The Effect of Intermittent Renewables on Electricity Forward Premia: Assessing the Role of Flexibility

The Effect of Intermittent Renewables on Electricity Forward Premia:

Assessing the Role of Flexibility

Bastiaan Bruijn*

Supervised by prof. dr. M. Mulder June 11, 2019

Abstract

Intermittent renewables, such as wind and solar power, are increasingly contributing to the world’s electricity supply. Intermittent renewables bring uncertainty of supply to the market, and therefore increase price-uncertainty and price volatility. This uncertainty can be hedged using forward contracts, but this comes at a price; the forward premium. The uncertainty inherent with intermittent renewables may increase this premium. Prior research has indeed found positive effects of intermittent renewables on the electricity forward premium, explained by the merit-order effect. In expectation, flexibility can mitigate the effect of uncertainty of intermittent renewables on the forward premium. A possible explanation is that flexible assets, such as flexible production plants and cross-border capacity, can be deployed to absorb the supply shocks of intermittent renewables. This paper examines the role of flexibility on this effect, using data from 2013 – 2018 from Belgium, Germany, France and the Netherlands. We contribute to the existing literature by testing if flexible assets can mitigate the positive effect of intermittent renewables on ex-ante forward premiums. We find a significant negative coefficient of the interaction term of intermittent renewables and flexible assets, implying that the presence of flexible assets can mitigate the positive effect of intermittent renewables on forward premiums. This underlines the importance of flexible assets in reducing uncertainty in the transition towards cleaner energy.

Keywords: Finance, Derivatives, Forward Premium, Energy, Electricity, Intermittent Renewables Word count: 15,368

This thesis will also be used for the Energy Certificate

* _{Faculty of Economics and Business, University of Groningen, PO Box 800, 9700 AV Groningen, the Netherlands.}

1

1 I

NTRODUCTION

In recent years, renewable energy sources (RES) have undeniably increased its presence in the world’s electricity supply. In the United States, RES consumption has grown with 13.9% annually between 2006 and 2016, and in China even with a staggering 42.6% per year. The European Union is not far behind the US with 13.2%. This does not mean that RES is of great significance yet, as RES still only supply 13.2%, 23.9% and 29.2% of total electricity in the US, China and the EU respectively1_.

Most of these RES carry the characteristics of having negligible to no marginal costs. The low marginal costs of production make renewables a competitive electricity source, at least in the short run. In a perfect market, electricity, such as any other normal good, is first supplied by the producer that supplies at the lowest marginal cost, then the producer with the second-lowest marginal cost comes in and this process continues until supply equals demand. This results in a convex supply curve where the industry marginal cost increases gradually with quantity produced, such as in any other market. The electricity price during this pricing regime is equal to the industry’s marginal cost; the so-called marginal cost pricing system. Electricity sources with production costs so high that they never enter the market naturally cannot have an impact on the pricing process in the market, but this is not the case for RES, given their low marginal costs, which thus do have an impact on this process. If demand stays equal, then varying levels of RES supply cause price changes through horizontal shifting of the supply curve. This shifting of the supply curve, also known as the merit order, is known as the merit-order effect. Also, as RES have low marginal costs, RES electricity generation should have a negative impact on electricity prices. Research using data from France, Germany and the Netherlands has indeed identified negative effects from RES generation on spot prices (Mulder and Scholtens, 2013; Dillig et al., 2016; Keppler et al., 2017).

Intermittent RES (IRES), such as wind and solar energy, rely on non-controllable inputs; wind speed and sunshine in the case of wind and solar energy. If these weather conditions are volatile, i.e. wind speed and sunlight vary over time, then so is electricity supply from these sources2_{. RES}

are thus less reliable than conventional electricity sources and, given the pricing process

2

explained above, varying supply (from RES) causes more variation in prices: price volatility. This variation in supply shifts the supply curve horizontally. Many papers, including Paschen (2018), Fischer (2010), Würzburg et al. (2013) investigate how RES generation affects electricity prices and explain this with the merit-order effect.

Volatile supply does not necessarily mean uncertain levels of supply, though. Weather forecasts can be made, albeit with decreasing accuracy the farther away from the present. Prices that are perfectly predictable can still be volatile. However, most electricity is traded for periods in the future for which weather conditions cannot be accurately predicted (Falbo et al., 2014). Energy companies are therefore not only exposed to more volatility of electricity supply and thus to price volatility (through the merit-order effect), but also to some extent to uncertainty of supply levels depending on the accuracy of weather forecasts.

The uncertainty of future supply leads to an increased demand for insurance against adverse price movements. Electricity market participants get this insurance by buying electricity for periods in the future, using forward contracts. In these forward contracts, the price is fixed at the time that two parties enter into such a contract. This price is then the forward price for a given period in the future over which electricity will be delivered. This forward price theoretically consists of two components: the expected spot price and an insurance premium. The buyer of the forward contract of course only sees the forward price. The process of insuring against adverse price movements by using financial products such as forward contracts is known as hedging. The demand for hedging in electricity markets is dependent on the level of uncertainty about future prices in a market, which among other things depends on the flexibility offered by the market, for example in the form of interconnections with grids from other regions or from flexible electricity sources such as coal- and gas-fired power plants. Hout et al. (2014) investigate how increased price volatility affects hedging behavior, and they expect increased volatility of future electricity supply to increase demand on day-ahead and intraday markets, where energy companies adapt their portfolios to new information on production and consumption, after buying or selling on forward markets3_.

The purpose of this research is to empirically analyze how flexibility affects the relation between RES generation and forward premia. Previous empirical research has reported a positive effect of RES generation on ex-post forward premia, which is explained by renewables decreasing the spot

3

price through the merit-order effect, resulting in a larger ex-post premium. This paper will assess the impact that flexible assets, consisting of cross-border capacity, coal- and gas-fired power plants and hydropower plants, may have on this relation. To the best of the author’s knowledge, there has been no research on how flexibility affects the effect of intermittent renewables on the electricity forward premium.

The central research question throughout this paper therefore is how the presence of flexibility in an electricity market affects the effect of RES generation on electricity forward risk premia. This will be done by analyzing panel data from countries in the Central Western European (CWE) market-coupling region: the Netherlands, Germany, Belgium and France. These countries are chosen because they are all interconnected but have very different characteristics in terms of generation mix and cross-border capacity and thus differ in the presence of flexible assets on supply side, where generally fossil fuels, hydropower and a high degree of interconnection with neighboring countries are the more flexible and nuclear and renewable power the less flexible means of electricity production (figure 1).

4

Figure 1: CWE electricity generation capacity (in %) and CWE production, import and export (in %), 2013

Sources: Elia, 50Hertz, Amprion, TenneT, TransnetBW, RTE

0 2 0 4 0 6 0 8 0 1 0 0 %

Belgium Germany France Netherlands

Generation Mix

Hydro Nuclear Wind Solar Fossil Other

0 2 0 4 0 6 0 8 0 1 0 0 %

Belgium Germany France Netherlands

Production, Import and Export

5

2 L

ITERATURE REVIEW

2.1 R

As described in the introduction, RES affects spot prices through the marginal cost pricing mechanism. In this mechanism, prices are equal to the marginal production cost of the marginal producer. Through the merit-order effect, the varying levels of supply of RES cause spot prices to vary as well.

6

Figure 2: Price divergence in CWE, Q3 2018

Source: Market Observatory for Energy of the European Commission (2018)

2.2 I

RES

,

Having determined that RES affect electricity spot prices, we will now briefly review literature that deals with how RES and flexible assets jointly affect price volatility.

The combination of intermittent renewables and flexibility in the market used by Wirdemo (2017) is particularly interesting for this research, as this combination is the main subject of this research. However, this research assesses the joint effect of intermittent renewables and flexibility on forward premiums, while Wirdemo (2017) investigates the impact of the joint effect on the Swedish electricity price volatility. A GARCH model is made, using daily data from the Nord Pool market from 2015 through 2017. He reports increased electricity price volatility due to wind power production, but this is partially counteracted by the flexibility supplied by hydropower reservoirs.

7

renewables. They use data for the years 2012 – 2014, predictions by Entso-e for the years 2013, 2015, 2016 and 2020, and linearly interpolate missing years to form a dataset covering 2012 – 2023.

2.3 F

-

The leading paper in the Finance literature about premia in electricity forward contracts is written by Bessembinder and Lemmon (2002). Although this is rather old literature, the theory explained below is still valid and most of the literature that followed in the field of forward premiums builds upon this paper, some of which are described below.

They argue that “the inability to store power means that the no-arbitrage approach to pricing derivative securities cannot be applied in the usual manner.”

8

Figure 3: No-arbitrage pricing visualization

Source: own production

The negative relation between variance and the premium seems counterintuitive, but the explanation is as follows. At times when more retail power is sold, wholesale prices are generally higher, for example during peak hours. As long as retail prices are above wholesale prices (which on average is the case), profits of power retailers are positively exposed to wholesale prices (high prices and high output means high profits). The assumption hereby is that large profits during base hours have been competed away. The positive exposure of revenues to power prices leads to a demand for hedging, where retailers want to take short positions in forward contracts, to offset their positive exposure in order to reduce their overall risk. By (short-)selling forward contracts, they place a downward pressure on the forward price and thus, ceteris paribus, on the

FOJ forward market

Counterparty

Short-sell forward contract for 100 liters of frozen orange juice, delivery in 1 month

gives rise to obligation to deliver in the future -->

Deliver 100 liters of frozen orange juice to

counterparty Buy 100 liters of frozen

orange juice on the spot market, financed with a loan at interest rate r

Store orange juice for 1 month, incurring storage costs c

Repay loan with proceeds from delivery of orange juice

+ interest

9

forward premium. However, the explanation by Bessembinder and Lemmon (2002) has not been undisputed in the empirical literature. Redl and Bunn (2012), for example, find a positive relation between variance and premiums and explain this by arguing that shocks creating high skewness (due to the convex supply curve) also cause higher variance and should therefore have the same sign; the positive one. High skewness and variance then cause market participants to hedge against adverse future price movements. Important to note is that both papers refer to the ex-post premium which is influenced by spot price movements that occur after forward contract closure. Supply shocks thus increase the ex-post premium while at the same time causing variance and skewness. There might therefore not necessarily be a causal relation.

In general, electricity becomes more expensive as time-to-delivery decreases. On the short term there are less possibilities to flexibly respond to changes in demand, which increases the chances of both scarcity (which can lead to scarcity prices) and overproduction (which can even lead to negative spot prices). It is also more expensive to change production levels as production plants suffer from start- and stop costs when they do change production levels. These costs will of course be incorporated in the marginal cost, which in turn sets the price.

10

Figure 4: Simple electricity supply-and-demand framework

Source: own production

Figure 5: Distribution of base load (left-hand side) and peak load (right-hand side) power prices (EUR/MWh)

Source: Redl and Bunn (2012)

An important electricity generation input is gas. So, electricity prices depend greatly on gas prices. Gas prices in turn depend on, among many things, gas storage levels. Douglas and Popova (2008) examine the relation between gas storage levels and the forward premium, and build upon the forward premium model by Bessembinder and Lemmon (2002) while adding variables related to gas storage levels and weather data (cooling and heating degree hours). They hypothesize that as

11

constraints in potential gas supply increase, gas price spikes become more likely, increasing the skewness of the gas price and in turn of the electricity spot price. Using hourly spot and day-ahead forward prices from PJM from January 2001 through December 2004, they find a significant negative relation between gas storage levels and the electricity forward premium. Low gas storage levels increase the probability of price spikes and thereby increase demand for insurance against these price spikes. Thus, a net increase in demand for hedging follows. Similarly, increased gas storage levels decrease these supply constraints in the gas market and therefore decrease the forward premium. Redl and Bunn (2012) also use the Bessembinder and Lemmon (2002) paper as a basis and then add variables. They expand the model by adding forward premium determinants capturing behavioral, dynamic, market conduct and shock components (e.g. supply shocks). They identify as forward premium determinants the gas forward premium, scarcity in the electricity supply system, variance, skewness, kurtosis and spikes of the electricity spot price, volatility of the Brent oil spot price, market power in the spot market, the electricity basis (the difference between the forward price and spot price) and shocks in supply margin during the delivery month. Some of these explanatory variables may affect the distribution of spot prices and thus influence skewness and kurtosis, but Redl and Bunn (2012), nor any authors citing this paper, do not comment on potential multicollinearity which would lead to errors in the results obtained with the regression.

They report only scarcity in the electricity supply system to have a negative impact on the premium, all other variables are positively related. However, this is because scarcity is measured by the generation/consumption ratio, which is essentially the inverse of scarcity. For this ratio a negative coefficient is obtained, so the conclusion from this is that scarcity has a positive effect on the power premium. They draw data from the European Energy Exchange (EEX) and use month-ahead forwards, covering the period from October 2003 through January 2010.

The gas forward premium directly affects the electricity forward premium due to the abundance of gas-fired generation plants in the EEX market. Higher gas premiums increase these plants’ expected future marginal costs of production and consequently the forward price.

Scarcity in the electricity supply system leads to a higher probability for spikes to occur. This probability for spikes should then increase the forward premium, so there is a negative relationship between the generation/consumption ratio (the inverse of scarcity) and the forward premium.

12

market participants want to hedge against extreme outcomes, both positive and negative (which is exactly what kurtosis measures), increasing the demand for hedging and therefore kurtosis will be positively related to the electricity forward premium.

Sentiment in the oil market, measured by volatility in that market, has a spill-over effect into the electricity market. This comes from the fact that oil prices, both directly and indirectly through gas prices, affect electricity prices. Volatility in the oil market increases the demand for hedging in the electricity market and therefore increases forward premiums.

Forward contracting can, among other things, serve to mitigate the strength of market power of players in the spot market. More concentrated market power should therefore induce parties to work around this market power and should therefore increase the demand for forward contracts, consequently increasing the forward premium.

An increasing basis, the difference between the spot price and the forward price, is expected to increase the forward premium, simply because the future basis is estimated based on the current base.

Margin shocks, measured by the change in the supply margin, are added to the model to account for unexpected consumption or generation shocks, that will affect the spot price and therefore the ex-post forward premium measured at that time. If, for example, total generation is unexpectedly high, spot prices should fall below predicted forward prices since the supply curve is shifted to the right. The phenomenon called the merit-order effect occurs. With equal forward prices, this increases the ex-post premium.

13

low and running out of water in the reservoir (this happens when reservoir levels are below the equilibrium level, corresponding to a positive convenience yield) and the risk of overflow for the hydro producers when reservoir levels are high (and when there will be a lot of supply). Another way of explaining that is that high (low) reservoir levels give rise to the expectation that in the future there will be more (less) supply of electricity and therefore a lower (higher) forward price is expected.

There is a theoretical equilibrium reservoir level at which the forward premium would be zero, and due to seasonal patterns, reservoir levels are more often above than below this equilibrium level throughout the year and therefore they find on average negative convenience yield. The premia with the corresponding reservoir levels can be found in figure 6.

Figure 6: Average convenience yields (in %) on weekly forwards with 1 week and 6 weeks holding periods compared to average reservoir levels in Norway, 1996-2006

Source: Botterud et al. (2010)

An important distinction to make, however, is that the timing of hydropower supply can be controlled, as opposed to the intermittent RES sources, which can only be switched off in case of oversupply to avoid grid congestion.

14

is realized (see figure 7 below). Effectively, the spot price is (unexpectedly) reduced and while the forward price is fixed, the differential between the, lower than anticipated, realized average spot price and forward price (the ex-post premium) increases. The increased ex-post premium can be interpreted as a risk mark-up on forward prices resulting from higher intermittent wind feed-in. His explanatory variables are derived from the multi-factor analysis of premia in Redl and Bunn (2012), which can be traced back to Bessembinder and Lemmon (2002). Paschen (2018) reports significant positive wind shock effects on both monthly and daily ex-post forward premia. A relationship between weather conditions and forward premia thus seems to exist, and therefore one would expect the increased RES generation to increase forward premia. Given that electricity markets in the CWE region are increasingly integrated, a spill-over effect of oversupply could occur from one country to another, increasing forward premia across borders. On the other hand, the country dealing with the oversupply should see less price shocks as part of the oversupply is exported. Overall more market integration means less volatile prices.

Figure 7: Illustration of the merit-order effect

15

due to convexity of the supply curve (similar changes in quantity result in greater price changes). Solar shocks show no significant effect.

2.4 E

-

17

3 E

LECTRICITY MARKETS

What follows now is a brief description of the organization of the European electricity network, an introduction to electricity markets, pricing on these markets and what determines volatility and uncertainty on these markets.

Figure 8: Organization of European electricity systems

Source: European Parliament (2016)

3.1 E

European electricity systems consist, as depicted in the simplified figure 8, of six main types of players: electricity suppliers, electricity generators, transmission system operators, distribution system operators, consumers and regulators. Electricity suppliers buy electricity from generators and sell it to consumers. Transmission system operators transport electricity across long distances and ensure stability of the system. Distribution network operators distribute electricity, supplied by transmission system operators, to consumers. Consumers buy electricity from suppliers in the retail market. Suppliers buy electricity from generating companies on the wholesale market. Large industrial consumers buy directly from generators on the wholesale market. The market of interest for this research, is the wholesale market.

18

Electricity trading can take place in various ways, such as through bilateral trade, over-the-counter and on exchanges. In the first two, arrangements can be made to tailor the product towards the needs of the customer. On exchanges, on the other hand, standardized products are offered, members trade anonymously, and financial risks are assumed by the exchange. Only exchange prices are publicly available, which is why, for the remainder of this paper, only exchange markets will be considered.

3.2 E

Electricity differs from other commodities mainly in the fact that it cannot be delivered until it is consumed (Mulder, 2017). The network must constantly be balanced (i.e. generation equals load), or the entire system will fail. This also means that, besides classification by geography, wholesale electricity markets are also classified by time scales. These shall be explained, working from the future to the present. For Germany, France, the UK, the Netherlands, Belgium, Austria, Switzerland and Luxembourg, the wholesale market is the European Power Exchange (EPEX). However, this does not mean that there is only one EPEX spot price. In fact, each market area has its own electricity price.

3.2.1 Futures markets

Futures in the CWE region are traded on the European Energy Derivatives Exchange (ENDEX). ENDEX offers two standardized products based on the delivery hours of electricity. Per country, the exchange offers a “Power Base Load” contract and a “Power Peak Load” contract. Base load contracts cover delivery for the entire delivery period of a base load. Peak load contracts deliver electricity from 08:00 to 20:00 on weekdays, regardless of public holidays4_{. Both contracts are}

offered for various lengths of time to maturity, such as month-ahead, quarter-ahead and year-ahead.

As figure 9 illustrates, prices and dispersion around the mean, increase with time to maturity. It is for that reason that 95% of electricity is traded on futures markets (Falbo et al., 2014).

19

Figure 9: Price and dispersion around the mean increasing with time to maturity

Source: Falbo et al. (2014)

3.2.2 Spot markets

Formally speaking, there is only one spot market comprising real-time trading, which is the imbalance market. However, in the context of electricity markets, the day-ahead market and the intraday market, which form part of the EPEX spot market. are regarded spot markets.

3.2.2.1 Day-ahead market

Once trading for the delivery period in question is no longer facilitated by future markets, traders turn to the day-ahead market. On the day-ahead market electricity is traded for physical delivery the next day. The day-ahead market closes around noon although this varies across markets. 3.2.2.2 Intraday market

The closing of the day-ahead market does not mean that there is no more trading possible for delivery the next day or even the same day. Market participants can update their positions up to 5 or 30 minutes before physical delivery takes place, depending on the country5_.

3.2.2.3 Imbalance market

Some countries have an imbalance market, too. In countries where there is no imbalance market, balance is centrally maintained by charging fixed rates to the parties.

20

On the imbalance market, grid balance is maintained by the transmission system operators who buys and sells any excesses or shortages from the party that is responsible for the region where the imbalance occurs.

3.3 P

As long as an electricity market has access to sufficient generating capacity, prices are based on supply and demand bids. The supply bids are based on the marginal costs from generation plus opportunity costs and a risk markup for technical breakdowns. The opportunity costs arise from missing out on revenues not gained in the future, as conventional generation plants can only use inputs (gas and uranium for instance) once. For example, if a gas-fired plant sells 1 MW of electricity in the day-ahead market and uses a certain amount of gas in the generation

process, it cannot use the same gas to sell electricity in the intra-day market, which may be more profitable. Prices and quantities of electricity are in this regime based on a supply-and-demand framework, as shown in figure 10. The pricing mechanism changes when shortage in capacity occurs. In order to set demand equal to supply, so-called scarcity prices emerge, which are not based on supply-side marginal costs anymore, but on demand-side willingness to pay (Mulder, 2017).

Figure 10: Simple electricity supply-and-demand framework

Source: own production

21

3.4 V

Price volatility arises from movements of the demand curve and/or the supply curve. It is thus either demand-side induced or supply-side induced.

3.4.1 Demand-side induced volatility

Demand for power exhibits strong seasonal patterns within a day, week, month and year. (Sedidi et al., 2014). Given an upward sloping supply curve, this variation in demand c.p. induces variation in prices. This change is illustrated in figure 10. A change in quantity from Q1 to Q2 c.p. leads to a price change from P1 to P2.

Since the supply curve is increasingly upward sloping the variation is bigger for changes in relatively high demand levels. These seasonal patterns in demand are quite predictable, but there can also be less predictable changes in demand, temperature and various market conditions. 3.4.2 Supply-side induced volatility

Electricity price volatility can also be a spill-over effect from volatility in inputs. At times when the marginal-cost-pricing regime prevails, input price volatility resonates in the electricity market though marginal prices. This will then shift (parts of) the supply curve vertically. Increased input prices c.p. result in higher prices and lower quantities of electricity.

The supply curve can also shift horizontally. This happens when a generation technique supplies different quantities of electricity than previously. An example of this is when wind turbines supply more (less) power as a result of more (less) wind velocity. The entire supply curve then shifts to the right (left), resulting c.p. in lower (higher) prices.

3.5 U

22

To some extent, we can predict the weather for some day next week, but there is still a lot of uncertainty involved. This uncertainty usually decreases as time brings us closer towards that day, as our ability to predict the weather increases as the forecast is closer to the present. Many participants in electricity markets enter into forward contracts for months, quarters or years ahead and therefore cannot use reliable weather forecasts. Also, on supply side, other supply-side uncertainties such as unexpected technical problems can occur.

Another major source of uncertainty is uncertain demand, depending on, among other factors, meteorological factors as well. The weather thus incorporates uncertainty through both the supply and the demand side of the market. Other sources of uncertainty can be macroeconomic, GDP growth is strongly associated with electricity usage, and regulation, for example support schemes (Hirsh and Koomey, 2015).

3.6 F

Hout et al. (2014) classify flexibility in electricity markets into two types, flexibility from generation, and flexibility from other options such as storage, demand response and interconnection. They predict that electricity increasingly generated by RES increases volatility in production. Flexibility is needed to accommodate this volatility. A significant part of this flexibility will have to be offered by flexibility from generation, in particular by gas-fired power plants. Other important, currently available sources of flexibility are interconnections. The importance of interconnections is also stressed by Child et al. (2019), who state that a total of 12% of end-user demand in Europe is provided by interconnections between regions. Interestingly, they find that the highest levels of interconnections are found in areas rich in wind, solar or hydropower resources, and in densely populated and industrial areas. Therefore, it seems to be the case that regions with high electricity supply from RES and high electricity demand in general have access to the highest capacities of interconnections.

23

4 T

HEORY

From the literature review above, theories will be extracted and explained in this chapter. The first theory concerns the relation between RES and electricity spot prices (section 4.1), the second theory deals with how flexibility in a market affects this relation between RES and electricity spot prices (section 4.2). Section 4.3 explains how intermittent renewables affect the forward premium, section 4.4 puts it all together and elaborates on the combination of intermittent renewables and flexible assets and the effect on forward premiums.

4.1 I

24

Figure 11: Merit-order effect

Source: own production

Increased IRES supply, indicated by the green line, causes the entire supply curve to shift to the right, c.p. intersecting the demand curve at a higher quantity and a lower price level.

price

intermittent renewables supply other supply demand P1 P2 Q1 Q2 quantity price

intermittent renewables supply other supply

demand

P1 P2

25

4.2 F

A market’s flexibility is characterized by the extent to which it can efficiently respond to changes in demand. A nuclear power plant (NPP), for example, has every possibility to ramp production up and down. However, economically it usually does not make sense to do so as NPPs require high amounts of capital expenditures and therefore the plant operators want to produce at maximum capacity all the time to cover investment expenditures6_.

The effect of flexibility on electricity spot prices can be best explained in cases of extremes. In chapter 3, scarcity prices were introduced. In a market with a low degree of flexibility, extremely high demand is more likely to induce scarcity prices than in a market with a high degree of flexibility (Wirdemo, 2017; Nicolosi, 2010).

This is because a market with a high degree of flexibility can, albeit at above-average marginal cost, still produce enough electricity to prevent scarcity prices.

In the other extreme, where there is unusually low demand, a market with a low degree of flexibility is overproducing, causing prices below marginal cost which are needed to set demand equal to supply, which is essentially the opposite of scarcity prices.

The mechanisms in the extreme cases also apply to non-extreme cases. The presence of flexible assets in a market opens up the possibility to adequately respond to changes in supply and demand and that way prevent more severe price changes. In general, less flexibility therefore causes higher price volatility (Wirdemo, 2017).

This effect is illustrated in figure 12 below. Suppose all electricity is supplied by intermittent renewables, (inflexible) nuclear power plants and flexible assets. Now, a negative supply shock from intermittent renewables occurs, due to less wind. If this market would not have flexible assets in place, scarcity prices would have to be charged, indicated by the yellow vertical line. However, this market does have sufficient flexible assets in place and can easily ramp up production from flexible power plants or use its cross-border capacity to supply additional electricity.

6 _{In some cases (e.g. France, later also in Germany) some NPPs are dedicated to following the so-called load-following}

26

Figure 12: Merit-order effect, negative supply shock

Source: own production

In this case a negative supply shock is illustrated. Flexible assets can swiftly adjust quantities supplied, as indicated by the dotted line. If quantities cannot be adjusted, scarcity prices will emerge, as indicated by the yellow vertical line.

4.3 I

The existing literature agrees that the intermittent renewables have a positive effect on the ex-post forward premium. This is caused by the increased uncertainty about future spot prices and increased spot price volatility, arising from supply shocks, that are inherent with intermittent renewables, and translated into price changes through the merit-order effect, described above. This increased uncertainty about and volatility of future spot prices increases the risk of market

price intermittent renewables nuclear flexible assets demand quantity I N F price intermittent renewables nuclear flexible assets demand

negative supply shock

quantity

27

participants and drives an increased desire to hedge against adverse spot price movements. This risk is priced in into the forward price by means of the forward premium.

4.4 I

,

Ultimately, the forward premium is a function of the forward price and the spot price, either expected (ex-ante premium) or realized (ex-post premium). Having explained the effects of intermittent renewables and flexibility on electricity spot prices individually, the combined effect of the two on electricity spot prices will now be explained, followed by the combined effect on the forward premium.

29

5 M

ETHODOLOGY

This chapter will describe the methodology used in this research. Section 5.1 explains the difference between ex-ante and ex-post premiums and discusses the pros and cons of both premiums. The economic model will then be explained in section 5.2.

To assess how flexible assets affect the effect of intermittent renewables on the ex-ante premium, one needs to have the ex-ante premium. In order to do this, future spot prices need to be estimated. This is the first step, and this is done using input prices for fossil fuels and other determinants such as nuclear power plant downtime, hydro reservoir levels, wind speed and sun hours. Exactly how this is done is explained in more detail in section 5.3.

Having estimated the future spot price, the ex-ante premium can be calculated by subtracting the estimated future spot price from the forward price. This process is discussed in section 5.4. Finally, the forward premiums will be regressed on the variables nuclear, intermittent assets, flexible assets, the interaction term between intermittent and flexible assets, and the control variables skewness, kurtosis and variance of the spot price in the month preceding the delivery month and monthly dummies. The interaction term between intermittent and flexible assets is of particular importance as this will define the impact of the two types of assets combined.

5.1 E

-

-

Forward prices in theory consist of two parts: the expected spot price plus a risk premium (the ex-ante forward premium). Of course, the breakdown of the forward price is not explicitly given – market participants only see the forward price – so the ex-ante forward premium is unobservable and can only be estimated.

Forward premiums can be calculated in two distinct ways: ex-ante and ex-post. The ex-ante premium is equal to the difference between the forward price 𝐹_𝑡 and the expected spot price 𝐸[𝑆_𝑡] and is thus determined when the contract is closed. Unfortunately, the expected spot price is unobservable, and therefore the ex-ante premium is unobservable. Alternatively, researchers use the observable ex-post premium as a proxy. The ex-post premium is the difference between the forward price 𝐹𝑡 and the realized spot price 𝑆𝑡. In summary, the ex-ante premium is equal to

30

It is the ex-ante premium that most researchers investigating forward premiums are interested in, as the ex-ante premium truly reflects the hedger’s willingness to pay for insurance. We are interested in how the interaction between RES generation and flexible assets affects hedgers’ willingness to pay for insurance, measured by the ex-ante premium. Hedgers determine the price they are willing to pay for this insurance at the time of closing the contract based on information available at that point in time. The forward price for that contract does not change subsequently, so if the realized spot price differs from the expected spot price, this difference will be reflected in the ex-post premium. A measurement error then exists as the ex-post premium will not be a reliable proxy for the ex-ante premium that the hedger was willing to pay at the time of entering into the contract.

5.2 E

The economic model will now be elaborated upon. The spot price regression serves as a preliminary step to obtain estimated future spot prices. The prices of gas, coal and CO2 are inputs for electricity generation plants and are, together with NPP downtime, hydro reservoir levels, wind speed and sun hours, the supply-side variables. Any seasonal variation in gas and coal prices is already captured directly in the price. The only demand-side variables are the monthly dummy variables. These capture the seasonal variation in demand, which, in the countries considered in this research, is usually higher in the colder months, when more electricity is consumed for heating, lighting and roasting turkey for Christmas dinner. The coefficients for gas, coal and CO2 are all expected to be positive. This is because of the marginal cost pricing mechanism, an increase in the price of these inputs will also increase the price of the output; electricity. The coefficient for NPP downtime is expected to be positive, as more downtime means capacity constraints. Coefficients for hydro reservoir levels, wind speed and sun hours are expected to be negative as higher levels of these variables are expected to lead to more supply from hydro-power and intermittent renewables.

31

parties selling forward contracts know that total supply has substantial variability and therefore there is more variability in the spot price (due to the merit-order effect, see section 4.2). The parties selling forward contracts face a higher risk and therefore require a higher risk premium. NPP usually does not have this variability in demand, but it is not well suited to cope with demand-shocks. If, for example, demand unexpectedly increases a lot (i.e. a positive demand shock), NPPs will usually not increase output to meet the increased demand levels and scarcity prices will emerge. The same argument goes for a negative demand shock. Therefore, NPPs add to the risk of scarcity prices which translates into price peaks. This too increases the risk for the parties selling forward contracts and requires a higher risk premium.

The effect of flexible assets on the premium is expected to be negative, as flexibility reduces the risk and severity of price peaks. In the case of supply or demand shocks, flexible assets, by definition, should have the flexibility to adequately respond to these shocks and prevent or minimize price shocks. For example, in the case of a positive supply shock, this can be done by increasing output from fossil or hydro power plants or by importing more electricity from neighboring countries.

The interaction term of intermittent RES and flexible assets is expected to be negative. If this is the case, then the presence of flexible assets mitigates the (expected) positive effect of intermittent RES. Any supply shocks from intermittent RES are absorbed by flexible assets, as explained above.

5.3 E

To estimate the ex-ante premium, we will subtract an estimate of the expected spot price from the forward price. In order to do this, we thus need an estimate for the future spot price. Mulder et al. (2019) regress Dutch daily year-ahead baseload electricity prices against gas- and CO2 prices for the period 2010 – 2018 with an 𝑅2 of 0.91. Two explanatory variables thus seem to have

32

by the number of hours of downtime). For Germany, where intermittent renewables are abundant, the wind speed and number of sun hours are included in the regression. For France reservoir levels for hydro-power (together dominating the French energy market) is included too. For the Netherlands no variables are added.

For this regression the Seemingly Unrelated Regression (SUR) model will be used, using daily prices covering January 2013 to May 2019. The SUR model estimates a model for each country, while allowing error terms to be correlated across equations (i.e. countries). We apply robust standard errors to deal with heteroskedasticity. The model used for estimation is defined as: 𝑃_{𝑝𝑜𝑤𝑒𝑟_𝑠𝑝𝑜𝑡,𝑡,𝑖} = 𝛼_𝑖 + 𝛽_1,𝑖× 𝑃_{𝑔𝑎𝑠_ 𝑠𝑝𝑜𝑡,𝑡} + 𝛽_2,𝑖× 𝑃_{𝑐𝑜𝑎𝑙_𝑠𝑝𝑜𝑡,𝑡}+ 𝛽_3,𝑖× 𝑃_{𝐶𝑂2,𝑡} + 𝛽_4,𝑖×

𝑁𝑃𝑃𝐷𝑜𝑤𝑛_𝑡,𝑖+ 𝛽_5,𝑖× 𝐻𝑦𝑑𝑟𝑜𝑠𝑡𝑜𝑐𝑘_𝑡,𝑖 + 𝛽_6,𝑖× 𝑊𝑖𝑛𝑑_𝑡,𝑖 + 𝛽_7,𝑖× 𝑆𝑢𝑛_𝑡,𝑖+ 𝛽₈× 𝐷_𝐹𝑒𝑏… + 𝛽₁₈×

𝐷𝐷𝑒𝑐+ 𝜖𝑡,𝑖 [1]

The variables are explained in the table below.

Table 1: Spot price regression variables

𝑃𝑝𝑜𝑤𝑒𝑟_𝑠𝑝𝑜𝑡,𝑡,𝑖 Power spot price for country i at time t

𝑃𝑔𝑎𝑠_ 𝑠𝑝𝑜𝑡,𝑡 Gas spot price at time t

𝑃𝑐𝑜𝑎𝑙_𝑠𝑝𝑜𝑡,𝑡 Coal spot price at time t

𝑃𝐶𝑂2,𝑡 CO2 price at time t

𝑁𝑃𝑃𝐷𝑜𝑤𝑛𝑡,𝑖 Nuclear power plant downtime for country i at time t

𝐻𝑦𝑑𝑟𝑜𝑠𝑡𝑜𝑐𝑘_𝑡,𝑖 Hydro reservoir levels for country i at time t 𝑊𝑖𝑛𝑑_𝑡,𝑖 Wind speed for country i at time t

𝑆𝑢𝑛𝑡,𝑖 Sun hours for country i at time t

𝐷𝐹𝑒𝑏… 𝐷𝐷𝑒𝑐 Monthly dummy variables

33

monthly product. We therefore need to convert the data from daily to monthly. In this conversion, the last quoted price of each month is kept as this is assumed to have the lowest forecast error, since it has the nearest time to delivery. For spot prices, an average is calculated for each month.

The coefficients obtained in [1] are multiplied with the forward prices for gas and coal, the CO2 price and the applicable monthly dummies, to obtain an estimated month-ahead spot price per country, as shown in equation 2.

𝐸[𝑃_{𝑝𝑜𝑤𝑒𝑟_𝑠𝑝𝑜𝑡,𝑡,𝑖}] = 𝛼̅_𝑖+ 𝛽̅_1,𝑖 × 𝑃_{𝑔𝑎𝑠_𝑓𝑤𝑑,𝑡}+ 𝛽̅_2,𝑖× 𝑃_{𝑐𝑜𝑎𝑙_𝑓𝑤𝑑,𝑡}+ 𝛽̅_3,𝑖× 𝑃_{𝐶𝑂2,𝑡} + 𝛽̅_4,𝑖×

𝑁𝑃𝑃𝐷𝑜𝑤𝑛_𝑡,𝑖+ 𝛽̅_5,𝑖× 𝐻𝑦𝑑𝑟𝑜𝑠𝑡𝑜𝑐𝑘_𝑡,𝑖 + 𝛽̅_6,𝑖× 𝑊𝑖𝑛𝑑_𝑡,𝑖 + 𝛽̅_7,𝑖× 𝑆𝑢𝑛_𝑡,𝑖+ 𝛽̅₈× 𝐷_𝐹𝑒𝑏… + 𝛽̅₁₈×

𝐷_𝐷𝑒𝑐 [2]

The variables are explained in the table below.

Table 2: Spot price estimation variables

𝐸[𝑃_{𝑝𝑜𝑤𝑒𝑟_𝑠𝑝𝑜𝑡,𝑡,𝑖}] Expected power spot price for country i at time t

𝑃_{𝑔𝑎𝑠_ 𝑓𝑤𝑑,𝑡} Gas forward price for time t

𝑃_{𝑐𝑜𝑎𝑙_𝑓𝑤𝑑,𝑡} Coal spot price for time t

𝑃_{𝐶𝑂2,𝑡} CO2 price at time t

𝑁𝑃𝑃𝐷𝑜𝑤𝑛_𝑡,𝑖 Nuclear power plant downtime for country i at time t 𝐻𝑦𝑑𝑟𝑜𝑠𝑡𝑜𝑐𝑘𝑡,𝑖 Hydro reservoir levels for country i at time t

𝑊𝑖𝑛𝑑_𝑡,𝑖 Wind speed for country i at time t 𝑆𝑢𝑛_𝑡,𝑖 Sun hours for country i at time t 𝐷_𝐹𝑒𝑏… 𝐷_𝐷𝑒𝑐 Monthly dummy variables

34

to be predictable, although in reality there will be some measurement error involved in estimating the values for these variables.

5.4 C

Once the month-ahead spot prices have been estimated, calculating the forward premia is a trivial exercise. For the ex-ante premium, the estimated spot prices are simply subtracted from the forward price, as described in section 5.1. The ex-post premium is calculating by subtracting the actual spot price from the forward price. The result is base load and peak load premia per country, which for comparison and completeness are calculated both ex-ante and ex-post.

5.5 E

At this point, the data is converted into panel data. The purpose of this research is to assess the role of flexibility in the effect of intermittent renewables on the forward premium in general, hence there is no need to analyze this on a country level. Furthermore, the use of panel data accounts for individual heterogeneity, and it gives a more accurate inference.

After converting the data into panel data, a variable is created that represents the total capacity (in MW) of flexible assets, consisting of hydropower and fossil fuels generation capacity, and cross-border capacity. NPP capacity is included separately in the regression as it belongs to neither intermittent nor flexible assets. Solar and wind powered generation capacity is aggregated to form intermittent renewable assets.

35

Table 3: Variables electricity forward premium regression

𝑝𝑟𝑒𝑚𝑖𝑢𝑚_𝑡 Electricity forward premium at time t (delivery month) 𝑁𝑡 Nuclear power plant capacity in MW at time t

𝐹_𝑡 Flexible assets (fossil, hydro, import, export) capacity in MW at time t

𝐼_𝑡 Intermittent renewables (wind, solar) capacity in MW at time t 𝐹_𝑡× 𝐼_𝑡 Interaction term between flexible assets capacity and

intermittent renewables capacity in MW at time t

𝑠𝑘𝑒𝑤𝑃𝑠𝑝𝑜𝑡,𝑡−1 Skewness of the electricity spot price in the month preceding the delivery month

𝑘𝑢𝑟𝑡_{𝑃𝑠𝑝𝑜𝑡,𝑡−1} Skewness of the electricity spot price in the month preceding the delivery month

𝑣𝑎𝑟_{𝑃𝑠𝑝𝑜𝑡,𝑡−1} Skewness of the electricity spot price in the month preceding the delivery month

36 The following three hypotheses will be tested.

Intermittent RES generation capacity has no significant effect on electricity forward premiums. This hypothesis will be tested by the coefficient from intermittent renewable generation capacity in the regression on forward premiums. For the null hypothesis to be rejected, this coefficient needs to be positive and significant.

The presence of flexible assets has no significant mitigating effect on the effect of intermittent RES generation capacity on electricity forward premiums.

This hypothesis will be tested by the interaction term between flexible assets and intermittent renewable generation capacity. This interaction term measures the effect of the two types of assets combined, while controlling for the individual effects and the effect of nuclear power plants (NPPs). If the coefficient of the interaction term negative, the null hypothesis will be rejected in favor of the alternative hypothesis.

The effect of intermittent RES generation capacity on electricity forward premiums is not significantly greater for peak load than for base load premiums.

37

6 D

ATA

In this section the data is described. This includes financial data in section 6.1, generation capacities in section 6.2 and cross-border capacities in section 6.3. Section 6.4 gives a brief description of the data used to enhance the spot price estimation model, and sections 6.5, 6.6 and 6.7 elaborate on test diagnostics.

6.1 F

All futures and forwards are referred to as forwards, to prevent confusion with future spot prices. As Redl and Bunn (2012) argue, month-ahead forward contracts are the most liquid contract and most price data is available for forwards with monthly delivery periods. Furthermore, due to the near-term delivery period, forecast errors should on average be low. They use one-month’s forward prices on the last trading day before the delivery month. In line with Redl and Bunn (2012) and Paschen (2018), this research will use ahead forwards as well. All forwards are month-ahead forwards for delivery during the next calendar month. All prices are drawn from Bloomberg. The coal price is the API2 Rotterdam coal forward. In the absence of coal spot prices, a (30-day) lagged coal forward price is used as a proxy. This variable will, however, be referred to as the coal spot price. Figures 13 and 14 below visualize the spot prices, base load and peak load, over the full period of January 2010 – May 2019. Figures 15 and 16 show the base load and peak load forward prices. All prices are relatively well available, with daily prices, except for Dutch electricity prices which are only available after 2013 and Belgian peak forwards which are not available at all. Hence, the dataset contains data from January 1st_{, 2013 up to and including May}

1st_{, 2019 and does not consider Belgian peak prices. Figure 17 shows gas spot and month-ahead}

forward prices, while figure 18 shows gas and coal month-ahead prices and CO2 emission allowance prices, the latter measured in euros per ton on the secondary axis. Summary statistics of the electricity, gas, coal and CO2 prices are given in table 4.

38

Figure 13: Daily electricity base load spot (day-ahead) prices (EUR/MWh), Jan 2010 – May 2019

Source: Bloomberg

Note: y-axes differ in range and scale

39

Figure 14: Daily electricity peak load spot (day-ahead) prices (EUR/MWh), Jan 2010 – May 2019

Source: Bloomberg

Note: y-axes differ in range and scale

40

Figure 15: Daily electricity base load month-ahead forward prices (EUR/MWh), Jan 2010 – May 2019

Source: Bloomberg

Note: y-axes differ in range and scale

41

Figure 16: Daily electricity peak load month-ahead forward prices (EUR/MWh), Jan 2010 – May 2019

Source: Bloomberg

Note: y-axes differ in range and scale

42

43

Figure 18: Daily gas (PEGAS NGC) and coal (API2) month-ahead forward prices (EUR/MWh, primary axis) and CO2 price7 _{(EUR/ton, secondary axis), Jan 2010 – May 2019}

Source: Bloomberg

7 _{CO2 price is the EU ETS price for emission of one ton of CO2, valid until the next January 1}st _00:00.

0 1 0 2 0 3 0 E U R /t o n 5 1 0 1 5 2 0 2 5 3 0 E U R /M W h 1/1/2010 1/1/2012 1/1/2014 1/1/2016 1/1/2018 1/1/2020

44

Table 4: Summary statistics financial data (EUR/MWh (CO2 in EUR/ton8₎₎

Belgium Germany France Netherlands

VARIABLES N mean sd min max mean sd min max mean sd min max mean sd min max Daily CO28 76 8.398 5.540 3.220 25.96 8.398 5.540 3.220 25.96 8.398 5.540 3.220 25.96 8.398 5.540 3.220 25.96 Daily coal month-ahead forward 76 7.416 1.460 4.699 10.34 7.416 1.460 4.699 10.34 7.416 1.460 4.699 10.34 7.416 1.460 4.699 10.34 Daily gas month-ahead forward 76 20.42 4.544 12.24 28.94 20.42 4.544 12.24 28.94 20.42 4.544 12.24 28.94 20.42 4.544 12.24 28.94 Daily gas spot 76 14.63 10.43 0 50 14.63 10.43 0 50 14.63 10.43 0 50 14.63 10.43 0 50 Daily power spot base 76 43.17 16.17 4.670 116.6 31.16 15.51 -25.30 61.24 36.95 15.02 6.740 80.79 38.79 9.228 20.89 60.30 Daily power spot peak 76 46.62 19.96 8.370 152.5 33.96 18.03 -31.73 75.87 41.89 19.31 5 125.2 45.61 12.45 18.49 82.77 Daily power month-ahead forward

base

76 48.05 18.28 24.10 149.5 35.59 7.911 21.99 59.33 42.14 14.11 22.37 98.33 43.50 9.419 25 72.50

Daily power month-ahead forward peak

76 44.12 10.11 25.90 70.42 52.60 19.20 28.30 135.3 51.12 11.44 29.13 86.70

Source: Bloomberg

45

6.2 C

-

Capacity data for interconnections is available as monthly data and is drawn from the four country’s transmission system operators (TSO)9_{. The capacity used for this research is the net}

transmission capacity (NTC). The NTC is the total transmission capacity minus the transmission reliability margin. These NTC values set the physical limit for cross-border power exchange. There exists a strong seasonal pattern in NTC which is mainly caused by patterns of generation and load that differ seasonally, and the thermal ratings of the transmission lines which vary across seasons, depending on the outside temperature. Usually the thermal ratings are higher in the wintertime, allowing for higher cross-border capacities10_{. Figure 19 below shows how the cross-border}

capacities vary over time and across countries.

Figure 19: Net Transmission Capacities (in thousand MW), 2013 – 2018

Sources: Elia, 50Hertz, Amprion, TenneT, TransnetBW, RTE

9 _{Elia (Belgium); 50Hertz, Amprion, TenneT, TransnetBW (Germany); RTE (France); TenneT (Netherlands)}

46

The summary statistics below in table 5 and figure 19 above show that France has the highest mean, maximum and standard deviation, followed by Germany. The Netherlands and Belgium follow with considerable distance. It is, however, important to keep in mind that consumption and generation in France and Germany is much higher than in the Netherlands and Belgium.

6.3 G

Data for installed net generation capacity (NGC) is available on an annual basis and is drawn from the European Network of Transmission System Operators for Electricity (Entso-e)11_{. Entso-e}

collects this data from the respective TSO’s. The 2013 generation mix is shown in figure 20 below.

Figure 20: Net generation capacity CWE countries, 2013

Source: Entso-e

47

Appendix I contains line charts that show per country how the NGC changes over time.

The graph above and the summary statistics on the next page show how the Belgium market is dominated by fossil and NPPs, the German market by intermittent RES and fossil power plants, the French market by nuclear and to a lesser extent hydro and fossil, and the Dutch market by fossil fueled power plants.

48

Table 5: Summary statistics cross-border capacity and generation capacity (in MW), 2013

Belgium Germany France Netherlands

VARIABLES mean sd min max mean sd min max mean sd min max mean sd min max Import 3.469 1.015 0 4.800 11.64 2.828 0 13.43 12.39 3.484 0 18.80 4.870 1.188 0 5.780 Export 2.052 0.622 0 3.200 9.725 2.552 0 14.09 13.80 3.706 0 18.93 3.196 0.830 0 4.020 Nuclear 5.923 0.00340 5.919 5.926 10.73 1.060 9.516 12.07 63.13 0 63.13 63.13 0.510 0 0.510 0.510 Fossil 7.246 0.312 6.639 7.599 83.09 2.157 78.75 85.80 21.71 2.576 18.59 25.71 27.18 1.995 23.83 29.72 Hydro 1.429 0.00323 1.424 1.433 10.31 0.379 9.610 10.78 25.34 0.167 25.09 25.51 0.0370 0 0.0370 0.0370 Wind 2.409 0.546 1.720 3.247 46.87 9.287 34.04 58.23 11.55 2.503 8.157 15.11 3.656 0.667 2.707 4.292 Solar 3.137 0.307 2.680 3.581 40.10 2.559 36.91 43.92 6.576 1.437 4.373 8.527 2.204 1.289 0.739 4.300 Other 1.082 0.196 0.806 1.340 7.785 1.400 5.856 9.397 1.746 0.289 1.190 2.026 0.0406 0.00858 0.0330 0.0530

49

6.4 N

,

,

Data for planned downtime of Belgium generation plants is available on Belgian TSO Elia’s website12_{. Data for French reservoir levels and planned downtime can be found on the data portal}

of the French TSO, RTE13_{. German weather data is drawn from the Deutscher Wetterdienst}14_.

6.5 D

Diagnostic tests have been performed on the SUR regression, on country level.

The White test and the Cook-Weisberg test reveal that the residuals are heteroskedastic. This is dealt with by using robust standard errors.

The Durbin-Watson test as well as the Breusch-Godfrey LM-test reveal that there is also evidence for autocorrelation in the standard errors. Unfortunately, STATA doesn’t offer the possibility to use autocorrelation consistent standard errors in combination with the SUR model. Therefore, we have to settle with including time-fixed effects in the form of monthly dummy variables. The Jarque-Bera test and visualization of the distribution of the error terms (the latter given in appendix II, the former in appendices IV through XI) reveal that the error terms are not normally distributed. However, considering the large amount of observations this is accepted.

VIF-values to detect multicollinearity are not concerning. All test results can be found in appendices IV through XI.

6.6 D

The same tests as above reveal that the premium regression model suffers from autocorrelation as well as heteroskedasticity in the residuals. Therefore, a cross-sectional time-series feasible generalized least squares (FGLS) regression is adopted. This adjusts standard errors to allow for

12 _{Visit http://www.elia.be/en/grid-data/data-download for more information}

50

the presence of autocorrelation within panels and cross-sectional correlation and heteroskedasticity across panels. The standard errors are thus heteroskedasticity and autocorrelation consistent. FGLS offers no possibility to include random or fixed effects.

Again, there is no evidence for multicollinearity and given the large sample the non-normality of the error terms is accepted.

6.7 N

The Augmented Dickey-Fuller test reveals that the CO2 price, the (lagged) coal forward price and French reservoir levels are nonstationary. The first-differences of these variables are stationary. Therefore, these variables will be replaced in the regression by their first-differences. This also means that prediction of the expected future spot price will be done using first-differences of the forward prices of these variables. Coefficients based on first-differences start with an increment-sign (∆) followed by the coefficient’s label.

The regressions of the premiums is done on data regarding capacities. The data-series for flexible assets capacity is stationary. The data-series for intermittent renewables is detrended to remove nonstationarity. The data-series for nuclear capacity is nonstationary, even when including a drift or a trend. Also, first-differences offer nu solution. Therefore, nuclear capacity is dropped from the regression. The results of the Augmented Dickey-Fuller tests can be found in appendix III. Model (1) thus becomes

𝑃𝑝𝑜𝑤𝑒𝑟_𝑠𝑝𝑜𝑡,𝑡,𝑖 = 𝛼𝑖 + 𝛽1,𝑖× 𝑃𝑔𝑎𝑠_ 𝑠𝑝𝑜𝑡,𝑡 + 𝛽2,𝑖× ∆𝑃𝑐𝑜𝑎𝑙_𝑠𝑝𝑜𝑡,𝑡+ 𝛽3,𝑖× ∆𝑃𝐶𝑂2,𝑡 + 𝛽4,𝑖×

𝑁𝑃𝑃𝐷𝑜𝑤𝑛_𝑡,𝑖+ 𝛽_5,𝑖× ∆𝐻𝑦𝑑𝑟𝑜𝑠𝑡𝑜𝑐𝑘_𝑡,𝑖+ 𝛽_6,𝑖× 𝑊𝑖𝑛𝑑_𝑡,𝑖+ 𝛽_7,𝑖× 𝑆𝑢𝑛_𝑡,𝑖+ 𝛽₈× 𝐷_𝐽𝑎𝑛… + 𝛽₁₉ ×

𝐷_𝐷𝑒𝑐+ 𝜖_𝑡,𝑖 [1]

And model (3) becomes

51

7 R

ESULTS

In this section the results will be discussed, starting with the spot price regression (section 7.1), followed by estimation of the future spot price (section 7.2), analysis of obtained ex-ante versus ex-post premiums (section 7.3) and ending with the forward premium regressions (section 7.4) and a robustness check (section 7.5).

7.1 S

The regression output of the Seemingly Unrelated Regression (SUR) results of the spot price base load and peak load regressions are given in tables 6 and 7 below. The coefficient for the gas price has the expected (positive) sign in all countries, the coefficient for first-differences of the CO2 price is only significant in Germany, the other coefficients for coal and CO2 are insignificant. The coefficient for wind has the expected (negative) sign: the more wind, the more electricity will be generated by wind turbines and the lower the electricity spot price will be due to the merit-order effect. The coefficient for sun hours is insignificant.

Downtime in Belgium has an unexpected significant negative coefficient. In theory, the more downtime, the less available generation capacity and the higher the price as electricity suppliers with higher marginal costs are needed to supply electricity. In France the coefficient for downtime is positive and significant.

Hydro reservoir levels are expected to be negatively related with the electricity price, as the higher the reservoir levels are, the higher the expected generation is from hydro power, but the coefficient is not significant.

52

Table 6: Seemingly Unrelated Regression: Base load electricity spot prices

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

VARIABLES Belgium Germany France Netherlands

Gas spot 1.277*** 0.688*** 1.053*** 1.382*** (0.0600) (0.0623) (0.0653) (0.0580) ∆.Coal spot -3.632 -3.328 -1.807 -1.822 (4.283) (2.935) (3.390) (1.346) ∆.CO2 0.920 3.299** 0.172 0.538 (1.246) (1.343) (0.877) (0.733) Downtime BE -0.627*** (0.129) Wind speed DE -0.748*** (0.0672) Sun hours DE -0.0659 (0.0739) Downtime FR 1.397*** (0.124)

∆.Hydro stock FR -7.30e-07

(1.55e-06) Constant 29.37*** 36.50*** 3.878 9.307*** (2.463) (2.233) (2.572) (1.376) Feb -4.357*** -0.819 -1.256 0.646 (1.593) (1.568) (1.714) (0.814) Mar -6.489*** -5.372*** -3.652** 1.729** (1.581) (1.577) (1.704) (0.870) Apr -5.622*** -7.186*** -6.066*** 1.815** (1.568) (1.489) (1.771) (0.774) May -7.114*** -9.173*** -11.71*** 1.235 (1.669) (1.496) (1.722) (0.833) Jun -8.830*** -7.668*** -11.02*** 0.836 (1.762) (1.472) (1.777) (0.865) Jul -7.868*** -3.551** -7.883*** 1.787** (1.712) (1.540) (1.723) (0.868) Aug -6.735*** -5.084*** -7.207*** 2.811*** (1.767) (1.541) (1.780) (0.887) Sep -1.909 -3.184** -4.669*** 3.888*** (1.664) (1.411) (1.721) (0.841) Oct 6.506** -2.285 3.159* 2.849*** (2.581) (1.564) (1.831) (0.797) Nov 9.786*** 3.391** 4.882** 2.781*** (2.647) (1.629) (1.976) (0.804) Dec -0.983 -0.726 0.442 0.214 (1.898) (1.681) (2.134) (0.898) Observations 1,321 1,321 1,321 1,321

BE = Belgium, DE = Germany, FR = France ∆ = first-differences

53

Table 7: Seemingly Unrelated Regression: Peak load electricity spot prices

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

VARIABLES Belgium Germany France Netherlands

Gas spot 1.407*** 0.828*** 1.321*** 1.731*** (0.0733) (0.0827) (0.0714) (0.0562) ∆.Coal spot -5.318 -3.714 0.291 -3.440 (5.587) (3.874) (4.470) (2.493) ∆.CO2 1.684 3.878** 1.218 0.697 (1.624) (1.549) (1.242) (0.826) Downtime BE -0.743*** (0.163) Wind speed DE -0.770*** (0.0823) Sun hours DE -0.154* (0.0906) Downtime FR 1.208*** (0.154)

∆.Hydro stock FR -2.67e-06*

(1.39e-06) Constant 35.43*** 41.95*** 13.89*** 16.56*** (2.996) (2.980) (3.048) (1.601) Feb -7.382*** -3.450* -6.084*** -4.654*** (1.835) (1.996) (1.931) (1.173) Mar -10.80*** -10.14*** -11.75*** -6.774*** (1.839) (1.968) (1.950) (1.204) Apr -11.22*** -13.59*** -16.14*** -6.840*** (1.858) (1.897) (1.984) (1.148) May -11.89*** -14.90*** -20.50*** -6.886*** (1.990) (1.914) (2.063) (1.286) Jun -12.35*** -13.12*** -18.21*** -4.793*** (2.039) (1.862) (2.055) (1.278) Jul -11.77*** -8.847*** -14.42*** -4.701*** (1.973) (1.929) (2.022) (1.238) Aug -11.06*** -10.11*** -14.41*** -5.677*** (2.044) (1.962) (2.074) (1.219) Sep -5.087*** -6.852*** -7.462*** -3.625*** (1.929) (1.835) (1.952) (1.184) Oct 6.801* -4.886** 0.845 -2.596** (3.569) (1.941) (2.115) (1.184) Nov 12.32*** 3.524* 8.245*** 2.315 (3.543) (2.095) (2.777) (1.472) Dec -1.429 -1.143 -2.075 -0.714 (2.228) (2.051) (2.392) (1.419) Observations 1,321 1,321 1,321 1,321

BE = Belgium, DE = Germany, FR = France ∆ = first-differences