Increasing renewable energy sources: The influence on European firm performance.

(1)

Increasing renewable energy sources:

The influence on European firm performance.

Master thesis, MSc. Finance

University of Groningen, Faculty of Economics and Business

June 4, 2020

Peter Sven Smilda S2776405 Saffierstraat 102 9743LK Groningen Tel: +31 654631361 E-Mail: p.s.smilda@student.rug.nl Supervisor: D.M. (Daniel) te Kaat, PhD. Key words:

(2)

1. INTRODUCTION

Until the beginning of the current century, almost no growth in the generation of renewable energy sources took place. Nowadays, as climate change is a topic to be concerned about, the situation has changed. As carbon emissions contribute to global warming, actions are being taken to reduce the amount of carbon emission. Carbon emissions from burning fossil fuels contribute to 76% of the total carbon dioxide emissions in the United States (U.S. Environmental protection agency, 2019), therefore a lot of measures are already taken in the energy sector. In this sector a shift towards replacing fossil fuels with renewable energy sources (RES) is taking place. In the last years, the overall share of electricity generated by RES increased from 8.5% in 2004 to 18% in 2018 (Eurostat, 2020), as is represented by figure 1. A main stimulating factor behind this shift are the policy-induced support schemes implemented by governments to reach the goals of the Paris agreement. Therefore, it is relevant to look at the consequences of these shifts to gain more knowledge about the consequences of governmental instruments. A study that investigated the economic consequences of the movement towards more RES has been performed by Jaraite, Karimu & Kazukauskasa (2017). This study found that the policy-induced expansion of RES promotes economic growth in the short-run, but leads to negative growth in the long-term. Our study will extend this research investigating the effects of the growth in solar and wind generation capacity at the firm-level. Using a panel-data set consisting of stock information on 374 firms from seven different European countries, a negative relation between the generation capacity of RES and the stock returns of electricity-generating firms is found. Furthermore, this study performs an explorative analysis on possible mechanisms behind the mentioned negative relationship. This is done by exploring the possible interaction effects of energy prices and its volatility on the relationship between the generation capacity of RES and firm-performance. This study contributes to existing literature by providing a sector-focused analysis of the firm-level effects caused by renewable policy.

(3)

University of Groningen | Peter Sven Smilda 2

As indicated above, the thesis will focus on the effect that the increase in the electrical generating capacity of RES may have on the financial performance of European firms. A main driving factor behind this increase are several different support schemes which are set up by European governments. As these support schemes are designed to increase the amount of renewable energy generation, they cannot be seen separately from this study. The most popular support scheme in the European Union is the so-called feed-in tariff (Ropenus & Jensen, 2009), accounting for the biggest share of renewable energy supporting (European Commission, 2017). According to Alizamir, Véricourt and Sun (2016), the feed-in tariff (FIT) allows governments to attract investments and stimulate demand for the technology by sponsoring a certain compensation level for purchasing electricity from those who have adopted the technology. Although these support schemes will impact the firm performance of electricity-generating firms and create different circumstances for various firms (Jaraite & Kažukauskas, 2013), the influence of these measurements is constant over time. Therefore, we assume that it will not be impacting the relationship that is examined in this research.

Besides the value of this study to policy makers, new insights into the drivers behind firm performance in certain sectors could be valuable to investors. By providing a broad view on the stock returns of energy-related firms, this research could also contribute by promoting investment in energy-related projects. In order to be able to explain the possible outcomes of this research, various drivers explaining the firm-level consequences of increased production capacity of RES will be highlighted. As will be explained in the upcoming part of this study, the generation capacity of RES is suggested to impact the energy prices as well as the volatility and securitization of the energy market. As it is noted by Ohler & Fetters (2014) that limited research is conducted investigating the connection between different RES and economic performance, the analysis of these drivers is relatively new and should, therefore, be regarded as being explorative. In order to guide the choices made in this study, an overview of the literature regarding this topic will be discussed before the research question is addressed.

2. LITERATURE REVIEW

(4)

2.1 Impact of increasing RES capacity.

When looking at the possible effects that an increase in the capacity of RES may cause, we will, at first, consider the paper of Ohler and Fetter (2014). This study looked after the long-term effects of an increase of various RES and found that this has a positive impact on GDP. This is partly contrary to the later findings of Jaraite, Karimu & Kazukauskasa (2017). They found, in a study looking at the long- and short-run economic effects of increasing generation capacity of wind and solar energy in the European Union, a negative economic effect in the long-run. Although they find a positive effect in the short run, they conclude that an increase in capacity does not stimulate economic growth on a long-term basis. These mixed findings could possibly be explained by different methodology, RES, countries, and, the time period that were being analysed by both papers. The paper of Ohler and Fetter (2014) looked at the generation of six RES (biomass, geothermal, hydroelectric, solar, waste, and, wind) in 20 OECD countries over the time period 1998-2008, whereas Jaraite, Karimu & Kazukauskasa (2017) looked at the generation capacity of two RES (solar and wind) in 15 EU-countries over a longer time (1990-2013). As is suggested by Jaraite, Karimu & Kazukauskasa (2017) generating capacity does not suffer from the high seasonal volatility of RES (the sun shines less in the winter, for example), making it less effective for fluctuations caused by external variables. Next to this, analysing European countries would be preferred because of the consistent energy support policies in the European Union, making the analysis less prone to changes in the underlying governmental support of a country. Based on these arguments, the current study will take firms into account from European countries and analyse the generation capacity rather than the total generation. As this study will, due to the availability of data, also consider solar and wind sources, it will be most comparable to the paper of Jaraite, Karimu & Kazukauskasa (2017). This would suggest that negative long-term effects of increasing solar and wind generation capacity on economic growth are to be expected.

(5)

have higher marginal costs and, therefore, creating lower marginal costs of generating energy (Figueiredo et al., 2015). These lower production costs would allow the electricity price to decrease. Another effect regarding electricity prices that can be noted in the literature is the intermittency effect. This effect is created by the intermittent generation pattern of solar and wind sources. As described by Wiser et al. (2017), problems may arise by the fact that wind speed, for example, is stronger at night, when electricity demand is lowest. This may be related to more price volatility and lower average prices in the short run when the relative amount of wind and solar generation sources increase. This is confirmed by the research of Johnson & Oliver (2019) which states that increased wind and solar generation can be linked to higher price risks in the electricity market. Both the merit-order effect and the intermittency effect are consequences of a relative increase of RES in the generation mix (the distribution of generation methods), rather than an absolute increase. Based on these findings it is to be expected that a relative increase of RES in the electricity mix will lead to lower electricity prices and more volatility in the electricity price. Summarizing, we expect that, based on previous literature, an increase in the solar and wind generation capacity will harm the economic growth in the long-term. Next to this, a lower electricity price and more volatility in the electricity prices is expected to be caused by the increase in RES. In the following part, we will analyse how these effects may impact the performance of various firms.

2.2 Impact on firm performance.

First, we will look at the effects that the expected negative economic growth in the long-term will have on firm performance. In the studies mentioned, economic growth is measured by looking at the aggregate output of a certain country. Logically, a decrease of the aggregate output in a country will affect the performance of a certain firm in that country negatively (due to lower demand, e.g.). However, based on Jaraite, Karimu & Kazukauskasa (2017), this negative economic effect is only expected in the long term. Considering that our study analyses firm performance by a measure of stock returns, one could argue that it is not necessary to consider a long-term vs. short-term effect. This is mainly due to the assumption of the efficient market hypothesis (Yaes & Bechhoefer, 1989) which states that stock prices will adapt quickly to the latest information in the market. Therefore, policy-induced changes in the generation pattern that could affect the performance on the long are expected to be incorporated in the stock prices. As positive short-terms are expected by the studies both Ohler and Fetter (2014) and Jaraite, Karimu & Kazukauskasa (2017) it will be interesting to find out the sign of this effect to see which of both results dominates the relationship.

(6)

lower electricity price on firm performance, a look will be taken at a study performed by Bagirov & Mateus (2019). This study looks at the relationship between oil prices and the firm performance. As they find a positive relationship between oil prices and firm performance, a decrease in the price of oil is expected to lead to worse firm performance. This effect differs, however, depending on the sector of the firm and seems to be stronger for oil and gas firms. Although this study analysed oil prices rather than electricity prices, the underlying relationship can be considered comparable to energy generating firms and the energy prices. Based on the findings of Bagirov & Mateus (2019), we infer that there should be a relationship between the energy price and firm performance of electricity-generating firms. However, for a firm which is not actively involved in the generation of electricity, this effect may not be discovered. For these firms, lower electricity prices can be thought to be beneficial as it will save costs on the electricity bill. This would mean that the decrease in electricity prices is expected to hurt the performance of electricity-generating firms, whereas this would not be the case for firms which are not involved in electricity generation. Congruent with this reasoning a paper of Lynch & Curtis (2016) mentioned that an increase in wind generation tend to benefit the consumers more than the generators. Based on these ideas, firms that generate more electricity are expected to face lower firm performance due to the lower electricity prices. This expectation fits the story on two sides. At first, firms importing energy will benefit more from lower prices, as their business costs will decrease. At second, firms that generate their energy will, due to a lower price, have to deal with lower margins, which may decrease their profitability. This expectation will be tested in this study to find out whether the data fit our story.

Next to the possible lower energy prices as a consequence of the increase in RES, an increase in volatility in this price, due to higher price risks is expected. As increased uncertainty discourages investment and, therefore, requires a higher return, the increased price risks are likely to increase the cost of capital (Stern, 2007, p. 366). This would mean that increased price risks will decrease firm performance. Furthermore, it can, based on this idea, be expected that an increase in price risks will have more impact on firms that are closely related to the energy sector as energy forms a bigger part of their operations. Next to this. it would fit the story if it seems that firms with relatively much renewable generation will face more negative consequences of an increase in the capacity of RES compared to firms with relatively less renewable generation. This is expected as having more RES may provoke more price risks for the firm. This expectation will also be tested to clarify the main results of this study.

(7)

hypotheses can be outlined: (1), We expect a negative relationship between the financial

performance of European firms and the increasing generation capacity of RES. And (2), we expect this relationship to be stronger for firms involved in electricity generation. Next to these

two main questions, two sub-hypotheses can be outlined: (3) It is expected that firms with

relatively more electricity generation will face a heavier impact of an increase in wind and solar generation on stock returns compared to firms with relatively more electricity import. And (4), we expect electricity firms with a higher amount of generation by RES, to show a heavier impact of an increase in wind and solar generation capacity on stock returns compared to firms with relatively less RES.

3. METHODOLOGY

In this study, the two key hypotheses mentioned before will be examined. Especially, we will customize the study of Jaraite, Karimu & Kazukauskasa (2017), discovering the negative long-term relationship between the expansion of renewable energy capacity and economic growth, towards a firm perspective. It will be interesting to discover what the impacts of significant support to the wind and solar electricity are on the well-being of firms in various sectors. Furthermore, to explore the possible findings of the two main hypotheses, the two other sub-hypotheses will be tested. Therefore, we will explore whether the effect differs depending on the relative generation of a company by looking at the relative amount of electricity supplied that has been imported and at the amount of generation that is due to renewable sources. Besides, as a robustness check, we will analyse whether the study suffers from using the wind and solar generation capacities of the country in which the company is listed as the main regressor. This will be done by plugging the relative amount of sales in the listed country into the regression. Now that the scope and the expectation of this study are known, it will be explained how the phrased hypotheses are going to be tested. In particular, the study concerns five explanatory variables: Wind generation capacity, Solar generation capacity, the relative amount of electricity generation that is renewable, the relative amount of total power supply that is generated inhouse and, the percentage of sales in the home country of a company. These variables will be used in three different panel-data models that are being estimated. The first model is estimated in order to look for the possible impact of installed wind and solar capacity in a country on the stock returns of companies listed in that particular country. This model will be estimated for all companies in the dataset to test for hypothesis 1. In order to analyse the model and as a robustness check, this model will also be estimated for all the different sectors and all the different countries in the dataset. After this, the focus will be targeted on the firms that generate energy. In order to test for hypothesis 2 this first model will be worked out more detailed for the energy generating firms. The main model can be written as:

(8)

Where Y refers to the stock return of company i at time t, Wjt is the increase in Wind generation

capacity in country j and, Sjt is the increase in solar generation capacity.

Thereafter, a second model is used to also explore the relationship that the relative amount of generation and the amount of generation by RES may have on the effect estimated in the first model. This model allows testing for sub-hypotheses 3 and 4. In order to see whether the results are robust over the different countries the firms are listed in, it will also be estimated for all the different countries in the dataset. This second model can be written as:

Yit = f (WNDjt, SOLjt, OGi, RGi) (2)

Where OGi refers to the percentage of energy that is supplied generated by the company and

where RGi refers to the relative amount of renewable sources in the total electricity supply of

the company.

Finally, a third model is imposed as a robustness check to see whether the relative amount of activity in the listed country has an impact on the effect estimated in the first model. This second model can be written as:

Yit = f (WNDjt, SOLjt, SHi) (3)

Where SHi represents the percentage of total sales in the country in which the firm is listed.

Consequently, the following three equations will be estimated to investigate the relationship between company performance and the growth of renewable resources in a country:

lnYi,t = α + β1WNDj,t + β2SOLj,t + ei,t (4)

lnYi,t = α + β1WNDj,t + β2SOLj,t + β3OGi + β4RGi + β5WNDj,t OGi + β6SOLj,t OGi + β7WNDj,t

RGi + β8SOLj,t RGi + ei,t ` (5)

lnYi,t= α + β1 WNDj,t + β2SOLj,t + β3SHi + β4WNDj,t SHi + β5SOLj,t SHi + ei,t (6)

The beta’s are the elasticities of stock returns with respect to wind power capacity, solar power capacity, sales in the listed country, generation inhouse and, renewable generation, which will be estimated using the data. Furthermore ei,t denotes the error term. Interaction effects are added

in formula (5) and (6) to find out in the possible impact OGi, RGi and, SHi on the relationship

estimated by formula (4).

We test the two main hypotheses and the two sub-hypotheses mentioned above by estimating equation 4,5 and, 6. In order to estimate these equations meaningfully, a look will first be taken at the properties of our panel-data. To start, a unit root or stationarity test as suggested by Pesaran (2007) is used to discover the time series characteristics of the variables and to check for stationarity. This test tests the H0: All panel units contain Unit root against Ha: Some panels are stationary. The test will be conducted for the dataset with all firms for all

(9)

be expected and possible trends are expected to disappear from the data. However, when our data shows to be non-stationary, making inferences of the models would be unreliable and transformation(s) should be made. At second, the cross-sectional dependency test suggested by Pesaran (2004) is used for all time-series data. This test is used to find out whether variables or residuals are correlated between groups in the panel setting. This would mean that shocks in one firm in our dataset are related to the shocks in another firm at the same time. The Cross-sectional dependence test developed by Pesaran (2004) tests the H0: Cross-sectional

independence. Rejecting the H0 would mean that the data is correlated between panel units. This

would mean that the individual observations are not independent, which could lead to making wrong inferences based on the results. The next test that is performed tests whether the models 1,1a and, 2 suffer from heteroskedasticity. This test is performed by comparing the model against a model with heteroskedasticity and performing an LR-test. Rejecting the H0 of

homoskedasticity, in this case, means that the model suffers from heteroskedasticity, meaning that the error terms of an estimation do not have constant variance (Brooks, 2019). Using ordinary least squares (OLS) estimation with heteroskedasticity could lead to inappropriate standard errors and hence any inferences made could be misleading. In order to tackle this problem the log-returns of the stock prices will be estimated in the models as is advised by Brooks (2019). The final test that is performed to analyse the properties of the data is used for testing autocorrelation in the data of each model. In order to test this, the test as is suggested by Wooldridge (2010) is used. This test tests the H0: no first-order autocorrelation. A detection of

autocorrelation could be a sign of inappropriate standard errors, potentially leading to wrong inferences. A solution to the problem of autocorrelation would be to add a lagged variable to the model. All the tests described so far, are used to discover the properties of the dataset. Based on these four tests conclusions can be made about the properties of the data, and the need for corrections to our models. This analysis of the results will be performed in the ‘results and

discussion’ part of the paper.

When coming to estimating the models, different estimation techniques can be used. For this study, pooled OLS, Random effect (RE) and, fixed effect (FE) models will be applied. To define the exact empirical approach for each equation, three steps are being taken to guide the choice on which estimation technique suits the dataset best. At first, the Lagrange multiplier test suggested by Breusch and Pagan (1980) is used to decide between a pooled OLS and a random-effects model. Rejecting the H0 of this test means that the random effect is different

from zero and a random-effects model would be preferred over a Pooled OLS estimation. Second, the F-test of the fixed-effect model will be used to access whether the fixed-effects model or and OLS estimation is desirable. Rejecting the H0 of this F-test would mean that the

(10)

means that there is a detection of endogeneity and that the fixed-effects model will be preferred. As the usage of panel data models (fixed-effects or random-effects) has the advantage of allowing for effects between entities and access the characteristics of the panel data (Brooks, 2019), both an OLS and the preferred panel data model will be regressed for each model. The estimation technique that suits the data best according to the three tests will then be used as the main estimation, whereas the other techniques will serve as a robustness check. The results of the abovementioned tests can be found in the ‘results and discussion’ part of this paper. Based on these results the exact empirical approach will be decided upon according to the mentioned procedure. Before turning to these results, a description of the used dataset will be given in the following part.

4. DATA

As is made clear in the methodology section, this study concerns six different variables, these variables are described in Table 1 below. In order to estimate model (1), (2) and, (3), data has been collected upon these variables. At first, monthly data on the historical stocks prices of all European firms listed at the stoxx600 index are collected from Thomson Reuters Eikon in March 2020. Data were collected for all 600 firms for January 2015 until February 2020, the sectors of these firms are based on the classification in this index. Next, monthly data on the solar and wind generation capacity of seven European countries (Austria, Belgium, France, Germany, Italy, Netherland and, the United Kingdom) has been collected from Thomson Reuters Eikon in March 2020, leading to a total of 427 individual observations on both solar and wind generation capacity.

(11)

University of Groningen | Peter Sven Smilda 10

Table 1

Description and measurement units of used variables

Note. This table describes the six variables that are used in this study. In the right-most column the measurement units of this specific variable means.

In the first model, the historical stock prices will be used to obtain the logarithmic stock returns for February 2015 until February 2020 and will be used as a proxy for firm performance. Next to this, the percentual growth in the generation capacity of solar and wind power will be used to represent the policy-induced growth in renewable resources. For the second model, the total amount of generation in terawatt-hour (TWh) of the individual electricity firms will be divided by the total amount of electricity in TWh that is supplied to both consumers and the wholesale market. For the analysis a dummy is created, which takes the value of 1 if the variables are higher or equal than the median (81.9%, see table 2) and 0 otherwise. Furthermore, the amount of renewable energy that is generated by the individual firms is divided by their total generation in TWh to obtain a proxy for the renewable- dependency in the generation process of the individual firms. Also, this variable is turned in to a proxy based on it’s median of 14.8%. Whenever the value is higher or equal to the median the variable will take value 1. Its value will equal 0 for the cases in which it’s lower. For the third model, we used the percentage of sales in the country the electricity firm is listed as a way to access the dependency of the firm to changes in that specific country. In the regression, this variable will be 1 whenever this percentage is 100%, and 0 otherwise. As these variables do not differ over time, these variables all consist of 14 different observations, 1 for each firm. Descriptive statistics on each of these variables can be found in Table 2.

Variable Description Measurement unit

Stock returns Return of historical stock prices of 376 firms from January 2015 until February 2020.

Euro & GBP

Wnd Percentual increase in historical wind generation capacity of 7 EU countries from January 2015 until February 2020.

Megawatt

Sol Percentual increase in historical solar generation capacity of 7 EU countries from January 2015 until February 2020

Megawatt

RenGen Generation capacity for wind and solar energy to total generation capacity for 14 electricity firms

% Megawatt

GenInhouse Generation of power to the total supply of power for 14 electricity firms.

% Twh

HomeSales Amount of power sales in the listed country to total power sales

(12)

Table 2

Descriptive statistics of the variables used in the study

Note. This table shows the number of obstervations, mean, standard deviation, minima, median and, maxima of the used variables in the study.

5. RESULTS AND DISCUSSION

Now that the methodological setup and the used dataset is introduced, a look will be taken at the results of the statistical tests before the results of the estimations will be discussed. The results from the various tests concerning the properties of the data series will be presented to form an argument about the chosen estimation methods. At first, the variables including time series (stock returns, wind generation capacity and, solar generation capacity) are tested for stationarity and cross-sectional dependence. Then, model 1, 1a and, 2 are tested for heteroskedasticity and autocorrelation. Finally, the optimal estimation technique for these models is accessed based on the Lagrange multiplier test of Breusch and Pagan (1980), the fixed-effects F-test and, the Hausman test.

5.1 Empirical models

The results of the stationarity test and the cross-sectional dependence test can be found in Table A1 and A2 of the appendix consecutively. From Table A1, it can be seen that the data follows an I(0) process. This means that the variables regarding the stock prices and the solar generation capacity seem to be stationary. The results of Table A2 show that all the null hypotheses of cross-sectional independence are rejected for all the time-varying variables. This means that the data is correlated across panel units, which should be taken into account when considering the estimation uses of the models in this study. Next, the results of the heteroskedasticity and autocorrelation tests are summarized in Table A3 and A4 of the appendix consecutively. As can be seen in Table A3, the H0 of homoskedasticity is rejected at 1%

significance, implying heteroskedasticity in all three models. The detected heteroskedasticity

Variable No. of Obs. Mean St. dev Min. Median Max.

(13)

would mean that the standard-errors of our estimations should be corrected in order to make adequate inferences about the models. When considering the autocorrelation test, it can be seen from table A4 that the of H0: no first-order autocorrelation, is rejected at the 5% significance

level for the first model, implying the presence of first-order autocorrelation. For model 1a and 2, the H0 could not be rejected, implying no first-order autocorrelation. Due to the detection of

this autocorrelation in model 1, this model will be estimated by including the first lagged variables in the estimation. Finally, to decide which model to use three more tests are performed. The results of these can be found in table A5 of the appendix. When considering the Lagrange multiplier test suggested by Breusch and Pagan (1980) to decide between a pooled OLS and a random-effects model, we see that in all three models the H0 cannot be rejected,

implying that pooled OLS would be more suitable than a random-effects model. Furthermore, the F-test for the fixed-effects model cannot reject the H0 in all three models, implying that OLS

estimation would be better suited than the fixed-effects model. When comparing fixed-effects with random-effects, the Hausman test shows us that the random-effects is more appropriate for model 1 and 2 and that a fixed-effects model should be used for model 1a.

Now that the several test outcomes are known it can be concluded that our data suffers from cross sectional-dependence, all the models suffer from heteroskedasticity and, Model 1 suffers from first-order autocorrelation. To solve these potential issues, the models will be estimated with the Driscoll and Kraay’s (1998) standard errors, which tend to be robust to autocorrelation, heteroskedasticity and cross-sectional dependency. When considering the main estimation technique, a pooled OLS estimation is preferred according to the results of table A5. For model 1 and 2, a random-effects model will be regressed next to the pooled OLS estimation as a robustness check. For model 1a, rejecting the null-hypothesis of the Hausman-test for endogeneity means that the fixed-effects model is preferred over random-effects. This may be the case due to the presence of an endogenous variable (Hausman, 1978). The possible presence of endogeneity in this model may affect the OLS estimation as the assumptions may not be met. The regression results of the OLS estimation will, therefore, be compared with the results of the fixed-effects estimation to access the severity of this issue.

5.2. Model 1

(14)

relationship, between generation capacity for the RES and firm performance, is contrary to the first hypothesis of this study. However, we cannot make inferences about this hypothesis as the results of model 1 do not show statistical significance. Despite, not being able to confirm the hypothesis, the expected positive effect, although it is small can be found in all the different estimations performed. It is interesting to see that the results show significance when the standard errors of Driscoll and Kraay are being relieved.

Table 3

Regression results of model 1, for all firms and electricity-generating firms specific.

Model 1: All firms.

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

VARIABLES OLS_DK RE_DK OLS RE

Wnd 0.035 0.036 0.035*** 0.036** (0.036) (0.036) (0.011) (0.015) Sol 0.065 0.066 0.065*** 0.066*** (0.069) (0.069) (0.021) (0.019) Constant 0.418 0.416 0.418*** 0.416*** (0.430) (0.444) (0.059) (0.062) Observations 22,904 22,904 22,904 22,904 R-squared 0.000 0.000

Model 1a: Electricity-generating firms

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

VARIABLES OLS_DK FE_DK OLS FE

Wnd -1.354*** -1.641* -1.354*** -1.641* (0.484) (0.973) (0.384) (0.854) Sol -0.297* -0.349** -0.297** -0.349*** (0.149) (0.161) (0.141) (0.081) Constant 1.601** 1.849** 1.601*** 1.849** (0.648) (0.879) (0.338) (0.672) Observations 854 854 854 854 R-squared 0.018 0.018 0.009

(15)

included in the study. From this table it can be seen that the expected positive relation for wind generation capacity and stock returns could be found in 4 out of the 7 countries. This is only 3 out of 7 when looking at the solar generation capacity. Based on this, it is not possible to make inferences about the relations between the generation capacity of RES and the stock returns of a company in general. A possible explanation of not finding effects for the sample including all the firms may be given in a paper by Kim, Rahman and Shamsuddin (2019). They, namely found in a US dataset that energy prices became a less predictor of excess firm returns over time. This may suggest that the impact of energy prices on firm performance in the last couple of years became smaller, which may contribute to not finding significant results in this study. This result does, however, not directly state that the generation capacity of RES does not influence the stock prices of individual firms. When a look is taken at table C in the appendix, it follows that in 13 of the 20 sectors no statistically significant effects of the generation capacity of Solar and Wind power on stock returns can be found at the 5%. However, in 7 sectors (basic resources, construction & materials, energy, healthcare, media, travel & leisure and, utilities) at least some statistically significant effects can be noted at the 5% significance level. When taking out the energy sector, as this sector will be discussed independently later in this paper, 5 of the 7 effects are positive of which 5 are related to solar generation capacity and 2 two wind generation capacity. The travel & leisure sector found both positive effects for wind generation capacity as well as solar generation capacity. As these sectors are not studied in depth it is hard to make appropriate inferences about these results and to determine which role negative economic growth and electricity prices may have played. In the upcoming part of the study the analysis will narrow down on firms that are actively involved in electricity generation processes. The estimation results for this sector can be seen in the bottom half of table 4.

5.3. Model 1a

(16)

larger negative effect. Also, the Driscoll & Kraay standard errors are slightly higher compared to the regular robust standard errors, as was to be expected. For this regression, comparing the results of the fixed-effects model with the results of the pooled OLS does also say something about the impact of possible omitted time-independent variables, as these variables would cancel out in case they do not differ over time. Based on the comparable results between the OLS estimation and the fixed-effects estimation an omitted variable that does not change over time does not seem to be influential in this case. In order to check whether the found results are consistent over the different countries studied, the results of the model are estimated for the different countries individually, these results can be found in Appendix B1. Although these estimation lack statistical significance due to the low amount of observations, it can be seen from this table that the effect of an increase in wind generation capacity on stock returns can be observed in all the six countries from which firm data was obtained. For the solar generation capacity, this comes down to four out of the six countries that show a negative relationship with stock returns. The results concerning the generation capacity of wind power is not only statistically relevant, but it is also economically applicable as a one per cent increase in the generation capacity of wind power in a specific country is estimated to cause, on average, a 1.354 per cent decrease in the stock prices of the electricity-generating firms in that specific country. It is interesting to see that the effect seems to differ between wind and solar power, as the effect is both significantly as economically less clear. As far as considered, there is no clear explanation between this discrepancy between the effects of an increase in wind and solar generation capacity. A possible explanation behind this result could be the differences in the marginal price of both the power generation methods. As discussed by Mulder & Scholtens (2013) solar generation has different marginal costs compared to wind energy, which may be a reason for the lower negative impact. Whether or not this is a main driver for the observed effect could be an interesting topic for future research.

(17)

(2019) when studying shocks in the oil prices. However, as this study does not test for the direct relationship with electricity prices, no strong inferences can be made about this reasoning at the moment. If the electricity prices would be added in a comparable analysis, the found results could be explained better. Another factor that could explain these results is the increase in volatility of the electricity price as suggested by Wiser et al. (2017). As this increased volatility would lower firm performance (Stern, 2007, p. 366), this could be another explaining factor behind the found negative effects. In order to further explore the possible relationship between lower electricity prices and more volatility in these prices, another model has been estimated taking the amount of (renewable) electricity generation into account.

Table 4

Regression results of model 2.

Note. This table shows the regression results of model 2. In this model the original model is extended with the dummy variables ‘GenInhouse & RenGen’ representing firm with relatively much generation performed inhouse and relatively much renewable generation consecutively. Next to this, the interaction effects with both, the wind and solar generation capacity are added to the original model. This model is estimated four times, with using pooled OLS and the random-effects estimation both, with and without using the standard errors suggested by Driscoll & Kraay (1998). Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10.

5.4. Model 2.

Model 2

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

VARIABLES OLS_DK RE_DK OLS RE

(18)

(19)

expectations to see that both (solar and wind) interaction effects are positive. Nevertheless, the statistical significance of these effects is low.

5.5. Explaining factors

An explanation for the finding that firms with more renewable generation do not seem to suffer from shocks due to more volatility in electricity prices can maybe be found when looking at the research conducted by Ng & Zheng (2018). This study concludes that energy-friendly firms can succeed greatly by the increasing demand for energy investments. A paper by Hartzmark & Sussman (2019) does also conclude that being related to sustainability increases the demand for investments. Therefore, we would expect that firms with relatively much production of RES will benefit from the increasing demand for investments, which may serve as a potential buffer to the higher volatility in the electricity prices. Another possible reason that the expected influence of lower electricity prices and more volatility in these prices has not been found may be due to the set up of this study. This study looked at the absolute increase in generation capacity. However, it should be noted that an increase in the total capacity of RES does not translate into a 1 on 1 increase of RES in the share within the generation mix. Firstly, this may be caused by the increasing demand for electricity we face at the moment. In the past years, changes are being made to more electric products in, for example, housing (heat pumps) or in the automobile industry. An increase in the total demand for electricity would mean that traditional gas plants would still be needed to meet the total demand for electricity (Moraga & Mulder, 2018). Furthermore, a situation in which the supply of electricity will be reliable for 100% on RES, is not thinkable of at the moment as is mentioned by Papaefthymiou and Dragoon (2016). The need for traditional generation sources is caused by the intermittent characteristics of generation wind and solar power. An increase by RES, namely, causes more volatility in the supply for energy (as we cannot choose when the sun shines). Because of this intermittency, flexible energy sources, such as gas plants, are needed to balance the supply and demand, which is important to deliver a constant amount of energy (Ecofys, 2014). Based on this it could be the case that the growth in the generation capacity of wind and solar power did not lead to a weighty increase in the share of RES in the generation mix. As the changes in prices and volatility were only expected as a result of changes in the electricity mix, this may contain the results. Another reason why the increase in generation capacity might not have lead to changes in the electricity price and volatility can be found in the study of Figueiredo et al., 2015. In their study, they namely found that an increase in RES in the electricity mix will increase the probability of market splitting (more transition of energy across borders). This may also contain the downward pressing effect on the electricity prices, as the demand for this cheaper energy will increase.

(20)

model. As not all the 14 firms that are included restrict their operations to their home country, this model analyses whether the results differ between firms with 100% of their activity (measured by sales) in the country they are listed in and firms with less than 100% of their sales in their home country. The results of this model are represented in table 6. This regression is purely estimated as a robustness check in order check whether the estimated models suffer from the assumption that the changes in the renewable generation capacity in the country the firm is listed are partly accountable for the changes in the stock prices of the firms. Somewhat of an impact can be noticed when including the interaction variable for the effect of wind generation capacity. Surprisingly, this effect is found to be positive and significant at the 10% significance level. Although this result is not shown in the fixed-effects model, the positive interaction effect may suggest that the results of the current study would be less comprehensive than suggested in the first place.

Table 5

Regression results of model 3

Model 3

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

VARIABLES OLS_DK FE_DK OLS FE

HomeSales -0.780 0.000 -0.780 (0.647) (0.000) (0.685) Wnd -1.878*** -1.879* -1.878*** -1.879 (0.663) (1.110) (0.533) (1.182) HomeSales*Wnd 1.341* 0.959 1.341* 0.959 (0.743) (1.342) (0.787) (1.195) Sol -0.312* -0.341* -0.312 -0.341*** (0.177) (0.189) (0.203) (0.101) HomeSales*Sol 0.007 -0.007 0.007 -0.007 (0.120) (0.143) (0.230) (0.101) Constant 2.047** 1.896** 2.047*** 1.896** (0.795) (0.885) (0.524) (0.762) Observations 854 854 854 854 R-squared 0.020 0.020 0.010

Note. This table shows the estimation results of model 3, which is an extansion to the first model serving as a robustness check. In this model the dummy variable ‘HomeSales’ is added to the model. This variable represents the firms that have all their sales in their listed country. The model is estimated four times by using pooled OLS and the fixed-effect, both with the Driscoll and Kraay standard errors and without. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10.

(21)

Jaraite, Karimu & Kazukauskasa (2017), technological gains may lead to higher utilization of installed capacity, allowing for higher production of electricity whilst the generation capacity stays the same.

6. CONCLUSION

In this study, research is done regarding the effects of policies supporting renewable energy on firm performance in seven European countries. The effects of the different policies are accessed by using the generation capacity of wind and solar sources, this is done because it is believed that generation capacity is a more direct measurement for sustainability policies compared to generation. As the generation capacity for RES has been increasing, this study tries to identify the impact of this increase on firm performances. This is done by accessing the main drivers that could affect the firm performance and their possible impacts.

Although no relations were found on firm performance in general, in some sectors results were being found. This study found that the firm performance for firms engaging in electricity production is negatively related to the increase in generation capacity of RES (wind and solar). This result was in line with the expectations and the estimations. In order to identify whether this negative relationship would be caused by the possible impact that the increase in generation capacity for RES could have on electricity prices and it’s volatility, two secondary analyses are performed on the relative import and the renewable generation of these firms. However, as both analyses do not represent any statistical significance, the main drivers of the found effect are still to be discussed. Therefore, it could be seen as a shortcoming of the current study that the electricity prices were not included. It will be interesting to explore this negative relationship further to explain the in-depth consequences on the effects of governmental policies stimulating more renewable energy on firm performance. For future research, it would be recommended to perform a comparative study in which electricity prices are included, to test whether this could be the main driver. Furthermore, using a sample of solemnly focussed on energy firms may be interesting to obtain more data and, therefore, a better understanding of the results of this study.

(22)

References

Alizamir, S., de Véricourt, F., & Sun, P. (2016). Efficient feed-in-tariff policies for renewable energy technologies. Operations Research, 64(1), 52–66.

Bagirov, M., & Mateus, C. (2019). Oil prices, stock markets and firm performance: Evidence from Europe. International review of Economics and Finance, 61, 270–288 Breusch, T. S., and A. R. Pagan. (1980). The Lagrange multiplier test and its applications to

model specification in econometrics. Review of Economic Studies 47: 239-253. Brooks, C. (2019). Introductory econometrics for Finance, 4th_edtion_{. Cambridge University}

Press, Cambridge.

Bohi, D. R. (1989). Energy price shocks and the macroeconomic performance. Washington D.C: Resources for the Future.

Curtis, J. & Lynch, M. A., J. & (2016). The effects of wind generation capacity on electricity prices and generation costs: A Monte Carlo analysis. Applied Economics, 48(1-3), 133-151.

Cludius, J., Forrest, S., & MacGill, I. (2014). Distributional effects of the Australian renewable energy target (ret) through wholesale and retail electricity price impacts. Energy Policy, 71, 40-51.

Driscoll, J., & Kraay, A. (1998). Consistent covariance matrix estimation with spatially dependent panel data. The Review of Economics and Statistics, 80(4), 549-560.

Ecofys, 2014. Flexibility options in Electricity Systems. March 2014, Berlin, Germany.

Eikon. (2020) Thomson Reuters Eikon. [Online]. Available at: Subscription Service (Accessed: March 2020).

European Commission (2017). Statement. <http://europa.eu/rapid/press-release_ STATEMENT-18-4155_en.htm>

European Commission (2008) Commission staff working document: The support of electricity from renewable energy sources.

Eurostat. 2020. “Summary results shares 2018.” [Online]. Accessed 21 February, 2020. Figueiredo, N. C., Silva, P. P. da, & Cerqueira, P. A. (2015). Evaluating the market splitting

determinants: Evidence from the Iberian spot electricity prices. Energy Policy, 85, 218–234.

Hartzmark, Samuel., & Sussman, Abigailb. (2019). Do investors value sustainability? a natural experiment examining ranking and fund flows. The Journal of Finance, 74(6), 2789–2837.

Hausman, J. A. 1978. Specification tests in econometrics. Econometrica 46: 1251–1271. IEA, 2020. “Renewable share (modern renewables) in final energy consumption (SDG 7.2),

World 1990-2016. Accessed 21 February, 2020.

Jaraitė, J., & Kažukauskas, A. (2013). The profitability of electricity generating firms and policies promoting renewable energy. Energy Economics, 40, 858–865.

Jaraite, J., Karimu, A., & Kazukauskasa, A. (2017). Policy-induced expansion of solar and wind power capacity: Economic growth and employment in EU countries. Energy Journal, 38(5), 197-222.

Johnson, E. P., & Oliver, M. E. (2019). Renewable generation capacity and wholesale electricity price variance. Energy Journal, 40(5), 143–168.

Kim, J.H., Rahman, M.L., & Shamsuddin, A. (2019). Can energy prices predict Stock returns? An extreme bounds analysis. Energy Economics, 81, 822-834.

(23)

Mulder, M., & Scholtens, B. (2013). The impact of renewable energy on electricity prices in the Netherlands. Renewable Energy, 57, 94–100.

Ng, A., & Zheng, D. (2018). Let's agree to disagree! on payoffs and green tastes in green energy investments. Energy Economics, 69, 155-169.

Ohler, A., & Fetters, I. (2014). The causal relationship between renewable electricity generation and GDP Growth: A study of energy sources. Energy Economics, 43, 125–139.

Papaefthymiou, G., & Dragoon, K. (2016). Towards 100% renewable energy systems: Uncapping power system flexibility. Energy Policy, 92, 69-82

Pesaran, M. H. (2007). A simple panel unit root test in the presence of cross section dependence. Journal of Applied Econometrics 22: 265–312.

Pesaran, M. Hashem (2004). General diagnostic tests for cross section dependence in Panels. CESifo Group Munich CESifo Working Paper Series 1229.

Ropenus, S., & Jensen, S. G. (2009). Support schemes and vertical integration-Who skims the cream? Energy Policy, 37(3), 1104–1115.

Stern, N., 2007. Stern Review: The economics of climate change. Cambridge University Press, Cambridge.

U.S. Environmental protection agency (2019), inventory of U.S. greenhouse gas emmissions c and sinks: 1990-2017.

Wiser, Ryan H., Andrew Mills, Joachim Seel, Todd Levin, and Audun Botterud (2017). Impacts of variable renewable energy on bulk power system assets, pricing, and costs. No. LBNL-2001082. Lawrence Berkeley National Lab.(LBNL), Berkeley, CA (United States).

Wooldridge, J. M. 2010. Econometric Analysis of Cross Section and Panel Data. Cambridge. US Environmental Protection Agency, 2015.

(24)

APPENDIX A.

Table A1

Unit root test at total firm set.

No Trend Trend

Lags Test Stats p-value Lags Test Stats p-value

Stocks 1 (91.665) 0.000 1 (85.429) 0.000

Wnd 1 (8.659) 0.000 1 (12.311) 0.000

Sol 1 (110.000) 0.000 1 (110.000) 0.000

Note. This table shows the results of the Unit root test for stationarity developed by Pesaran (2007). In this test the H0: All panels contain unit root, is tested. The results show that for all time-series the H0 is

rejected. It does not affect the results whether a trend is included or not.

Table A2

Cross-sectional dependence test.

Note. This table shows the results of the cross-section dependence test developed by Pesaran (2004). This test tests the H0: Cross-sectional independence. The tested time-series do all reject the H0.

Table A3

LR- test for homoskedasticity.

Note. This table shows the results of the homoscedasticity test performed on our models. For all three models the H0 of homoskedasticity is rejected, meaning that there is a sign of heteroskedasticity. Table A4

Wooldridge test for first order autocorrelation.

Wooldridge test p-value

Model 1 4.246 0.040

Model 1a 2.282 0.155

Model 2 2.282 0.155

Note. This table shows the results of the Woolridge (2010) test for autocorrelation. The results show that only model 1 rejects the H0: No first-order autocorrelation

Variable CD test p-value Average correlation

(25)

Table A5

Results of BP Lagrange multiplier test, F-test and Hausman test for the three models.

Note. This table shows the test results of the various tests that are used to decide upon the empirical approach together with their p-values. At first the Lagrange multiplier test developed by Breusch & Pagan (1980) is used to decide between pooled OLS and random-effects estimation. As none of the models shows a significant result, the H0 cannot be rejected, meaning that for all the models a pooled

OLS estimation would be preferred over random-effects. At second, a F-test is used to decide between pooled OLS and fixed-effects estimation techniques. As the H0 is not rejected in one of the tests, it can

be concluded that a pooled OLS estimation would be preffered over fixed-effects. Finally, a Hausman (1978) test is used to decide between random-effects and fixed-effects. As this test is significant at the 1% level for model 1a, it suggests that fixed-effects would be better suited for this model, compared to random effects.

BP Lagrange multiplier test

(26)

APPENDIX B. Table B1

Robustness check for different countries.

Model 1 (1) (2) (3) (4) (5) (6) (7) VARIABLES AT BE DE FR GB IT NL Wnd -2.304 0.001 1.591 1.515 0.808 -1.034 -0.030 (2.819) (0.003) (2.030) (1.813) (0.544) (16.53 0) (0.434) Sol 1.386 -0.075 2.418 -2.381*** 0.162** -62.459 -0.124 (1.011) (0.045) (1.887) (0.754) (0.079) (41.23 2) (0.855) Constant -0.017 0.618** * -1.399 1.504 -0.903 10.084 1.037 (1.877) (0.092) (1.657) (1.672) (0.975) (7.643) (1.738) Observations 488 1,098 4,087 5,238 8,903 1,820 1,270 R-squared 0.005 0.007 0.005 0.026 0.006 0.016 0.000 Number of groups 8 18 67 86 146 30 21 Model 1a (1) (2) (3) (4) (5) (6) VARIABLES DE FR GB IT AT BE Wnd -9.758** -0.780 -0.361 -10.692 -9.757* -3.156 (4.366) (2.924) (0.966) (13.887) (5.087) (2.263) Sol 4.521 -1.542 -0.239 -78.997*** 0.778 7.353*** (7.019) (1.615) (0.181) (26.271) (1.437) (2.753) Constant 6.173* 2.019 0.233 14.487*** 7.321* * 1.571 (3.510) (2.628) (1.545) (5.150) (3.505) (2.037) Observations 122 183 183 244 61 61 R-squared 0.064 0.013 0.016 0.058 0.066 0.076

(27)

Table B2.

Robustness check for the different sectors.

Note. This table shows the results of model 1, when the equation is estimated for the different sectors in the dataset. The estimation technique used is pooled OLS with the Driscoll & Kraay standard errors. Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10.

Model 1: Results per sector.

INDUSTRY Wnd Sol Constant Observations R-squared

Automobiles & parts 2.069 -1.379 -0.848 732 0.006

(2.272) (1.057) (1.686) Banks 0.122 -0.000 -0.422 1392 0.000 (-0.074) (0.139) (0.826) Basic resources 3.006*** -0.311 -3.454** 610 0.038 (1.111) (0.286) (1.600) Chemicals 0.011 0.084 0.310 1037 0.000 (0.020) (0.104) (0.591)

Construction & materials 3.563* -1.258** -1.250 366 0.029 (1.842) (0.525) (1.834)

Energy -1.354*** -0.297** 1.601*** 854 0.018

(0.429) (0.133) (0.572)

Financial services 0.049 0.128 0.640 1342 0.003

(0.054) (0.077) (0.461)

Food & beverage -0.021 -0.133* 0.982** 732 0.003

(0.027) (0.066) (0.399)

Healthcare -0.012 0.142** 1.004** 1769 0.001

(0.013) (0.069) (0.478)

Industrial goods & services 0.523 0.116 0.031 4079 0.003 (0.380) (0.088) (0.723)**

Insurance -0.018 0.055 0.394 1332 0.001

(0.092) (0.063) (0.434)

Media -0.025 0.278*** -0.163 1034 0.009

(0.056) (0.054) (0.394)

Oil & gas 0.147 -0.004 -0.706 671 0.000

(0.625) (0.235) (0.921)

Personal & household goods -0.182 0.048 1.136 1403 0.001 (0.428) (0.096) (0.680) Real estate -0.028 -0.015 0.461 1403 0.000 (0.043) (0.089) (0.486) Retail 0.033 0.020 0.402 1098 0.000 (0.098) (0.083) (0.553) Technology -0.809 0.083 2.015** 1159 0.002 (0.653) (0.126) (0.973) Telecommunications 0.013 0.065 -0.388 671 0.000 (0.061) (0.121) (0.497)

Travel & leisure 1.491** 0.220*** -1.789* 854 0.013

(0.668) (0.079) (1.052)

Utilities -0.331 -0.205** 1.128 305 0.009

(28)

APPENDIX C. Table C1

Information on energy generating firms in the data set.

Firm Sector Country %sales in

country % Generation inhouse % renewable generation A2A Energy IT 100% 93% 0% CENTRICA Energy GB 46% 49% 0% E.ON Energy DE 30% 38% 30% EDF Energy FR 75% 116% 7%

ELIA GROUP Energy BE 41% 87% 47%

ENEL Energy IT 51% 85% 11% ENGIE Energy FR 41% 130% 7% HERA Energy IT 100% 89% 9% RWE Energy DE 37% 77% 19% SCOTTISH & SOUTHERN ENERGY Energy GB 88% 79% 22% TERNA Energy IT 100% 90% 22%

UNITED UTILITIES GRP Energy GB 100% 21% 23%

VEOLIA

ENVIRONNEMENT

Energy FR 30% 55% 26%

VERBUND Energy AT 55% 49% 5%