = 17.67

𝑃𝑟𝑜𝑏 > 𝜒2= 0.1703

for model (1), and for model (2):

𝜒2 (14) = 21.05

(17)

For the 5% significance level the null hypothesis cannot be rejected for both time series models, so we assume that the residuals are homoscedastic. The robust option is not required.

Next to the regression of model 1 and 2 there are several other versions of these models regressed. By analysing models with some variables added or other omitted, the models are checked for possible improvements.

In the second version of the model the dummies for seasonal effects are omitted in order to avoid the ‘dummy-trap’, because the inclusion of many dummies can result in multicollinearity between them. For example, the dummies for seasonal effects can correlate with the crisis dummy.

In the third version of the model without the seasonal dummies, the regression is run with fewer lagged dependent values. Although the optimal amount of lagged values has been determined for each model, the lags could capture a great deal of the variance at the expense of the variable of interest.

Another adjustment to model 1 and 2 is the division of the ETS dummy in ETS1, ETS2 and ETS3 dummies to account for phase 1, 2 and 3 respectively. The reason for this is the difference in severity of the policy during the three phases, especially between phase 1 on the one hand and phase 2 and 3 on the other hand. It could be possible that the effect of ETS1 is insignificant, but the effects of ETS2 and ETS3 are significant. The three versions above (1. including all variables, 2. seasonal dummies omitted and 3. seasonal dummies omitted and fewer lags) have also run been run with 3 ETS-dummies instead of one. The last version of the model takes account of the possible multicollinearity between the proxy for demand (production value in fabricated metal products and in the construction sector) and the producer price. This is done by omitting the proxy for demand, since the producer price index is deemed to be relatively more relevant for the industry of interest. From the six previous versions of the model the most significant regression is selected for this adjustment.

Non-metallic mineral products industry

To estimate the effect of the ETS on the performance of the non-metallic mineral products industry exact the same steps are followed. The results of the ADF tests and selections of the optimal amount of lagged dependent values are listed in appendix A. After differencing for stationarity and addition of the optimal amount of lags two models are constructed:

𝑃𝑉𝑡 = 𝛽0+ 𝛽1𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛽2𝑃𝑉𝐶𝑆𝐷𝑡+ 𝛽3𝐷𝑃𝑃𝑡+ 𝛽4𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛽5𝐸𝑇𝑆1𝑡+ 𝛽6𝐸𝑇𝑆2𝑡+ 𝛽7𝐸𝑇𝑆3𝑡+

𝛽8𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽9𝑄1𝑡+ 𝛽10𝑄2𝑡+ 𝛽11𝑄3𝑡+ 𝛽12𝑃𝑉𝑡−1+ 𝛽13𝑃𝑉𝑡−2+ 𝛽14𝑃𝑉𝑡−3+ 𝛽15𝑃𝑉𝑡−4+ 𝛽16𝑃𝑉𝑡−5+ 𝜀𝑡

(3)

𝐷2𝐸𝑡= 𝛼0+ 𝛼1𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛼2𝑃𝑉𝐶𝑆𝐷𝑡+ 𝛼3𝐷𝑃𝑃𝑡+ 𝛼4𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛼5𝐸𝑇𝑆𝑡+ 𝛼6𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛼7𝑄1𝑡+

𝛼8𝑄2𝑡+ 𝛼9𝑄3𝑡+ 𝛼10𝐷2𝐸𝑡−1+ 𝛼11𝐷2𝐸𝑡−2+ 𝛼12𝐷2𝐸𝑡−3+ 𝜀𝑡 (4)

The results of the Breusch-Pagan test for model 3 are: 𝜒2 (14) = 23.96

𝑃𝑟𝑜𝑏 > 𝜒2= 0.0463

And for model 4:

𝜒2 (12) = 27.30

(18)

For the 5% significance level the null hypothesis has to be rejected for both model 3 and 4, so we assume that the residuals are heteroscedastic. The robust option is therefore required in this case.

For the non-metallic mineral products industry not all the different versions of model 3 and 4 are regressed, as is done for the basic metals industry. From the 7 different versions of the models on the basic metals industry, the regression with the highest proportion of explained variance is chosen for the comparison with the models on the non-metallic mineral products industry.

4.2 Extended time series (1994q1-2016q2)

As a result of differencing, 5 data points are lost in the first analysis, which makes the counterfactual period before the implementation of the ETS 5 quarters shorter. In model 1, 2, 3 and 4 the data before the launch of the trading system is outnumbered by the data treated by the ETS. To bring the data from before the ETS more in balance with the data from the years during the scheme, the time series in the second analysis is started in the first quarter of 1994. Due to a lack of data the regression on employment cannot be performed. The same goes for the data on production price. For these two variables there is no data available for years before 2000.

The same steps as for the first analysis are carried out, but now including time series from the first quarter of 1994 till the second quarter of 2016. The results are attached in appendix B. Ultimately, after

differencing and addition of the optimal amount of lags, the models for both industries are constructed. Basic metals industry

𝑃𝑉𝑡 = 𝛽0+ 𝛽1𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛽2𝑃𝑉𝐹𝑡+ 𝛽3𝑅𝐸𝐸𝑅𝑡+ 𝛽4𝐸𝑇𝑆1𝑡+ 𝛽5𝐸𝑇𝑆2𝑡+ 𝛽6𝐸𝑇𝑆3𝑡+ 𝛽7𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽8𝑄1𝑡+

𝛽9𝑄2𝑡+ 𝛽10𝑄3𝑡+ 𝛽11𝑃𝑉𝑡−1+ 𝛽12𝑃𝑉𝑡−2+ 𝛽13𝑃𝑉𝑡−3+ 𝛽14𝑃𝑉𝑡−4+ 𝛽15𝑃𝑉𝑡−5+ 𝛽16𝑃𝑉𝑡−6+ 𝜀𝑡 (5) Non-metallic mineral products industry

𝑃𝑉𝑡 = 𝛽0+ 𝛽1𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛽2𝑃𝑉𝐹𝑡+ 𝛽3𝑅𝐸𝐸𝑅𝑡+ 𝛽4𝐸𝑇𝑆𝑡+ 𝛽5𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽6𝑄1𝑡+ 𝛽7𝑄2𝑡+ 𝛽8𝑄3𝑡+

𝛽9𝑃𝑉𝑡−1+ 𝛽10𝑃𝑉𝑡−2+ 𝛽11𝑃𝑉𝑡−3+ 𝛽12𝑃𝑉𝑡−4+ 𝛽13𝑃𝑉𝑡−5+ 𝜀𝑡 (6)

The Breusch-Pagan test on heteroscedasticity gives the following results for model 5: 𝜒2 (16) = 63.54

𝑃𝑟𝑜𝑏 > 𝜒2= 0.0000

And for model 6:

𝜒2 (13) = 22.34

𝑃𝑟𝑜𝑏 > 𝜒2= 0.0503

At the 5% significance level the null hypothesis has to be rejected for model 5, so homoscedastic standard errors cannot be assumed. Therefore the option robust is applied when running the regression. For model 6 the null hypothesis cannot be rejected at a significance of 5% and homoscedastic standard errors can be used.

(19)

4.3 Differences across 4 industries (2000q1-2016q2)

With the last analysis a comparison is made between 4 sectors: the basic metals industry, the non-metallic mineral products industry, the paper industry and the wood industry. The latter two sectors are both less energy-intensive than the industries from the first two analyses, which allows to investigate whether less energy-intense industries are differently affected by the ETS than the basic metals industry and non-metallic mineral products industry.

Not for all industries the data used in previous models was available, such as the production volume of downstream industries serving as proxies for the demand. But since the producer price is correlated with the demand as well, the lost explanatory power due to the omission of the proxies for demand is partly captured. Moreover, the data on employment and producer prices was not available for years before 2000, so the time series begins at the first quarter of 2000 and ends at the second quarter of 2016. The results of the ADF tests and selections of the optimal amount of lagged dependent values of all the data from the 4 industries are listed in appendix C.

To determine the optimal models for a comparison, the different versions of models used in analysis 1 are regressed using the data on the basic metals industry again, but now without the demand proxy (PVF). The basic metals industry is chosen because there is more data available and the effect of the ETS is considered to be the most significant on this industry. Again all the variables are transformed into stationary variables and the optimal amount of lagged dependent variables is determined. The results of the 16 regressions can be found in appendix C. For the model of production volume, the regression is chosen with the most significant effects of the ETS dummies. The basic metals industry is considered as the most heavily affected by the ETS, so significant effects of the ETS are important for the comparison with other industries. For the model of employment, the model with the highest adjusted R2 is picked, because there was no significant difference between the effects of the ETS.

The following models are used for the comparison between the industries. Basic metals industry

𝑃𝑉𝑡 = 𝛽0+ 𝛽1𝐸𝑇𝑆1𝑡+ 𝛽2𝐸𝑇𝑆2𝑡+ 𝛽3𝐸𝑇𝑆3𝑡+ 𝛽4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽5𝐷𝑃𝑃𝑡+ 𝛽6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛽7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+

𝛽8𝑄1𝑡+ 𝛽9𝑄2𝑡+ 𝛽10𝑄3𝑡+ 𝛽11𝑃𝑉𝑡−1+ 𝛽12𝑃𝑉𝑡−2+ 𝜀𝑡 (7)

𝐸𝑡 = 𝛼0+ 𝛼1𝐸𝑇𝑆1𝑡+ 𝛼2𝐸𝑇𝑆2𝑡+ 𝛼3𝐸𝑇𝑆3𝑡+ 𝛼4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ +𝛼5𝐷𝑃𝑃𝑡+ 𝛼6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛼7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+

𝛼8𝑄1𝑡+ 𝛼9𝑄2𝑡+ 𝛼10𝑄3𝑡+ 𝛼11𝐸𝑡−1+ 𝛼12𝐸𝑡−2+ 𝛼13𝐸𝑡−3+ 𝜀𝑡 (8)

Non-metallic mineral products industry

𝑃𝑉𝑡 = 𝛽0+ 𝛽1𝐸𝑇𝑆1𝑡+ 𝛽2𝐸𝑇𝑆2𝑡+ 𝛽3𝐸𝑇𝑆3𝑡+ 𝛽4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽5𝐷𝑃𝑃𝑡+ 𝛽6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛽7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+

𝛽8𝑄1𝑡+ 𝛽9𝑄2𝑡+ 𝛽10𝑄3𝑡+ 𝛽11𝑃𝑉𝑡−1+ 𝜀𝑡 (9)

𝐸𝑡 = 𝛼0+ 𝛼1𝐸𝑇𝑆1𝑡+ 𝛼2𝐸𝑇𝑆2𝑡+ 𝛼3𝐸𝑇𝑆3𝑡+ 𝛼4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ +𝛼5𝐷𝑃𝑃𝑡+ 𝛼6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛼7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+

(20)

Paper industry 𝐷𝑃𝑉𝑡= 𝛽0+ 𝛽1𝐸𝑇𝑆1𝑡+ 𝛽2𝐸𝑇𝑆2𝑡+ 𝛽3𝐸𝑇𝑆3𝑡+ 𝛽4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽5𝐷𝑃𝑃𝑡+ 𝛽6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛽7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛽8𝑄1𝑡+ 𝛽9𝑄2𝑡+ 𝛽10𝑄3𝑡+ 𝛽11𝐷𝑃𝑉𝑡−1+ 𝛽12𝐷𝑃𝑉𝑡−2+ 𝜀𝑡 (11) 𝐷𝐸𝑡 = 𝛼0+ 𝛼1𝐸𝑇𝑆1𝑡+ 𝛼2𝐸𝑇𝑆2𝑡+ 𝛼3𝐸𝑇𝑆3𝑡+ 𝛼4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ +𝛼5𝐷𝑃𝑃𝑡+ 𝛼6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛼7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛼8𝑄1𝑡+ 𝛼9𝑄2𝑡+ 𝛼10𝑄3𝑡+ 𝛼11𝐸𝑡−1+ 𝛼12𝐸𝑡−2+ 𝛼13𝐸𝑡−3+ 𝛼14𝐸𝑡−4+ 𝜀𝑡 (12) Wood industry 𝐷𝑃𝑉𝑡= 𝛽0+ 𝛽1𝐸𝑇𝑆1𝑡+ 𝛽2𝐸𝑇𝑆2𝑡+ 𝛽3𝐸𝑇𝑆3𝑡+ 𝛽4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ 𝛽5𝐷𝑃𝑃𝑡+ 𝛽6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛽7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛽8𝑄1𝑡+ 𝛽9𝑄2𝑡+ 𝛽10𝑄3𝑡+ 𝛽11𝐷𝑃𝑉𝑡−1+ 𝛽12𝐷𝑃𝑉𝑡−2+ 𝜀𝑡 (13) 𝐷2𝐸𝑡= 𝛼0+ 𝛼1𝐸𝑇𝑆1𝑡+ 𝛼2𝐸𝑇𝑆2𝑡+ 𝛼3𝐸𝑇𝑆3𝑡+ 𝛼4𝐶𝑅𝐼𝑆𝐼𝑆𝑡+ +𝛼5𝐷𝑃𝑃𝑡+ 𝛼6𝐷𝑅𝐸𝐸𝑅𝑡+ 𝛼7𝐷𝐺𝐷𝑃𝑆𝐷𝑡+ 𝛼8𝑄1𝑡+ 𝛼9𝑄2𝑡+ 𝛼10𝑄3𝑡+ 𝛼11𝐷2𝐸𝑡−1+ 𝛼12𝐷2𝐸𝑡−2+ 𝛼13𝐷2𝐸𝑡−3+ 𝜀𝑡 (14)

(21)

5. Results

5.1 Effect on industries deemed to be exposed at risk of carbon

leakage (2000q1-2016q2)

The first analysis investigates the effect of the ETS on the production volume and the employment in the basic metals industry and the non-metallic mineral products industry. The regression output can be found after the report of the results of each analysis.

Basic Metals Industry

In table 1 the output of all the regressions on the production volume of the basic metals industry is listed. According to the results of model 1, the implementation of the EU ETS has had a positive effect on the production volume in the basic metals industry during the first and the third phase of the trading scheme. When the entire period is investigated, no significant impact is shown.

The positive sign of the insignificant coefficient of the ETS-dummy in regression 1 does not correspond with the economic intuition, because the ETS is expected to be rather harmful for the industrial production than beneficial. Apparently, the policy does not constrain the industry to an extent that adversely impacts the production volume.

The dummy variable for the financial crisis has indeed a significant negative effect on the production value, as was expected. The production in the downstream industry has a significant positive effect on the production value, but the negative effect of the producer price is contrary to the expectations as higher prices should decrease demand and therefore the output. The sign of the GDP is corresponding with the economic theory, since a decrease in GDP reduces the overall economic activity. Though the effect of the real effective exchange rate is insignificant, the positive sign of its coefficient goes against economic theory, because a rise in the REER means a loss of competitiveness.

The regression with the highest explanatory power is number 4. ETS1 and ETS3 do significantly affect the production volume at the 10% level, but the effect of ETS2 is insignificant. The positive effects of the ETS dummies could be explained by a too generous initial allocation of free EUA’s by the EC. In this way, the firms do not have to buy allowances and could even increase profits by a slight price increase due to increased opportunity costs.

Although they are insignificant, the effects of the crisis-dummy, REER and GDP in regression 4 do not have the signs that would be expected according to economic intuition. However, the coefficients of the proxy for demand and the producer price are significant at the 5% level and meet the expectations.

Table 2 gives the results of the regressions of model 2 on employment. None of the ETS-dummies do significantly differ from zero, so no effect on employment could be observed during the time series. In contrast to the effect on production volume, the coefficient 𝛼5 in the first regression has a negative

sign which corresponds with the intuition.

The crisis-dummy is significant, though it has the wrong sign. The coefficient of the production in the downstream industry is significant with the expected sign, but the rest of the control variables do not have a significant effect.

The fourth regression on employment exhibits the highest explanatory power and has three ETS-dummies with negative signs. The coefficients of the crisis-dummy and the proxy for demand are significant, but the crisis-dummy has a counterintuitive sign. The producer price and GDP have signs that meet the

(22)

expectation, but are insignificant. The REER has an insignificant effect, which cannot be explained by economic reasoning.

The R2 of 0.9409 and 0.9493 of the first and fourth regression in model 1 mean that 94.09% and 94.93% of the variation in respectively production volume and employment is explained by the variation in the variables used in the models. In model 2 the R2 of 0.9888 for the first and 0.9890 for the fourth regression show even higher explained variance in the model. These large measurements of the goodness of fit show a high proportion of the variance that is explained by the model.

Non-metallic Minerals Industry

In contrast to the basic metals industry in table 1, the results of models 3 suggest that the production volume in the non-metallic mineral products industry are negatively impacted by the scheme. All the phases are significant and only phase 1 not at the 5% level, which meets the expectations since this phase was less stringent. It is possible that the implementation of the ETS has led to a decrease in

competitiveness of the German firms operating in this industry, resulting in this negative effect. However insignificant, the coefficients of the crisis-dummy, the producer price and GDP have

counterintuitive signs. The production in construction has a significant positive impact, with a sign that meets the expectations. The insignificant coefficient of the REER is in correspondence with the economic theory.

From the results on employment in the non-metallic mineral products industry no significant effect can be observed. Although all control variables have signs that correspond with economic theory, only two are significant.

Again, the adjusted R2 of 0.9474 for the production volume and 0.8647 for the employment indicate a high goodness of fit for these models.

From the comparison between the two industries regarding the production volume and employment it can be concluded that the production in the non-metallic mineral products industry is more severely impacted by the policy than the basic metals industry for the time series 2001q2-2016q2. The effect of the ETS on employment cannot be proved for both industries.

(23)

Table 1

Model 1. Basic Metals Industry. 2001q2-2016q2.

Dep. Var OLS

PV

1

2

3

4

5

5

7

ETS 0.5224 (0.685) 0.4804 (0.823) 2.8919** (1.325) ETS1 1.5038* (0.894) 1.9291* (1.075) 5.5378** (1.692) 1.0548 (1.290) ETS2 -0.9648 (0.794) -1.2572 (0.941 -0.2446 (1.551) 0.8496 (1.088) ETS3 1.3718* (0.776) 1.4080 (0.943) 3.8158** (1.529) 2.4415** (1.101) CRISIS -2.8075** (1.358) -2.0576 (1.599) 1.0547 (2.634) 0.5286 (1.700) 1.8607 (1.965) 7.2907** (3.079) -5.3514** (2.152) PVF_SD 0.4773** (0.076) 0.6581** (0.081) 0.2531** (.1141) 0.5318** (0.074) 0.7114** (0.079) 0.3886** (0.116) DPP 0.6038** (0.128) 0.5810** (0.151) .7561** (0.253) 0.5872** (0.134) 0.5468** (0.157) 0.6251** (0.264) 1.0196** (0.173) DREER 0.0930 (0.236) 0.1240 (0.277) 0.6600 (0.452) -0.0069 (0.221) 0.0345 (0.260) 0.4981 (0.421) .1678737 DGDP_SD -1.6165* (0.916) -2.2177** (1.103) -0.6013 (1.776) -0.9592 (0.875) -1.3914 (1.059) 0.2437 (1.670) -1.9605 (1.249) Q1 3.3657** (0.952) 3.0193** (0.890) 5.4791** (1.187) Q2 -0.9478 (1.080 -1.3135 (1.012) -1.3400 (1.463) Q3 1.3764 (1.218) -1.1466 (1.143) -1.5810 (1.651) L1PV 0.5877** (0.111) 0.3173** (0.093) 0.6036** (0.132) 0.5380** (0.105) 0.2502** (0.089) 0.4899** (0.126) 0.9924** (0.120) L2PV -0.1575 (0.153) -0.0917 (0.108) -0.0719 (0.128) -0.1952 (0.144) -0.0916 (0.102) -0.1608 (0.125) -.2609 (0.208) L3PV -0.086 (0.140) 0.0093 (0.101) -0.0943 (0.130) -0.0460 (0.095) 0.0003 (0.187) L4PV 0.5490** (0.096) 0.6086** (0.085) 0.5389** (0.090) 0.5966** (0.079) 0.2844** (0.119) R2 0.9537 0.9274 0.7841 0.9620 0.9399 0.8232 0.9188 Adj. R2 0.9409 0.9129 0.7509 0.9493 0.9249 0.7878 0.8941 No obs. 61 61 61 61 61 61 61 Notes: Significance: * at 5% and ** at 10%.

(24)

Table 2

Model 2. Basic Metals Industry. 2001q2-2016q2.

Dep. Var. OLS

E

1

2

3

4

5

5

7

ETS -0.2503 (0.189) -0.5050** (0.230) 0.0683 (0.327) -0.1695 (0.261) ETS1 -0.1910 (0.206) -0.3911 (0.254) 0.1848 (0.365) ETS2 -0.2744 (0.211) -0.5353** (0.254) 0.0735 (0.375) ETS3 -0.2960 (0.206) -0.5826** (0.251) -0.0479 (0.364) CRISIS 0.6580** (0.208) 0.8282** (0.252) 0.6013 (0.386) 0.6865** (0.236) 0.8677** (0.282) 0.5930 (0.428) 0.0303 (0.257) PVF_SD 0.0730** (0.011) 0.0844** (0.011) 0.0598* (0.017) 0.0742** (0.011) 0.0863** (0.012) 0.0622** (0.017) DPP -0.0094 (0.0203) -0.0326 (0.026) -0.0269 (0.040) -0.0155 (0.022) -0.0432 (0.028) -0.0408 (0.044) 0.0405 (0.026) DREER 0.0140 (0.037) 0.0003 (0.046) 0.0358 (0.070) 0.0109 (0.039) -0.0041 (0.047) 0.0348 (0.072) 0.0173 (0.051) DGDP_SD 0.2240 (0.151) 0.3905** (0.189) 0.3483 (0.278) 0.2199 (0.156) 0.3765* (0.193) 0.3183 (0.284) -0.0724 (0.200) Q1 -0.9497** (0.219) -0.9342** (0.2236) -0.8445** (0.303) Q2 -0.7347** (0.299) -0.7253** (0.305) 0.1509 (0.371) Q3 0.4149* (0.218) 0.4087* (0.222) 1.2223** (0.251) L1E 0.7738** (0.132) 0.9463** (0.106) 1.1982* (0.135) 0.7627** (0.134) 0.9223** (0.109) 1.1691* (0.141) 1.3420** (0.139) L2E 0.4107* (0.180) -0.1772 (0.015) -0.2375 (0.140) 0.4133** (0.183) -0.1558 (0.167) -0.2097 (0.146) 0.0184 (0.236) L3E 0.0173 (0.178) 0.1599** (0.165) 0.0167 (0.181) 0.1530 (0.166) -0.3607 (0.234) L4E -0.1290 (0.184) 0.5934** (0.165) -0.1122 (0.181) 0.6041** (0.168) -0.2712 (0.253) L5E -0.1732 (0.116) -0.6727** (0.094) -0.1825 (0.119) -0.6752** (0.095) 0.1774 (0.143) R2 0.9888 0.9800 0.9467 0.9890 0.9805 0.9474 0.9780 Adj. R2 0.9854 0.9755 0.9386 0.9850 0.9751 0.9369 0.9719 No obs. 61 61 61 61 61 61 61 Notes: Significance: * at 5% and ** at 10%.

(25)

Table 3

Model 3 and 4. Non-Metallic Mineral Products Industry. 2001q2-2016q2

Dep. Var OLS Dep. Var OLS

PV

(robust)

D2E

(robust)

ETS -0.4190 (0.408) ETS1 -2.4431* (1.245) ETS2 -3.8416** (1.261) ETS3 -2.2350** (0.877) CRISIS -2.3894 (1.666) CRISIS -0.8663** (0.331) PVC_SD 0.4199** (0.089) PVC_SD 0.0903** (0.027) DPP 0.8304 (1.329) DPP -0.1665 (0.333) DREER -0.1896 (0.287) DREER -0.093 (0.089) DGDP_SD -2.0896 (1.311) DGDP_SD 0.6038 (0.491) Q1 -10.8442** (6.271) Q1 0.0883 (0.506) Q2 9.5217 (7.878) Q2 2.1898** (0.655) Q3 8.5388 (5.154) Q3 1.7176** (0.476) L1PV 0.4487** (0.170) L1D2E -0.7637** (0.101) L2PV 0.0684 (0.146) L2D2E -0.5171** (0.106) L3PV 0.0823 (0.171) L3D2E -0.5139** (0.118) L4PV 0.2880 (0.176) L5PV -0.2521** (0.120) R2 0.9614 R2 0.8918 Adj. R2 0.9474 Adj. R2 0.8647 No obs. 61 No obs. 61 Notes: Significance: * at 5% and ** at 10%. The standard errors are reported in brackets.

(26)

5.2 Extended time series (1994q1-2016q2)

As a result of the differencing to transform the non-stationary level data into stationary data there are data points lost. The counterfactual years under absence of the ETS were already underrepresented in the time series of model 1 and 2. After the loss of 5 data points the extension of the time series is desirable, but as mentioned in the methodology, the availability of data on employment is a limiting factor. Accordingly, a regression on solely the production volume was run using a time frame set at the first quarter of 1994.

Basic Metals Industry

In table 4 the output of the regressions of the basic metals industry are listed. The results suggest that phase 1 has positively affected the production volume, whereas phase 2 and 3 have had a negative impact.

Even as in the first analysis, the entire trading period in regression 1 does not significantly differ from zero. The effect is however negative, as economic intuition suggests. The insignificant effect of the entire ETS could suggest that the separate phases offset each other.

The crisis-dummy and proxy for demand have an economically reasonable sign, but only the proxy for demand is significant at the 10% level. The REER and GDP have both insignificant effects with

counterintuitive signs.

In contrast to the first regression, the regression 4 does show significant effects of the three

ETS-dummies. Contrary to expectation, the coefficient of the first phase is positive, but the impact of the last two phases of the ETS have a high negative and significant effect, which corresponds with the difference in stringency of the phases. The results indicate that the production volume was only negatively affected by the ETS when the policy became more stringent in the second phase. Competitiveness loss could have impacted the demand for German products in the industry, resulting in decreased production volume. However, the effect of the crisis-dummy is positive and significant, which is not expected from economic theory. Furthermore, the proxy for demand exhibits an expected positive effect that is significant. Both the coefficients of the REER and GDP are insignificant, but the sign of the REER is counterintuitive. The goodness of fit measured by a R2 of 0.8685 and 0.9493 for respectively regression 1 and 4 shows a high proportion of explained variance for the model.

Non-Metallic Mineral Products Industry

Compared with the fourth regression on the basic metals industry, the results of the model of non-metallic mineral products industry exhibits significant but smaller impacts of the ETS on the production volume, but now the first phase of the ETS has a negative sign as well. The basic metals industry could have benefited more from the generous free allocation of allowances than the non-metallic mineral products industry in the first trading period of the ETS, which could explain the difference in the coefficient of the first ETS-dummy.

The financial crisis and the production in the construction sector have significant impacts that are economically reasonable. The negative effect of the REER does not significantly differ from zero, whereas GDP does but with a counterintuitive sign.

Even as model 5, the regression on model 6 exhibits a high adjusted R2 of 0.9520.

Compared to the model of the non-metallic mineral products industry the model of basic metals industry shows much larger effects on the production volume in this time series, except for the first phase.

Referenties

GERELATEERDE DOCUMENTEN

This study population had a low consumption of red and organ meat (total animal protein intake in cases accounted for <31 g/d and <2 % of total energy intake) and may

Although this approach only leads to a lower bound for the critical temperature of the isotropic 3D Ising model, it gives an asymptotically exact result for the anisotropic

peu de sable gris, avec rognons de silex dans la masse; vers le bas, la charge sableuse augmente, parfois sous forme de Jentilles ondulées, tandis que

Other family names that Julaiga Gomez can remember in the Constantia community included “the Langeveldts, the Bowmans, and the Adams’s.” 18 They were part of the original families

Maar één welbepaalde maat lijkt wel bijzonder geschikt om de complexiteit van open systemen te karakterizeren aan de hand van de vrije energie die ze verwerven,

The Action Plan enumerates priority actions that must be attended to: (1) protecting the habitats of the Spanish Imperial Eagle through natural resource management

Others stressed the importance of education of parents and guardians to be able to disclose to their children, “It is really helpful to educate parents, we are