THE EFFECT OF RENEWABLE ELECTRICITY GENERATION ON ELECTRICITY PRICES

(1)

Author: Daniël Schans, S2469308 Thesis supervisor: dr. P. Rao Sahib Co-assessor: dr. R.K.J. Maseland Date: 18 June 2019

MSc Thesis International Economics & Business

Faculty of Economics and Business, Rijksuniversiteit Groningen Word count: 12,530

THE EFFECT OF RENEWABLE

ELECTRICITY GENERATION ON

ELECTRICITY PRICES

(2)

2

Abstract

A fixed-effects regression model is constructed in order to determine the merit-order effect in Central Western Europe. In order to do this, the effect of electricity demand, solar generation, wind generation, gas prices and the CO2 price on wholesale electricity prices was determined. The results indicate that a merit-order effect is present and that the magnitude of the merit-order effect differs per country. Also, it was found that the merit-order effect differs over time for solar generation, but not for wind generation.

JEL classification: Q21, Q42

(3)

3

1. Introduction

The European electricity markets have undergone significant changes over the last two decades. Around the start of this millennium, Europe experienced a transition in electricity markets. This transition primarily focused on liberalization of electricity markets and therefore withdrawal of the state’s role in the electricity industry. Liberalization of the electricity industry involved creating competitive wholesale and retail markets, while transmission and distribution activities remained regulated. It encompassed selling off public enterprises and allowed market entrance of new private entities, both domestic and foreign. By allowing domestic and cross-border competition, electricity markets became more integrated. This increased integration made it possible to buy electricity in foreign countries and therefore the market could respond to price signals, which in turn should lead to greater electricity price convergence across European countries (Jamasb & Pollitt, 2005).

Currently, Europe is experiencing another electricity market transition. The primary focus of this transition is decarbonizing electricity markets, thus decreasing the amount of greenhouse gas emissions as a result of electricity generation. Decarbonizing the electricity system should be achieved by moving away from fossil fuels and increasing renewable electricity production. A large collaborative effort in order to increase deployment of renewable energy sources (RES) is the Paris Agreement, where a landmark agreement was established in order to mitigate climate change. Here, it was agreed that countries will take substantial efforts in order to limit global warming to 1.5 °C by 2100 as compared to preindustrial times. The European Union (EU) formally ratified the Paris Agreement on 5 October 2016, implying that it is now rule of law for EU countries. Member states are individually responsible for complying to the Paris Agreement, but also have a shared responsibility as European Union. Following from the EU’s wider 2030 climate and energy framework, the aim is to reduce greenhouse gas (GHG) emissions by at least 40% in 2030, compared to 1990. This implies a reassessment of the current power system in order to guarantee a stable, efficient and sustainable energy supply in the future (Zahedi, 2011).

(4)

4 tomorrow). This additional electricity supply coming from RES shifts the supply curve of electricity to the right, and the most expensive marginal plants are driven out of the market. In other words, fossil fuels are being replaced by RES in electricity generation. This, ceteris paribus, lowers electricity prices, which is referred to as the merit-order effect of renewables (Sensfuß et al., 2008). In this thesis, solar and wind will be the RES regarded. Thus, the negative effect of solar and wind generation on electricity prices is referred to as the merit-order effect.

Merit order effects have been investigated quite extensively across countries. Jensen & Skytte (2003) were the first to show that increasing renewable electricity generation levels in Denmark might reduce electricity prices. They pointed out that wholesale electricity prices would be reduced if fossil fuel plants are replaced by RES. Various authors have found evidence favoring the merit-order effect in different countries: Germany (Sensfuß et al., 2008; Cludius et al., 2014), Spain (Sáenz de Miera et al., 2008; Gelabert et al., 2011), Italy (Clò et al., 2015) and Texas (Woo et al., 2011). In other words, the majority of the authors find evidence for the merit-order effect, but their analyses is often limited to one country. This thesis will investigate the merit-order effect in a multi-country setting. The geographical scope of the thesis is the Central Western European (CWE1) electricity market. As a consequence, the main research question is:

What is the magnitude of the merit-order effect in Central Western Europe?

This is particularly relevant, because of the increased deployment of RES at the expensive of fossil fuels. In other words, fossil fuel generation capacity with high marginal costs that is replaced by RES. Because of the recent efforts for integration of the CWE electricity market, the relationship between the CWE electricity markets is becoming more important. Integrating these markets has led to an increased number of hours with equal wholesale electricity prices across countries. However, most of the time these prices still differ. Since these electricity markets are becoming more integrated, it would also be interesting to see whether merit-order effects differ. Therefore, a sub-question regarded in this thesis is: Is

the merit-order effect different across CWE countries? Also, it might be that the effect of RES

generation on electricity prices changes over time. Therefore, another sub-question addressed in this thesis is: Is the merit-order effect different across years?

The methods utilized for studying merit-order effects are quite different, which means findings for individual countries are hard to compare. This thesis fills this gap by investigating merit-order effects in a country setting, with an estimation method that allows for multi-country analyses. Regression models including fixed effects were specified in order to determine the merit-order effect across countries and throughout time. Data is extracted for the

(5)

5 period 2015-2018 and the sample consists the following countries: Belgium, France, Germany and the Netherlands. As discussed, the electricity markets of these countries became more integrated in recent years and furthermore they have different in electricity generation mixes. Therefore, in the context of investigating different merit-order effects, the CWE countries are interesting to study.

My main finding is that a merit-order effect is present in CWE countries, and that this merit-order effect differs across countries. Here, it was found that the negative effect of solar generation on electricity prices is larger than the negative effect of wind generation on electricity prices. Furthermore, the results indicate that the merit-order effect differs across countries. The largest negative effect of both solar- and wind generation on electricity prices was found in Belgium. Last, I examined whether the merit-order effect increases or decreases over time. For solar generation, the results indicate that the negative price effect is larger in 2015, as compared to the other years. For an increase in wind generation, no supportive evidence was found in order to conclude that the negative price effect in the other years was either larger or smaller than in 2015.

This thesis is divided into five sections, and will proceed as follows. First, in section 2, an overview is provided of the approach taken thus far in estimating merit-order effects. In section 3, the theoretical background of electricity markets will be discussed in order to ensure the reader fully grasps the research question of this thesis. Additionally, in section 3 factors affecting electricity prices will be described. Next, in section 4 the data and method used in this thesis will be elaborated upon. Section 5 discusses the empirical results. Finally, in section 6 the conclusions of this thesis and recommendations for future research are provided. Also, some limitations of the thesis will be discussed.

2. Literature review

(6)

6

2.1 Simulation-based studies

The German electricity market has received ample attention of researchers in the past decade, mainly because of the energy transition, the Energiewende, that occurred. Because of the high penetration of renewables into the electricity mix, it makes the country a well-suited case for investigating the merit-order effect. Sensfuß et al. (2008) do this using an agent-based simulation platform and find that the electricity price was reduced by 6.1€/MWh in the period 2001-2006 due to increased renewable electricity generation. The authors argue that the costs for subsidizing the Energiewende are dramatically reduced once the reduction in electricity prices is recognized. They further find that the merit-order effect shifts profits from generation companies to consumers.

In a slightly different analysis, Weight (2009) investigates the extent to which wind energy is able to replace fossil fuels in Germany for the period June 2006 to 2008. No evidence favoring the hypothesis that wind generation capacity is able to replace fossil fuels is found, but the author does find evidence for a merit-order effect. The average electricity price is about 10€/MWh lower during the observed period, and the magnitude of the reduction typically increases when wind generation increases. Germany was not the only country where simulation exercises focusing on the merit-order effect have been applied. Sáenz de Miera et al. (2008) estimate the reduction of wholesale electricity prices in Spain due to increased renewable electricity generation from 2005 until 2007. Here, by focusing on wind as the RES, they find that an absolute negative correlation exists between wind electricity generation and retail electricity prices. They find magnitudes of the merit-order effect ranging from 4.75€/MWh until 12.44€/MWh for the years 2005 until 2007.

In conclusion, the authors that conducted simulation-based studies all found evidence favoring the merit-order effect, but the magnitude of this effect differs. Also, different time frames are used. Simulation-based studies might take different forms. As mentioned, Sensfuß et al. (2008) use an agent-based simulation model. Agent-based modelling allows for capturing complex dynamics and adaptive systems (Macal & North, 2010). Furthermore, parameters values and scenarios can be changed in order to see what effect this has on the total analysis. Sáenz de Miera et al. (2008) simulate the electricity price in absence of wind generation and compare this to a situation with wind generation. Simulation-based models thus allow for comparing different situations under different market conditions.

2.2 Regression analyses

(7)

7 electricity market for the period 2005-2013 and find empirical evidence of the merit-order effect. Using a least squares method, they find than an increase of 1 GW solar and wind production has, on average, reduced wholesale electricity prices by 2.3€/MWh and 4.2€/MWh respectively. Also, it is found that this increase amplifies price variance significantly. The authors also determine monetary savings due to solar- and wind production, leading to the conclusion that in the case of solar the monetary savings are not enough to offset the costs of the related support schemes. Since these costs are internalized entirely within end-user tariffs, consumer surplus is reduced. For wind production, they find that the opposite is true.

Gelabert et al. (2011) conducts an ex-post analysis to study the effect of increasing renewable energy production and cogeneration (where electricity and heat is produced at the same time) on wholesale electricity prices in Spain. They do this for the period 2005-2009 using a multivariate regression model and find that the merit-order effect is indeed present. The paper reports that a marginal increase of 1 GWh renewable electricity production will decrease prices with 1.9€/MWh. Next to this, the authors find that the magnitude lowers over the years, probably due to a higher contribution of natural gas to the Spanish electricity generation mix. Würzburg et al. (2013) follow a similar empirical model and estimate the merit-order effect in the German-Austrian electricity market for the period 2010 until 2012. Their major finding is that the day-ahead electricity price fell by roughly 1.0€/MWh for each GWh of renewable electricity generation. This effect is stable throughout different model specifications and the effect of solar and wind generation on electricity prices is rather similar.

The geographical scope of the research on merit-order effects has been limited mainly to European countries, though some efforts have been made to investigate countries on other continents as well. Woo et al. (2011) investigate the merit-order effect in the state of Texas, where they use high frequency data and separate Texas in four zones. On average, the authors find that a 1 GWh increase in wind generation decreases the Texan electricity price in the range of 13 US$/MWh to 44 US$/MWh. Cludius et al. (2014) examine the merit-order effect of wind in the Australian electricity market by using a time-series regression. For the years 2011-2012 and 2012-2013 they find that increasing wind generation by 1 GWh will decrease wholesale electricity prices by 2.30$/MWh and 3.29$/MWh respectively. These results are then used together with information on costs for RES support schemes in order to determine the effect on differing electricity users.

2.3 Comparing the approaches

(8)

8 Given the different approaches in analyzing the merit-order effect, the conclusions drawn from the existing studies are difficult to compare. However, this is not only the case for comparing simulation-based and regression analyses with one another. It is also difficult to compare studies that were conducted with similar approaches. For simulation-based studies, this difficulty might stem from the use of different assumptions and models. As discussed, Sensfuß et al. (2008) use an agent-based simulation platform in order to analyze the German merit-order effect. Weigt (2009) constructs a market model for the German electricity market, with and without the input of wind. Both papers find evidence for a merit-order effect, but given the different approaches the magnitude of the effect is difficult to compare. Also, for the regression analyses a generally agreed upon set of explanatory variables is missing. An example of this is that some authors included the price of fossil fuels (see Clò et al., 2015) as a determinant of electricity prices, whereas others did not (see Cludius et al., 2014).

Last, the merit-order effect has been analyzed for a couple of countries and differing time frames. Germany has been a major country of interest, probably due to their well-known

Energiewende. Spain has also proved itself to be a suitable candidate, because active support

for renewables has considerably expanded renewable electricity production (Gelabert et al., 2011). Though these countries are indeed interesting due to the high penetration levels of renewables in the electricity mix, it might also be interesting to study the opposite situation. Up until this moment, not much is known about the magnitude of the merit-order effect in countries with relatively low levels of RES in electricity generation. Also, the time frames studied differ across the studies. It is therefore hard to determine whether the merit-order effect is different across years. A summary of the existing literature can be found in table 1.

Table 1

Overview of the literature

Authors Method Country Timeframe Magnitude MO effect

Sensfuß et al. (2008) Simulation Germany 2001-2006 1.7-7.8€/MWh

Weight (2008) Simulation Germany 2006-2008 10€/MWh

Sáenz de Miera et al. (2008)

Simulation Spain 2005-2007 4.7-12.44€/MWh

Cludius et al. (2014) Regression Germany 2010-2012 6-10€/MWh Clò et al. (2015) Regression Italy 2005-2013 2.30-4.2€/MWh Gelabert et al. (2011) Regression Spain 2005-2009 1.90€/MWh Würzburg et al.

(2013)

Regression Germany-Austria

2010-2012 1.00€/MWh Woo et al. (2011) Regression Texas (USA) 2007-2010 13-44US$/MWh Cludius et al. (2014) Regression Australia 2011-2013 2.30-3.29$/MWh

(9)

9 setting. The objective is to determine the magnitude of the merit-order effect and see whether it differs across countries. It is relevant to compare the merit-order effect across countries as European electricity markets are becoming more integrated. This integration resulted in prices being equal some hours of the year, but it must be noted that most of the time prices are still different (TenneT, 2017). This is mainly influenced by the cost of producing electricity and by determining merit-order effects in different countries, one can draw inferences regarding European electricity price integration.

3 The electricity market

In order to grasp a better understanding of the problem discussed in this thesis, the working of the electricity market will be discussed. A fundamental property of electricity systems is that demand and supply must equal one another, because otherwise the electricity network will not be stable and power outages occur. This has some important implications for the way in which prices are formed in the electricity market. Section 3.1 will discuss the different markets in which electricity prices are formed, in order to clarify what is meant by ‘the electricity price’. In section 3.2, recent developments in the electricity market will be

described, with a focus on RES adoption and integration of electricity markets. In section 3.3, RES adoption in CWE countries will be discussed and in section 3.4, factors influencing electricity prices will be elaborated upon.

3.1 Pricing and electricity markets

Primarily, a market price establishes an equilibrium between demand and supply. Especially in the electricity market this is crucial, because it is (up until now) not possible to store large amounts of electricity. Therefore, careful planning is required in order to make sure the system will not collapse. This key task of balancing supply and demand is carried out by the day-ahead market, with an additional focus on the role of forward markets. Moreover, there is a final balancing process for adjustments in the real-time balancing market (APX Group, 2018). In case the system balance is distorted by market parties, they are penalized. Market parties have to communicate their electricity flows upfront in order to make sure the system remains balanced. These market parties can buy or sell electricity in the real-time balancing market, in order ensure they deliver the electricity they are supposed to deliver (i.e. the amount they communicated). The market parties are obliged to deliver the amount of

electricity they scheduled to deliver, otherwise they will have to pay an imbalance fee. In the end, this imbalance fee will be is channeled through to consumers, meaning they will pay higher prices for their electricity (Mulder, 2018).

(10)

10 from producers and consumers and based on this a market price is calculated. Day-ahead markets trade one day before actual delivery and are facilitated by a power exchange. This power exchange assists in buying and selling electricity for immediate delivery, as well as future delivery. Here, spot transactions ensure that electricity is delivered one day later, whereas futures transactions have a delivery date further away in the future. Electricity is delivered one day later, because otherwise a buyer had to immediately consume it after the purchase. Therefore, buying electricity on the spot market means it is delivered one day later.

The trade in electricity thus occurs at power exchanges, but it can also be traded (over-the-counter) by market participants without the power exchange as intermediate. This situation allows the market parties to establish non-standardized contracts, whereas the contracts traded at the exchange are often highly standardized. The former is the description of forward contracts, whereas the latter are called futures contracts. Basically, they are financial products that establish an agreement about the delivery of a certain amount of electricity in the future for a price agreed upon today. Mostly, market participants trade in these contracts to reduce their risk, typically known as hedging. Here, vulnerability to changing electricity prices is reduced. Electricity generators use these markets in order to guarantee future sales and protect themselves against price decreases. In contrast, large consumers might secure future consumption, while at the same time protecting themselves against price increases (TenneT, 2017).

The real-time balancing market is also referred to as the intraday market. Electricity is traded on the delivery day itself and allows market participants to correct for changes in their day-ahead electricity profile. These changes might occur because of unexpected generation problems or demand changes. Therefore, the intraday market is a last resort before market participants get penalized.

The day-ahead market is the market with the highest volume of electricity traded and number of participants. Therefore, the balancing market price in the day-ahead market is often referred to as the (wholesale) electricity price (TenneT, 2017). A well-functioning and competitive day-ahead market generally has two features representing the balancing market price. First, it is influenced by the cost of producing one MWh of electricity from the most expensive source required for market balance. Second, the price consumers are willing to pay for a final MWh is taken into consideration (Nordpool, 2017).

3.2 Developments electricity markets

(11)

11 markets. This reform of the electricity markets is particularly focused on fossil-free electricity generation. Consequently, the amount of electricity generated by large and renewable power plants, like large solar farms and wind parks, increased in the last decade. In other words, the energy transition we are undergoing currently is focused on the increased usage of

renewables in the electricity generation mix at the expense of fossil fuels.

As countries are gradually reducing electricity generation from fossil fuels, the focus should not be on domestic integration of renewables only. Another goal of the European Commission is to harmonize European power markets (European Commission, 2018). Here, the focus is on creating a pan-European market with increased electricity transmission capacity in order to improve the efficient use of electricity across borders.This process is the result of the liberalization of electricity markets, which started around the beginning of this millennium. In 2010, this resulted in the launch of the association of day-ahead markets of the CWE region (Tennet, 2019). In the years to follow, other countries joined and the agreement was renamed to Multi Regional Coupling (MRC). Market coupling optimizes the allocation process of cross-border capacities by coordinating price calculation and electricity flows between countries. Such a market coupling mechanism ensures that the market

responds to price signals. Therefore, market coupling might lead to price convergence across member countries of a market coupling initiative (EPEX Spot, 2019).

The extent to which integration is feasible, is constrained by electricity

interconnection capacity (the extent to which electricity can flow across borders) (Jamasb & Pollitt, 2005). Electricity interconnectors ensure a physical link between countries and therefore allow for electricity to be traded across borders. In order to achieve market integration and climate goals, Europe should improve these cross-border electricity

connections. If these interconnections function properly, surplus renewable electricity in one country could be used by another country. This way, Europe might be able to cope with the different energy realities its member states are currently in. Some countries are way ahead of others in terms of RES electricity generation, and allowing renewable electricity to flow across borders might speed up the adoption of renewables in Europe (European Commission, 2017).

3.3 Renewable electricity in CWE

(12)

12 Different geographical features or weather conditions affect the way in which a country generates its electricity. France is a rather mountainous country, making the country particularly well-suited for electricity generation by hydropower. As electricity is basically generated by moving water, benefitting from height differences, it makes sense that France implemented hydrogen into its electricity generation mix. Therefore, geographical

characteristics of countries partly determine in which form electricity is generated. It also seems rather intuitive that it is most efficient to build wind turbines at locations where the wind often blows and the same holds for installing solar panels in areas that receive a lot of sunshine. Also, is seems plausible that countries with relatively high average wind speeds tend to focus on wind turbines as their source of renewable electricity generation (Simoes et al., 2017). The potential for installing wind turbines in Belgium, France, and the Netherlands is high, given that these countries are located in an area in Europe with high average wind speeds (EEA, 2009). However, wind turbines are not adopted on a large scale yet, even though the countries plan on building them. This would increase electricity generation capacity by wind and therefore contribute to a less polluting electricity generation mix.

Thus, there are some country-specific geographic characteristics that made it

attractive to invest in renewable electricity generation units, but the reverse also holds. Ever since the Groningen natural gas field has been discovered, the Netherlands is highly reliant on natural gas (McKinsey, 2016). Natural gas is seen as a fossil fuel and therefore the Netherlands ranks rather low in terms of renewable electricity generation, as seen in table 2. The Dutch government has announced to put an end to gas extraction in the Groningen area, which means there is less supply for electricity generation by natural gas (Mulder, 2018). In order for the Netherlands not to lose generation capacity, electricity must be produced differently and renewable energy sources might be the alternative here.

Also, electricity generation mixes might be different due to differences in the policies implemented, designated to impose a shift to an electricity system with more renewable energy sources. The most prominent example is the Energiewende, to which has been referred to earlier in this thesis. The Energiewende encompassed an increased penetration of renewables at the expense of the traditional generation sources like coal and nuclear reactors, which substantially increased the likelihood of negative electricity prices (Fanone et al., 2011). Indeed, German data on electricity prices shows that the price is sometimes negative. Here, inflexible, non-programmable generation sources (like solar and wind) meet low demand and the market price decreases. Negative electricity prices are not a theoretical concept, because buyers actually receive money and electricity from sellers. However, producers will only produce electricity (thus sell electricity) if it is less expensive to keep the plant in operation (and selling at negative prices) rather than shutting it down and restarting it again later.

(13)

13 fossil free generation mix in 2050, meaning a lot will have to change in the coming years. Multiple European countries do this by phasing out coal-fired power plants in the coming years, because it is the most polluting electricity generation source. In the Netherlands, all coal-fired power plants will be decommissioned by 2030, whereas in Germany the coal-fired power plants will be phased out by 2038. Coal-fired power plants make up for a relatively large share of the electricity generation mix, so alternatives have to be found in order to keep electricity generation levels the same.

Not only coal-fired power plants are expected to be phased out in the near future. Even though they are not producing air pollution while operating, different countries are intended to decommission their nuclear fleet in the coming years. Germany decided it will phase out nuclear power production by 2022 (MCS, 2017). Nuclear safety is the main argument, following from the March 2011 Fukushima disaster. In Belgium and France, nuclear power plants are the source of the majority of the country’s electricity supply. Therefore, phasing these out requires alternatives for electricity production, because otherwise electricity supply will not be sufficient for the existing demand. In France, the nuclear fleet is ageing and the government intends to cut the share of atomic energy by 50% in 2025 (IEA, 2017). Similarly, Belgium wants to phase out all nuclear power plants by 2025 (IEA, 2017). However, this planned decrease in electricity production by nuclear power plants has to be offset by other sources. As has been stated earlier, a mismatch in demand and supply will result in power outages. The additional required capacity might come from renewable sources, but the countries have to be really careful in the way in which they substitute for the generation capacity provided by nuclear plants.

(14)

14 Table 2

Electricity generation by fuel in 2016 (Source: International Energy Agency)

Netherlands Germany France Belgium

Coal 34,3% 42,2% 1,9% 3,1% Oil 1,1% 0,9% 0,5% 0,2% Gas 46,9% 12,7% 6,3% 26,0% Biofuels 2,5% 7,0% 0,9% 5,2% Waste 3,2% 2,0% 0,8% 2,5% Nuclear 3,4% 13,1% 72,6% 51,2% Hydro 0,1% 4,0% 11,7% 1,8% Geothermal 0,0% 0,0% 0,0% 0,0% Solar 1,4% 5,9% 1,5% 3,6% Wind 7,1% 12,1% 3,9% 6,4% Tide 0,0% 0,0% 0,1% 0,0% Total 100% 100% 100% 100%

3.4 Factors influencing electricity prices

3.4.1 Electricity demand

Demand for electricity in itself is influenced by many factors. These can be climate factors, but also economic factors are important to distinguish. One of the most important climate factors is air temperature, since temperature determines the amount of cooling and heating required in a certain country over a certain time period. Furthermore, as nations grow, they tend to consume more electricity (Psiloglou et al., 2009).

Temperature might be a major determinant for electricity demand due to its clear link with heating and cooling. As temperatures get lower, mainly during winter periods, more electricity is demanded in order to heat buildings. This is referred to as the heating effect in literature. When temperatures are high, as is generally the case in summer periods, it might be that additional electricity is demanded for cooling purposes due to the increased usage of air-conditioning. This is the so-called cooling effect (Bessec & Fouquau, 2008). Mostly, authors use heating degree days (HDD) and cooling degree days (CDD) in order to capture this non-linearity (Valor et al., 2001; Staffell & Pfenninger, 2018). Valor et al. (2001) find that

sensitivity to weather changes is more significant in the cold season than in the warm season, thus implying that more electricity is used in colder periods. Staffell & Pfenninger (2018) find that weather has an increasing impact on electricity demand, and that this effect is amplified by electrification. As more buildings are heated with electricity, the demand for electricity will also rise in colder periods. Also, one can imagine that as countries grow in size (both in terms of people and GDP per capita), more electricity will be demanded. Therefore, larger (developed) countries often demand more electricity than smaller (developed) countries (Narayan et al., 2007). The way in which the economy is organized also affects the amount of electricity consumed, since specific sectors differ in their

(15)

15 is displayed (Psiloglou et al., 2009).

Mosquera-López & Nursimulu (2019) show that electricity demand positively affects electricity prices. They suuggest that this is particularly relevant in the spot market, and do not find that the effect of electricity demand on electricity prices is relevant in the futures market. Also, Clò et al. (2015) include demand as an explanatory variable for wholesale electricity prices and find that a positive relationship exists between electricity prices and electricity demand. Similar results were presented by Cludius et al. (2014), who find a positive relationship between load levels (so electricity demand) and electricity prices. 3.4.2 The merit order

The merit order ranks the available electricity generation units from low, to high marginal costs. Basically, the merit order describes the sequence in which electricity generation units are designated to deliver electricity, while supply is economically optimized. Firms in the electricity industry will sell output as long as their marginal costs are below the market price. These marginal costs typically include input costs (type of fuel) and maintenance costs, which obviously vary with production. Costs that do not vary with the quantity of the electricity produced, like investments, are not taken into consideration when making short-run productions decisions. Therefore, it must be noted that the merit order model is particularly well-suited for short-term electricity price formation (Cludius et al., 2014). Electricity generation units that produce continuously at very low marginal costs are the first to be called upon (i.e. supply electricity), and plants with higher marginal costs will be added until demand and supply are equalized.

As has been discussed, RES like solar and wind have approximately zero marginal costs. Therefore, these renewable sources are the sources with lowest input costs, implying that they will be called upon first in price formation. Imagine a world without solar and wind, where electricity production from nuclear sources would have the lowest marginal costs. As has been explained, the power plant with the lowest marginal cost will be the first to supply electricity, which is the nuclear power plant in this case. However, the nuclear power plant only has a certain production capacity and therefore it does not produce sufficient electricity in order to satisfy demand. The residual demand should therefore be met with other power plants, like gas-fired, coal-fired and oil-fired plants (Mulder, 2018). In this example, oil has the highest marginal costs for producing electricity, but it is still needed in order to satisfy demand.

(16)

16 wind production leads to a decrease in electricity produced by gas and oil fueled plants. However, nuclear electricity production and electricity produced with coal as input are still required in order to satisfy demand, because they have lower marginal generation costs than gas and oil. For sake of clarity, the demand curve was not depicted in this graph. As solar and wind generation increases, we see that the merit order is pushed to the right, meaning that oil and gas are no longer necessary for supplying electricity. As the most expensive producers are driven out of the market, ceteris paribus, the electricity price will decrease. This is referred to as the merit-order effect of renewables.

Figure 1

Stylized merit order effect (Source: Own illustration)

In section 3.3, the electricity generation mixes of the CWE countries regarded in this thesis were elaborated upon. It was discussed that all countries face the issue of power plants being phased out. Belgium and France decided to phase out nuclear power plants, whereas the Netherlands and Germany decided to phase out coal-fired power plants. The capacity decrease due to phasing out these plants should be balanced by an increase in electricity generation from other sources. In case these other sources are RES, merit-order effects might arise.

3.4.3 European emissions trading scheme (EU ETS)

(17)

17 of CO₂ and is referred to as a European Union Allowance (EUA). Within the cap in place, market players can freely trade these contracts in secondary markets. In case a market player does not utilize all its permits (so emits less than expected), it can either keep the permits for a future period or sell them to other market players .

In the first years after establishment, the EU ETS did not work as expected. The price of certificates remained lower than intended, with a large surplus of allowances, partly because of the economic crisis. This was during the first two phases of the EU ETS, which are often referred to as the learning by doing phase. Currently, the EU ETS is in its third phase, and is showing signs of becoming an efficient mechanism. The price of the certificates was still rather low at the beginning of the phase, but it has increased over the years (World Bank Group, 2018). This means that it has become more expensive to emit 1 ton of CO2,

which might promote the adoption of renewable energy sources. Under the EU ETS, prices are allowed to fluctuate, because the cap-and-trade system ensures the quantity is fixed (Mason, 2009). However, if it is not costly to emit 1 ton of CO₂, the system does not work properly. The desirable amount of emission reduction is namely found at the intersection between marginal costs and marginal benefits to additional greenhouse gas emissions.

As stated previously, allowances can be traded freely on the secondary market (within the boundaries of the cap-and-trade system). In the case of electricity generators, this means they have to carefully schedule upfront how much CO₂ they want to emit. When it is

expected that more CO2 is emitted, the electricity generators have to buy these certificates on

the secondary market. In other words, in case the process of electricity generation emits more CO₂, more certificates have to be bought. This will make it relatively more expensive to produce electricity with highly polluting sources, especially when the price of a EUA increases. So in case the price of EU ETS certificates increases, producing electricity with coal, gas and oil will become more expensive. The trade-off is between the marginal costs of renewable electricity generation and the marginal costs of fossil-fuel based electricity

(18)

18

4. Data & method

The dataset for the analyses comprises data from 1/1/2015 – 31/12/2018. In all cases, it consists of daily observations, which thus provides a comprehensive foundation for the analyses. It includes the day-ahead electricity price of different European countries; the estimated day-ahead load levels per country; the estimated day ahead feed-in of solar and wind; the day-ahead gas price in each country; the price of EU ETS certificates; a set of year dummies; a set of day of the week (DOTW) dummies. Table 3 provides the variable

definitions. Paragraph 4.1 includes the operationalization of the variables and paragraph 4.2 entails stationary tests of the variables, because this might have implications for the way in which I model the variables.

4.1 Sources and operationalization

The dependent variable is the day-ahead wholesale electricity price 𝑷𝒓𝒊𝒄𝒆_𝒄,𝒕 for each CWE

country c at time t. Though different electricity markets exist, most electricity is traded in the day-ahead market and the day-ahead price is often referred to as the electricity price.

Therefore, in this thesis the day-ahead price is referred to as the electricity price. In addition, it is typically utilized for maintaining a balanced portfolio (so that electricity demand and supply are equalized), while future and forward products are typically used for hedging or speculative reasons. The dataset of electricity prices contains wholesale electricity prices in EUR/MWh for every day of the year and data is extracted from the Bloomberg terminal. For some countries, hourly prices were available, so these had to be converted to average daily prices2.

The estimated day-ahead load for each country 𝑳𝒐𝒂𝒅_𝒄,𝒕 is included in order to

determine the influence of electricity demand on electricity prices. By using load levels, I can take seasonal variation into account. Al quantities are specified in MW and in order to make interpretation more logical, they are transformed to quantities in GW. This thesis focuses on a national scale, which means it makes more sense to work with GW. Data is available for every day of the year and was extracted from the ENTSO-E transparency platform, which receives data on electricity markets from the transmission system operator (TSO) of each country. Some countries contained data on an hourly level, but since this thesis concerns daily electricity prices, the average estimated daily day-ahead load was calculated. A benefit of calculating daily prices is that excessive noise is reduced (Gelabert et al., 2011). This day-ahead forecast is calculated on historic load profiles on similar days, taking into account factors that affect electricity demand like weather conditions, socio-economic factors and other climate factors. Day-ahead rather than actual load levels should be used, since they day-ahead load levels are relevant in day-day-ahead electricity price formation.

2_{Average daily prices were calculated as follows: (∑} _𝑃

𝑐,ℎ 24

ℎ=1 )/24, where h is one of the hours of the day. The

(19)

19 In order to determine the effect of increased solar electricity production, the estimated day-ahead electricity generation by solar 𝑺𝒐𝒍𝒂𝒓_𝒄,𝒕 will be used. Data was extracted from the ENTSO-E transparency platform and quantities are transformed from MW to GW in order to facilitate interpretation. Again, some countries have data available on an hourly level, so these had to be transformed to average daily levels. Next to the argument made earlier of reducing excessive noise, it also reduces intra-day price volatility that arises from intermittent electricity generation (Gelabert et al., 2011). The same data description applies to the variable 𝑾𝒊𝒏𝒅_𝒄,𝒕, which is the day-ahead generation forecast for electricity generation by wind. Data was extracted in MW, but rescaled to GW in order to facilitate interpretation.

A relevant variable for inter-fuel competition is the gas price. In order to determine the effect of changes in the gas price, 𝑮𝒂𝒔_𝒄,𝒕 is added as an explanatory variable. In most countries, gas makes up for a majority of the input for electricity generation, and it is added to the analysis in order to determine the effect of fossil fuel prices on electricity prices. The daily day-ahead price of gas for each individual country is included, which is relevant in my estimation of day-ahead electricity prices. Data was extracted from Bloomberg in

EUR/MWh, only for trading days. Therefore, weekends, national holidays and other days at which no trade was recorded are excluded from the sample.

In order to capture the effect of changes in the price for EU emission certificates on electricity prices, the variable 𝑪𝑶𝟐_𝒕 is included as an explanatory variable. Data was

extracted from Bloomberg in EUR/metric tonne CO2 and prices were determined on the

European Energy Exchange (EEX). These are spot auction prices for phase three of the EU ETS system and the prices are quoted for next day delivery. Therefore, it is suitable to include these prices in the analysis of day-ahead electricity prices.

Day of the week (DOTW) dummies are included (𝑫_𝒒𝒅, 𝑞 = 1,2, . . ,5) because there might be daily factors that are not observed. By doing this, weekly patterns can be controlled for (Mulder & Scholtens, 2013). Similarly, yearly indicators are included (𝑫_𝒒𝒚, 𝑞 = 1,2, . .4) in order to take yearly effects into account. The indicator variables capture yearly and DOTW effects that will otherwise not be measured in the model (Hill et al., 2012). Note that seasonal patterns are captured by the variable 𝑳𝒐𝒂𝒅𝒄,𝒕, because the forecast takes into account weather

(20)

20

Table 3

Variable definitions

Description Source Unit of measurement

Price Day-ahead electricity prices for CWE countries

Bloomberg €/MWh

Load Average daily load of CWE countries

ENTSO-E GW

Solar Average daily solar generation in CWE countries

ENTSO-E GW

Wind Average daily wind generation in CWE countries

ENTSO-E GW

Gas Day-ahead gas prices for CWE countries

Bloomberg €/MWh

CO2 Spot prices EU ETS system Bloomberg €/metric tonne

In appendix A table 7, the summary statistics for the variables are provided. Here, a distinction between the four countries is made. These summary statistics might indicate differences between country. It is shown that the mean electricity price over the sample period is highest in Belgium. Load levels are highest in Germany, and so are solar and wind generation. Also, the German gas price is highest of all four countries. Some countries have maximum values that are way above the average, which might indicate that outliers are present. An example of this is the Belgian electricity price. However, given the large amount of observations, outliers are not expected to be a problem. Furthermore, the summary statitics shown that the sample is unbalanced, as I do not have the same amount of observations for every country. The amount of observations for Belgium (N=720) is lowest, and Germany and the Netherlands (N=780) have the highest amount of observations. This it not a major issue, but might have implications for some tests I intend to use.

Appendix A, table 8 shows the correlation matrix between the explanatory variables. In order to determine whether the explanatory variables are correlated with one another and therefore detect multicollinearity problems, a correlation threshold level of 0.8 is used. In other words, when correlation is higher than 0.8 multicollinearity might be an issue. A fairly high correlation coefficient (0.6348) for the variables Gas & 𝐶𝑂2 is found, but this does not surpass the threshold of 0.8. At first glance, multicollinearity is not expected to be an issue.

4.2 Stationarity tests

(21)

21 regression model is specified, the least squares estimators do not have their usual properties, and t-statistics are not reliable. Clò et al. (2015) show that stationarity might be an issue when estimating merit-order effects, which is why stationarity tests are included in this analysis.

In order to determine stationarity in unbalanced panel data sets, a Fisher type unit-root test has to be used (Maddala & Wu, 1999). Fisher unit-root tests allow for using an

Augmented Dickey Fuller (ADF) specification and consequently an ADF test will be used. It tests the null hypothesis that all panels contain unit roots. In case this hypothesis is rejected, we can state that at least one panel is stationary. This does not necessarily imply that all panels are stationary, but according to the Fisher test this conclusion is sufficient for proceeding with the regression analysis.

Appendix B, table 9 shows the results of the ADF tests for stationarity. Given the low p-values, it is shown that the null hypothesis can be rejected at 1% significance level. In other words, the variables are stationary at level for at least one panel. No lags were added and given that the null hypothesis is rejected under these conditions there is no need to add lags to the model. In conclusion, under the given test conditions there are no unit roots in the panel, meaning this cannot cause spurious regressions.

4.3 Estimation method

This section describes the estimation method used in this thesis. As has been elaborated upon in the literature review, generally two methods exist for assessing merit-order effects, namely simulation-based studies and regression analyses. While simulation-based studies generally allow for ex-ante analyses of the merit-order effect, regression analyses estimate it ex-post. One important precondition for ex-post estimation is the availability of a large dataset. As described in section 4, an extensive dataset is used, which increases the reliability of the estimation. Also, the extensive dataset allows for utilizing a regression model. Furthermore, simulation-based studies are based on certain scenarios. My thesis does not concern a certain scenario to be investigated, but a general merit-order effect. Therefore, a linear regression model will be utilized.

Such a linear regression model helps to investigate in what way the variables mentioned contribute to the development of electricity prices. By doing this, the existence and strength of the relationship between electricity prices and the explanatory variables will be identified, with the aim of determining order effects. In order to determine the magnitude of the merit-order effects and compare it across countries, it would be convenient to have the effect of the explanatory variables on the dependent variable expressed in percentage changes. A way to do so is by transforming the dependent variable 𝑃𝑒_𝑐,𝑡 into a logarithm, while keeping the

(22)

22 (1) 𝑙𝑛𝑃𝑟𝑖𝑐𝑒𝑐,𝑡= 𝛽0+ 𝛽1𝐿𝑜𝑎𝑑𝑐,𝑡+ 𝛽2𝑆𝑜𝑙𝑎𝑟𝑐,𝑡+ 𝛽3𝑊𝑖𝑛𝑑𝑐,𝑡+ 𝛽4𝐺𝑎𝑠𝑐,𝑡+ 𝛽5𝐸𝑇𝑆𝑡+𝛿1𝐷_𝑞𝑑

+ 𝛿₂𝐷_𝑞𝑦+𝜀𝑡

A minor alteration will be made to the regression equation when estimating the country-specific magnitude of the merit-order effect. Slope-indicator variables are added to the equation in order to determine the country differences. A slope-indicator variable is similar to an interaction effect, and is determined by multiplying Solar and Wind by the country dummies. By doing this, I will be able to capture a shift in the slope of the relationship. When including indicator variables in log-linear regression specifications, one has to be careful with the interpretation. Approximation errors increase if differences between the regression coefficients increase (Hill et al., 2012). In case large differences between regression coefficients exist, the results should be carefully scrutinized. In order to capture the merit-effect order effect across countries, Germany will be taken as the base country. The log-linear regression equation will therefore look as follows:

(2) 𝑙𝑛𝑃𝑟𝑖𝑐𝑒_𝑐,𝑡= 𝛽0+ 𝛽1𝐿𝑜𝑎𝑑𝑐,𝑡+ 𝛽2𝑆𝑜𝑙𝑎𝑟𝑐,𝑡+ 𝛽3𝑊𝑖𝑛𝑑𝑐,𝑡+ 𝛽4𝐺𝑎𝑠𝑐,𝑡+ 𝛽5𝐶𝑂_2𝑡 + 𝛾1(𝑆𝑜𝑙𝑎𝑟 ∗ 𝐷1𝑐) + 𝛾2(𝑆𝑜𝑙𝑎𝑟 ∗ 𝐷2𝑐) + 𝛾3(𝑆𝑜𝑙𝑎𝑟 ∗ 𝐷3𝑐) + 𝜃1(𝑊𝑖𝑛𝑑 ∗ 𝐷1𝑐) + 𝜃2(𝑊𝑖𝑛𝑑 ∗ 𝐷2𝑐) + 𝜃3(𝑊𝑖𝑛𝑑 ∗ 𝐷3𝑐)+ 𝛿1𝐷𝑞 𝑑_{+ 𝛿} 2𝐷𝑞 𝑦_{+ 𝜀} 𝑡

Last, a slightly different regression model is estimated in order to determine whether the merit-order effect differs over time. The way in which this is approached will be similar to equation 2, the only difference being that the slope-indicator variables are calculated by multiplying yearly dummies with the variables Solar and Wind. The regression equation will therefore look as follows:

(3) 𝑙𝑛𝑃𝑟𝑖𝑐𝑒𝑐,𝑡= 𝛽0+ 𝛽1𝐿𝑜𝑎𝑑𝑐,𝑡+ 𝛽2𝑆𝑜𝑙𝑎𝑟𝑐,𝑡+ 𝛽3𝑊𝑖𝑛𝑑𝑐,𝑡+ 𝛽4𝐺𝑎𝑠𝑐,𝑡+ 𝛽5𝐶𝑂_2𝑡 + 𝛾1(𝑆𝑜𝑙𝑎𝑟 ∗ 𝐷1 𝑦 ) + 𝛾2(𝑆𝑜𝑙𝑎𝑟 ∗ 𝐷2 𝑦 ) + 𝛾3(𝑆𝑜𝑙𝑎𝑟 ∗ 𝐷3 𝑦 ) + 𝜃1(𝑊𝑖𝑛𝑑 ∗ 𝐷1 𝑦 ) + 𝜃2(𝑊𝑖𝑛𝑑 ∗ 𝐷2 𝑦 ) + 𝜃3(𝑊𝑖𝑛𝑑 ∗ 𝐷3 𝑦 ) +𝛿1𝐷_𝑞𝑑+ 𝛿2𝐷_𝑞𝑦+ 𝜀𝑡

(23)

23 characteristics. In this thesis, different countries are regarded and therefore it is important to capture this heterogeneity. Countries are different and these differences should be accounted for in the model. Therefore, FE models might be appropriate. Another model that might be used is the random effects (RE) model, where individual differences are treated as random rather than fixed (Hill et al., 2012). I can argue intuitively which model to use, but it can also be tested.

In order to determine whether a FE or RE model should be used, the Durbin-Wu-Hausman (DWH) test (or Durbin-Wu-Hausman specification test) is used. The test evaluates the consistency of a FE model in comparison to a RE model. Essentially, the DWH test checks for any correlation between the error term and explanatory variables (thus endogeneity). The null hypothesis states that a RE model can be used. In case the null hypothesis is not rejected, both RE and FE are consistent for estimating the model, but FE is said to be inefficient. When the null hypothesis is rejected, it is concluded that fixed effects should be used rather than random effects. Basically, if the term and the explanatory variables are correlated, a FE model should be used. Therefore, the DWH test can be used to differentiate between RE and FE models. In appendix B, table 10 the results for the DWH test are reported, and it becomes clear that the null hypothesis is rejected (p-value = 0.000). Therefore, it is concluded that (country) fixed effects should be utilized.

5. Results

In this section, the regression results will be provided and interpreted. First, the panel regression model for all countries together will be provided. Here, a country fixed-effects model is utilized. After this, the merit-order effect for each individual country will be

identified. It must be noted that all interpretations are based on the notion that all else remains equal. This line of reasoning will be used throughout the entire section.

5.1 Regression model general merit-order effect

The results of the regression are reported in table 4. In column 1, results are reported for a model including 𝐿𝑜𝑎𝑑 as explanatory variable. Before describing the results for the variables, it must be noted that the overall adjusted r-squared of the model is rather low (Adjusted R2 = 0.0051). Adjusted r-squared represents the proportion of sample variance in the dependent variable that is explained by the explanatory variables. A small adjusted r-squared is not necessarily an issue, but it does imply that the model might not be estimated correctly. It is shown that the regression coefficient (β = 0.0019) for 𝐿𝑜𝑎𝑑 is close to zero and significant (p = 0.000), meaning that the magnitude of the effect of electricity demand on electricity prices is limited.

In column 2, I add two variables and estimate the model again. For Solar (β =

(24)

24 case the average generation of electricity by solar increases with 1 GW, electricity prices are expected to decrease by 2.24%. Similarly, when the production of electricity by wind increases with 1 GW, electricity prices are expected to decrease by 1.26%. These results suggest that a merit-order effect is present.

Column 3 reports the regression results, including the variables for the gas price (𝐺𝑎𝑠) and for the CO2 price (𝐶𝑂2). The magnitude of the effect of a change in electricity demand on electricity prices is approximately zero (β = 0.0015) and significant (p = 0.000). The coefficients of 𝑆𝑜𝑙𝑎𝑟 (β = -0.0366) and Wind (β = -0.0200) remain significant (p = 0.000 for both). A 1 GW increase in solar production implies that electricity prices will decrease by 3.66% and a 1 GW increase in wind production is expected to lead to a decrease in electricity prices of 2.00%. Gas is significant (p = 0.000) and the regression coefficient (β = 0.0289) indicates a positive relationship between the gas price and the electricity price. Put differently, an increase in the price of gas by 1 EUR/MWh, will increase electricity prices by 2.89%. Furthermore, the estimation shows that CO2 is significant and positively affects electricity prices. The regression coefficient (β = 0.0176) indicates that an increase of the price of CO2 by 1 EUR/metric ton will increase electricity prices by 1.76%, all else equal.

Last, in column 4 DOTW and yearly indicators are included in the regression. The results are comparable to the third column in terms of signs and magnitudes. However, the model estimated in column 4 is the model with highest adjusted r-squared (0.4729), meaning that this is the preferred model. The magnitude of the effect of a change in electricity demand on electricity prices remains approximately zero (β = 0.0017) and significant (p = 0.000). The coefficients of 𝑆𝑜𝑙𝑎𝑟 (β = -0.0380) and 𝑊𝑖𝑛𝑑 (β = -0.0216) remain significant (p = 0.000 for both). Thus, a 1 GW increase in solar production will decrease electricity prices by 3.80% and a 1 GW increase in wind production decreases electricity prices by 2.16%. Gas is significant (p = 0.000) and the regression coefficient (β = 0.0306) is positive. The same holds for CO2 (β = 0.0276; p = 0.000). The interpretations of the coefficients for Gas and CO2 are comparable to the interpretations corresponding to column 3.

These results indicate that a merit-order effect is present in the Western European electricity market, given the negative relationship between RES generation and electricity prices. As electricity generation by solar or wind increases, electricity prices are expected to decrease. Furthermore, the results show that increase in solar generation will result in a larger negative effect than an increase in wind generation. This is in line with what was found by Clò et al. (2015), who investigated the merit-order effect in the Italian market. They also seperated renewable electricity generation in solar and wind, and found that an increase in solar generation will lead to a larger decrease in electricity prices, compared to wind. Würzburg et al. (2013) also separate electricity generation by solar and wind, but they did not find a notable difference between the two sources and their impact on electricity prices.

(25)

25

Table 4

Regression results merit-order effect

(1) (2) (3) (4) Load 0.0019*** (0.0004) 0.0019*** (0.0004) 0.0015*** (0.0003) 0.0017*** (0.0003) Solar -0.0224*** (0.0047) -0.0366*** (0.0037) -0.0380*** (0.0035) Wind -0.0126*** (0.0016) -0.0200*** (0.0012) -0.0216*** (0.0012) Gas 0.0289*** (0.0012) 0.0306*** (0.0013) CO2 0.0176*** (0.0013) 0.0276*** (0.0020) Constant 3.6007*** (0.0162) 3.6807*** (0.0199) 3.0329*** (0.0231) 2.9038*** (0.0328) Adjusted R2 0.0051 0.0249 0.4096 0.4729 N observations 3035 3035 3035 3035 Country fixed effects

YES YES YES YES

DY indicators NO NO NO YES

Note: Standard errors in parentheses

***, **, * indicate significance at 1%, 5% and 10% levels, respectively

5.2 Regression model country-specific merit-order effect

The results of the regression for estimating the country-specific merit-order effect are reported in table 5. The effects of Load, Solar, Wind, Gas and CO2 on electricity prices are comparable to the effects found in section 5.1. All explanatory variables are significant and the magnitudes are similar. Therefore, the interpretations of these coefficients can be found in section 5.1. Recall, the regression coefficients for the indicator variables are all relative to Germany. This means that the coefficients reported for each individual country will represent the additional increase or decrease as compared to Germany. In interpreting the coefficients, ‘additional’ will thus mean additional to the German effect.

(26)

26 to an additional 26% decrease, or a total effect of 29%. Last, for the Netherlands it was found that a 1 GW increase in solar production will lead to a larger electricity price decrease than in Germany, but the coefficient is not significant. A similar conclusion can be drawn when looking at the effect of wind production on electricity prices. Wind has a negative and significant coefficient (𝜃 = -0.0182; p = 0.000); WindBE has a negative and significant coefficient (𝜃 = -0.1247; p = 0.000); WindFR has a negative and significant coefficient (𝜃 = -0.0375; p = 0.000); and WindNL has a negative and and insignificant coefficient (𝜃 = -0.0106; p = 0.347). Thus, in Germany a 1 GW increase in wind generation will lead to a decrease in electricity prices of 2%. For Belgium, an additional price decrease of 12% is expected when wind production increases by 1 GW, or a total effect of 14%. For France, a 1 GW increase in wind production will result in an additional price decrease of approximately 4%, or a total effect of 6%. Similar to the case of solar production, the coefficient for wind production in the Netherlands is not significant, and therefore the effect of wind production on electricity prices in the Netherlands cannot be determined.

After adding the yearly and DOTW indicators to model (reported in column 2), the implications slightly change. The coefficients for SolarBE (𝛾 = -0.4590; p = 0.000), SolarFR (𝛾 = -0.2978; p = 0.000), WindBE (𝜃 = -0.1421; p = 0.000) and WindFR (𝜃 = -0.0440; p = 0.000) change, but remain significant. Furthermore, the coefficients of SolarNL (𝛾 = -0.1281; p = 0.008) and WindNL (𝛾 = -0.0296; p = 0.006) are significant now. Thus, a 1 GW increase in solar production in the Netherlands will decrease Dutch electricity prices with an

additional 13%, or a total of 16%. A 1 GW increase in wind generation, will lead to an additional electricity price decrease of 3%, or a total of 5%, in the Netherlands.

In conclusion, after including DOTW and yearly effects, all indicator variables are significant and have a negative coefficient. This means the effect of a 1 GW increase in solar or wind generation will have a stronger negative effect on electricity prices in Belgium, France and the Netherlands than in Germany. Especially in the case of solar generation, the effect on electricity prices is stronger for these three countries. Thus, the results indicate that a 1 GW increase in renewable electricity generation (both solar and wind) will have the strongest negative effect on electricity prices in Belgium. Furthermore, the magnitude of the merit-order effect is in all countries highest for an increase in solar generation.

(27)

27

Table 5

Regression results country specific merit-order effect

5.3 Regression model time-varying merit-order effect

Table 6 reports the results of my third regression model. Again, the main coefficients I am interested in are the coefficients corresponding to the indicator variables. The year 2015 serves as the base year, which means all changes are relative to 2015. This means that the coefficients reported for each individual year represent the additional increase or decrease compared to the year 2015.

(28)

28 Column 1 reports the findings for the first regression. Solar has a negative and

significant coefficient (𝛾 = -0.0607; p = 0.000), Solar2016 has a positive and significant coefficient (𝛾 = 0.0203; p = 0.002), Solar2017 has a positive and significant coefficient (𝛾 = 0.0424; p = 0.000) and Solar2018 has a positive and significant coefficient (𝛾 = 0.0231; p = 0.000). All coefficients for the indicator variables corresponding to Solar are positive and significant. Thus, 1 GW increase in solar generation will decrease electricity prices in 2015 by 6.07%. The positive coefficients for Solar2016, Solar2017 and Solar2018 imply that the negative effect of solar generation on electricity prices is lower in these years than in 2015. In other words, the decrease in electricity prices due to a 1 GW increase in solar generation is not as large as in 2015. Furthermore, Wind has a negative and significant coefficient (𝜃 = -0.0234; p = 0.000), Wind2016 has a positive and insignificant coefficient (𝜃 = 0.0044; p = 0.106), Wind2017 has a positive and significant coefficient (𝜃 = 0.0050; p = 0.036) and

Wind2018 has a negative and insignificant coefficient (𝜃 = -0.0007; p = 0.771). Given that the coefficient for Wind2017 is positive and significant, the electricity price decrease due to a 1 GW increase in wind generation in 2017 will be smaller than in 2015.

After including the DOTW and yearly indicators (in column 2), the coefficients slightly change. The coefficients for Solar (𝛾 = -0.0524; p = 0.000); Wind (𝜃 = -0.0216; p = 0.000); Solar2016 (𝛾 = 0.0170; p = 0.008); Solar2017 (𝛾 = 0.0225; p = 0.000); Solar2018 (𝛾 = 0.0168; p = 0.005); Wind2016 (𝜃 = 0.0035; p = 0.211); Wind2018 (𝜃 = -0.0011; p = 0.593) remain similar in terms of (in)significance to the model estimated in column 1, only the magnitudes differ a bit. The coefficient for Wind2017 is not significant anymore (𝜃 = -0.0013; p = 0.644). Therefore, it cannot be concluded that the negative effect of wind generation on electricity prices is either larger or smaller across years.

(29)

29

Table 6

Regression results time-varying merit-order effect

(30)

30

6. Conclusion

This thesis examined the effect of solar and wind generation on wholesale electricity prices in CWE countries. Due to the liberalization of European electricity markets, the CWE countries became more integrated. This resulted in an increasing number of hours with equal wholesale electricity prices. Also, as a result of the Paris Agreement, countries are increasing the RES share in electricity generation. CWE countries intend to phase out large fossil fuel and nuclear power plants in the coming years, but they must ensure that this decrease in electricity generation is offset by other sources. In this case, RES might be the alternative. However, replacing existing generation capacity with RES generation capacity might give rise to merit-order effects. Given the integration of CWE electricity markets, it is relevant to investigate this merit-order effect across countries and see whether it is different. In order to do this, I made use of a linear regression model including country fixed-effects. This way, I was able to allow control for unobserved heterogeneity across countries. Different research methods were used to study merit-order effects, which makes it hard to compare findings. I fill this gap in the literature by researching the order effect in a multi-country setting, which means merit-order effects can be compared across countries.

First, the results presented in this thesis suggest that a merit-order effect exists in the CWE electricity market. Thus, an increase in renewable electricity generation has a negative effect on wholesale electricity prices. It was found that solar generation has a larger negative effect on wholesale electricity prices than wind generation. A 1 GW increase in solar generation will result in a 3.80% decrease in electricity prices, whereas a 1 GW increase in wind generation will lead to a 2.16% decrease in wholesale electricity prices. This is in line with the findings of (Clò et al., 2015), who also seperated renewable electricity generation into solar and wind and found that the negative effect is largest for solar. Second, my findings suggest that this merit-order effect differs across CWE countries. The negative effect of an increase in solar or wind generation on wholesale electricity prices is smallest in Germany and largest in Belgium. Furthermore, in all CWE countries the negative effect of solar generation on electricity prices is larger than the negative effect of wind generation on electricity prices. Third, it was shown that the negative effect of solar generation on wholesale electricity prices is consistent for the period 2015 until 2018. The negative effect was largest in 2015, since a 1 GW increase in solar led to a 5.24% decrease in electricity prices. For wind generation, it cannot be concluded that the effect on electricity prices is different throughout years.

(31)

31 that electricity prices are reduced. Therefore, it might be that the merit-order effect is larger for countries with a relatively high share of fossil fuels. Also, future research might focus on studying why the merit-order effect differs over time. Gelabert et al. (2011) found that the merit-order effect decreased over time due to an increase of natural gas in the electricity generation mix in those years. This might also be the case for the countries in this analysis. Also, the analysis could be extended to retail electricity prices. In case the effect of renewable electricity generation on retail electricity prices can be determined, welfare effects for end-users might be calculated.

THE EFFECT OF RENEWABLE ELECTRICITY GENERATION ON ELECTRICITY PRICES

THE EFFECT OF RENEWABLE

ELECTRICITY GENERATION ON

ELECTRICITY PRICES

Table of Contents

Abstract

JEL classification: Q21, Q42

1. Introduction

2. Literature review

3 The electricity market

4. Data & method

5. Results

6. Conclusion