Understanding the determinants of electricity prices

and the impact of the German Nuclear Moratorium

in 2011
Stefan Thoenes∗†

Abstract
This paper shows how the effect of fuel prices varies with the level of elec-
tricity demand. It analyzes the relationship between daily prices of electric-
ity, natural gas and carbon emission allowances with a semiparametric varying
smooth coefficient cointegration model. Different electricity generation technolo-
gies have distinct fuel price dependencies, which allows estimating the structure
of the power plant portfolio by exploiting market prices. The semiparametric
model indicates a technology switch from coal to gas at roughly 85% of maxi-
mum demand. This model is used to analyze the market impact of the nuclear
moratorium by the German Government in March 2011. Futures prices of elec-
tricity, natural gas and emission allowances are used to show that the market
efficiently accounts for the suspended capacity and correctly expects that several
nuclear plants will not be switched on after the moratorium.

Keywords: Electricity market, merit order, cointegration, varying coefficient, nuclear

moratorium
JEL-Classification: G14, L94, Q41, Q48

∗
Institute of Energy Economics at the University of Cologne (EWI), Vogelsanger Str. 321, 50827
Cologne, Germany. Email address: [email protected]
†
I am grateful to Felix Höffler, Christian Growitsch, Marc Oliver Bettzüge, Jörg Breitung, Jeffrey
S. Racine, Caroline Dieckhöner, Jürgen Kruse and Johannes Viehmann for their helpful comments
and suggestions. I would also like to thank the participants at the Energy & Finance Conference
2011 in Rotterdam and at the International Ruhr Energy Conference 2012 in Essen.

1
1 Introduction

Electricity is a homogeneous good that cannot be stored at reasonable economic costs.

However, the demand is highly seasonal and needs to be satisfied at all times. Hence,

it is most efficient to generate electricity with a mixture of various technologies with

different properties regarding capital costs and marginal costs. These technologies also

differ in terms of input fuels and carbon emissions.

Therefore, how input price variations affect the electricity price critically depends

on the marginal technology used; and the marginal technology used depends on the
level of the residual demand.1 The present paper tries to investigate exactly this
effect. To illustrate the point, consider the ‘merit order’, i.e., an ordering of fossil

power plants from those with low marginal cost (like lignite or hard coal) to high
marginal cost (natural gas). If the residual demand is low (e.g. because electricity

demand is low in the night; or because there is a lot of wind feed-in), the marginal
power plant will be a coal fired power plant, and we expect that changes in the gas

price will not affect the electricity price. This will be the case only if demand is high.
The approach in the present paper allows to identify how the fuel price effects vary
with the size of the residual demand.

This is analyzed empirically using data from the German electricity market and ap-
plying a semiparametric cointegration model. In order to measure how the fuel price

sensitivity changes throughout the merit order, it is necessary to use a model that
allows the parameters of the fuel price sensitivity to vary freely. The semiparamet-

ric varying smooth coefficient model, which was introduced by Hastie and Tibshirani
(1993), allows for straightforward analysis of the relationship between fuel price sen-
sitivity and load. The main advantage of the model is that the nature of the varying
1
The residual demand is the electricity demand minus the feed-in of renewables, like wind or solar
power.

2
effect is directly derived from the data, which means that there is no need for ad-hoc

assumptions or restrictive functional specifications. Recent work by Cai et al. (2009)

and Xiao (2009) shows that such a model can be used to estimate the nonlinear func-
tional coefficients of a cointegration relationship. The application of this estimator

is novel for modeling the dynamics of electricity markets. This method indicates a
technology switch from coal to gas fueled power plants at around 60 gigawatt (GW)

average non-wind daily peak generation. The estimated input price sensitivities are
used to simulate the merit order for different natural gas and carbon price scenarios.2

The usefulness of this approach can be illustrated by analyzing a specific policy

intervention like the German nuclear power suspension in March 2011. After the
incident in Japan’s Fukushima nuclear power plant, the German government decided

to put the so called ‘Nuclear Moratorium’ in place. Seven nuclear power plants, all built
before 1980, had to be switched off from 03/15/2011 to 06/15/2011 to examine the

security of these plants. After the announcement, the market reacted with immediate
price increases of electricity, gas and carbon emission allowance futures. Using only

these futures prices, the proposed model is able to split the electricity price increase
into a fuel price component and a capacity effect. It is also possible to measure the
expectations of the market for the period after the end of the moratorium. The results

of the event study show that the market accounts for most of the capacity effect during
the period of the moratorium and expects that several nuclear power plants remain

closed. This expectation proved to be correct as all affected nuclear power plants were
permanently decommissioned after the end of the moratorium.

The approach in this paper relates to two distinct strands of the literature on
empirical modeling of energy prices. The first strand focuses solely on the electricity
2
Carbon prices refer to EU emission allowance certificates under the European Emission Trading
Scheme phase II.

3
market and tries to resemble the stochastic characteristics of the typical price patterns.

Driven by capacity constraints, hourly and daily prices have a high volatility and
spikes. There are also hourly, daily and monthly seasonalities that reflect demand
patterns of consumers and industry. The two most prominent approaches are the

‘Mean Reverting Jump Diffusion Model’ and the ‘Markov Regime Switch Model’,
which are both described by Weron et al. (2004). These models can also be extended

by additionally accounting for fundamental factors like load (see Mount et al. (2006),
Kanamura and Ohashi (2007)). However, this class of models has the drawback that

the relationship between the electricity price and input fuel prices is not analyzed.
The second strand of literature consists of studies that broadly analyze the in-
terdependencies between different energy commodities, but fail to account for the

aforementioned specific fundamentals of the electricity market. Mohammadi (2009)

uses a vector error correction model (VECM) to analyze the long-term relationship

between fuel prices and electricity prices in the United States. Mjelde and Bessler
(2009) indicate that fossil fuels are weakly exogenous and electricity prices adapt to

re-establish the equilibrium. Similar results hold for the European electricity markets.
Bosco et al. (2010) employ a set of robust tests to show that European electricity time
series have a unit root and are cointegrated. Electricity prices seem to share a common

trend with gas prices, but not with oil prices. Ferkingstad et al. (2011) also find that
gas prices have strong instantaneous and lagged causal effects on electricity prices,

while coal and oil prices are less important. Furthermore, coal, oil and gas prices are
weakly exogenous. Fell (2010) finds evidence that the effect of fuel prices varies with

the level of demand. The author estimates a VECM for the Scandinavian electricity
spot market and several inputs. The short-term impact of the carbon price on the
electricity price is higher in off-peak hours than in peak hours. Coulon and Howison

(2009) account for this effect by directly modeling different parts of the supply stack.

4
The actual bids are split into clusters, which are governed by different fuels.

The present paper advances the current literature by showing how exactly the
natural gas and carbon price sensitivities vary with load. It fills the gap between
models that focus on idiosyncratic effects of the electricity market and models that

focus broadly on interdependencies between energy markets. The remainder of this

paper is organized as follows. Section 2 describes the data sets that are used for

the analysis. Section 3 outlines the semiparametric varying coefficient cointegration

model and discusses the empirical results. This part includes the semiparametric

estimates of the gas and carbon price sensitivity functions as well as the predicted
merit order simulation for different input price scenarios. In Section 4, the proposed
semiparametric model is used to analyze the market impact of the German nuclear

moratorium in March 2011. The conclusion is given in the final section.

2 Data

This study focuses on electricity, natural gas and carbon prices in Germany. The data
consists of daily observations from 2008/04/01 to 2010/09/29. All price time series

were obtained from the European Energy Exchange (EEX). This analysis uses day-
ahead base, peak and off-peak electricity prices on weekdays. The peak block covers the

hours from 8 am to 8 pm, while the off-peak block covers the remaining time. The base
block is the daily average price. Daily day-ahead EEX gas prices are quoted from July
2007 onwards. Both Gaspool and NetConnect Germany (NCG) contracts are traded,

but I choose NCG because of the higher liquidity in this market. NCG gas prices are
denominated in Euro/MWh and will be used as an indicator for the gas market as a

whole. For carbon prices, the EEX Carbix index of the EU Emission Trading Scheme

5
phase II is used.3 All prices are transformed into their natural logarithms.

The choice of price time series used for the analysis is driven by the consistency
of both the geography of the German market and the exchange itself. However, the
liquidity of the natural gas and carbon market at the EEX remains an issue for this

study, which could potentially affect the results. While the Belgian natural gas hub
Zeebrugge or the Dutch Title Transfer Facility (TTF) are sometimes considered more

important markets with a higher liquidity, it is a priori not clear whether they are bet-
ter proxies for the German natural gas price. Furthermore, due to a high convergence

of European natural gas prices, as shown by Renou-Maissant (2012) or Asche et al.

(2013), the choice of the natural gas price time series does not have a relevant impact
on the results.

Lignite, coal and oil prices are not included for several reasons. First, the oil fueled
electricity generation capacity in Germany is rather small, as it is shown in Table

1. Moreover, the trading and transportation properties of the coal market do not
match the daily frequency setup of this study. Lignite is not actively traded and is

usually not the marginal technology, which also holds for nuclear power. Adjustments
for electricity ex- and imports as well as reservoir power stations can be neglected,
because the observed relationship between load, input prices and electricity prices

implicitly accounts for their influence. Several comparable studies, including Fezzi and
Bunn (2009) and Zachmann and von Hirschhausen (2008), choose a similar approach

and focus on the cointegration relationship between electricity, gas and EU emission
allowance prices. The analysis of detailed cross-commodity relationships for a system

of all different energy commodities is not the aim of this study, but can be found in
Ferkingstad et al. (2011) and Mjelde and Bessler (2009).
3
The gas prices are taken from the trading day that is closest to delivery to match the trading
structure of the electricity market. Carbon spot prices are taken from the same trading day as the
gas prices. The delivery day of gas and electricity contracts is the same.

6
 Despite the fact that coal markets are biased towards over-the-counter trading and

long-term contracts, there are some proxies for spot coal prices in northwest Europe,
such as the McCloskey Coal Marker for the Amsterdam-Rotterdam-Antwerp (ARA)
region. However, in order to achieve a coherent framework for the event study, it

would be necessary to have a full set of actively traded futures for monthly, quarterly
and yearly contracts at the EEX. During the period of the announcement of the

moratorium, the spot coal price showed a strong co-movement with both natural gas
and carbon emission allowance futures prices, which all increased substantially less

than electricity prices. This observation indicates that the coal price did not reflect
additional information, and therefore, including coal prices into the analysis of the
moratorium would not change the main conclusions. However, as thoroughly argued

by Dyckman et al. (1984) and Armitage (1995), the market model and data are very
important for the event study methodology due to the fact that the choice of the

respective approach may have an impact on the final results.

Germany’s diversified technology and fuel mix is shown in Table 1. Electricity

from renewable energy sources enjoys a preferred feed-in policy. The remaining load
is covered by other technologies and cross-border exchange. Nuclear and lignite fueled
plants satisfy the base load, while coal and especially gas fueled power plants cover

the peak demand during the day. Generators have to buy EU emission allowances for
their carbon emissions.

ENTSO-E provides hourly load data for Germany.4 Wind forecasts and realized
wind production were obtained by aggregating publically available data from the major

transmission system operators (TSO), Amprion, 50Hertz and Transpower.5 Wind

4
ENTSO-E is the abbreviation for the European network of transmission system operators for
electricity. Data is publically available from www.entsoe.eu.
5
The data can be downloaded from: www.amprion.net/en/wind-feed-in, www.50hertz.com/
en/1983.htm and www.transpower.de/site/en/Transparency.

7
 Table 1: The German generation portfolio by technology

Technology Installed Capacity (in MW)

Wind 25,848
Nuclear 20,441
Lignite 20,375
Coal 16,158
Gas 13,094
Solar 10,392
Oil 1,826
Hydro 1,678
Waste 496
Total 110,307
Source: German Federal Ministry of Economics and Technology, 2010.

power production from EnBW has been neglected because of the unavailability of
forecasts and the small capacity.6 Daily wind in-feed and load data was derived by

averaging the quarter-hourly and hourly data. Day-ahead load forecasts are necessary
to model day-ahead electricity prices. I assume that the realized load is the best proxy

for this variable, because there is no publically available and generally accepted load
forecast. The realized load is adjusted by the official wind production forecasts of the
major TSOs. This adjusted load is called residual load. Summary statistics of the

price and load variables are given in Table 2.

For the event study of the impact of the nuclear moratorium, a range of different

EEX future contracts are used for a period from 2012/02/28 to 2012/04/18. The
analysis includes monthly electricity futures settlement prices with delivery in April,

May and June 2011, quarterly futures with delivery in the second, third and fourth
quarter of 2011 and yearly futures for 2012 and 2013. The analyzed electricity and gas
prices are futures with the same delivery period. The carbon price is the EU emission
6
EnBW accounted for 1.86% of the total German wind power production in August 2010. Data
was obtained from www.enbw-transportnetze.com.

8
allowance future for delivery in mid-December of the corresponding year.

Table 2: Summary Statistics

Variable Unit Mean Median Minimum Maximum Std. Dev.

Base Electricity e/MWh 53.71 47.68 17.06 131.40 18.71
Peak Electricity e/MWh 65.03 55.92 33.15 177.49 24.45
Off-peak Electricity e/MWh 41.67 38.73 -11.25 87.08 13.61
EU Emission Allowance e/t CO2 16.30 14.53 8.02 28.75 4.74
NCG Gas e/MWh 17.91 17.14 6.90 32.04 6.82
Base Residual Load MW 53,449 53,446 37,773 63,978 4,332
Peak Residual Load MW 59,303 59,394 41,445 69,255 4,619
Off-peak Residual Load MW 46,973 46,951 33,566 60,699 4,309

3 Semiparametric varying coefficient model

This section analyzes the relationship between natural gas, carbon emission allowances

and electricity prices. Given the fact that electricity is generated with different tech-
nologies, the relationship between fuel prices and the electricity price should depend

on the marginal technology used. It is necessary to assume that fuel price changes
are passed through to electricity markets. In this case, the carbon sensitivity for coal
driven parts should be higher than for gas. The dependence on gas prices should be

higher for periods with high load.

Thus, I use a semiparametric varying-coefficient model, which was introduced by

Hastie and Tibshirani (1993) as a generalized class of regression models. It measures

explicitly how the fuel price sensitivity varies with load, which means that the model

directly accounts for the underlying merit order. It is very flexible, because it does
not assume any functional specification of how the fuel price sensitivity varies, but
estimates it directly from the data. The model is given as

Yi = β(Zi )′ Xi + ui (1)

9
which seems to be rather specific. However, the model is very flexible, because Z is

a vector of so-called effect modifiers. The beta coefficients vary freely as a smooth
function depending on the effect modifier. This function does not need any further
specification and is estimated only from the data. The model proposed by Hastie and

Tibshirani (1993) is a static approach that is not necessarily capable of estimating

parameters in a time series context.

The literature on energy markets suggests that fuel prices and electricity prices
are cointegrated. There is a unilateral effect from fuel prices to electricity prices in

all markets. These results are robust for different regions and model setups.7 Thus,
the existence of a cointegration relationship is relevant for the following analysis and
also has to be examined in this article. The Johansen test indicates that there is

exactly one cointegration relationship and that both gas and carbon prices are weakly
exogenous. These prices do not adapt to the long-term equilibrium, indicating that the

electricity price follows the natural gas and carbon prices in a unilateral relationship.
Thus, it is possible to estimate this relationship in a single equation model with the

electricity price as endogenous variable. As the Johansen cointegration analysis and

the obtained findings are standard in the literature, the results of this preliminary step
are presented in Appendix A.

Recent studies by Cai et al. (2009) and Xiao (2009) expand the semiparametric
approach and analyze the properties of similar varying coefficient models for non-
7
Mohammadi (2009) finds that there is one cointegration vector in his model for annual electricity,
gas and coal prices in the United States. The error correction term is only significant for electricity.
Mjelde and Bessler (2009) use weekly data and find that only electricity and uranium prices adapt to
re-establish the equilibrium in the long-run relationship. Using a different methodology, Ferkingstad
et al. (2011) find a strong causal link from gas prices to electricity prices, while the German electricity
market does not have a causal effect on any fuel market. Fezzi and Bunn (2009) analyze daily spot
prices and show that gas and carbon prices drive the electricity price in the UK. Furió and Chuliá
(2012) use forward prices of Spanish electricity, Brent crude oil and Zeebrugge natural gas. Similarly
to the other studies, they also find a cointegration relationship where causation runs from fuel prices
to the electricity market.

10
stationary time series and cointegration settings. Xiao (2009) proves that a kernel

estimator of the varying cointegration coefficients is super-consistent. A kernel es-

timator is used to estimate this regression by locally weighing all observations with


K zth−z . The estimator of βb is defined as

n
X  
zt − z 2
βb (z) = arg min K {yt − x′t β} (2)
β
t=1
h

In this paper, the kernel estimator and bandwidth selection of the semiparametric

varying smooth coefficient model is implemented as given in Li and Racine (2007) and
in the np package by Hayfield and Racine (2008). The semiparametric varying smooth
coefficient model is then given as yt = β(zt )′ xt + ut . The electricity price is defined

as yt , while xt is a matrix of a constant and of gas and carbon prices. The regression
coefficient β (zt ) is a vector of unspecified smooth functions of z, which is the residual

load.8 In this model, the gas and carbon price dependence of the electricity price
varies with the effect modifier z. This means that the cointegration coefficients change

throughout the assumed underlying merit order.

I estimate different models for base, peak and off-peak electricity prices to account
for different underlying fundamentals. The semiparametric cointegration coefficients

for gas and carbon are shown in Figure 1. These functions measure the input price
sensitivity of the electricity price depending on the residual load.9

A visual inspection shows that the parameters vary throughout the merit order and
that there are two distinct parts. The first part has a higher carbon sensitivity, while
the second part has a higher gas sensitivity. The transition point lies at around 55 GW

average daily residual load for the base electricity price and at around 60 GW average
8
In Xiao (2009), the process zt is required to be stationary, which is the case for all residual load
processes of the base, peak and off-peak blocks. See Table 3 for the according unit root tests.
9
Due to the estimation procedure, parameters at the fringe of the load spectrum are unstable and
therefore omitted in the graphs.

11
 Beta parameters for gas Beta parameters for carbon
0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Beta parameter

Beta parameter
Off−Peak

40000 45000 50000 55000 40000 45000 50000 55000

Residual load Residual load

0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Beta parameter

Beta parameter
Base

45000 50000 55000 60000 45000 50000 55000 60000

Residual load Residual load

0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Beta parameter

Beta parameter
Peak

50000 55000 60000 65000 50000 55000 60000 65000

Residual load Residual load

Figure 1: Semiparametric cointegration parameter estimates of fuel prices

Notes: This figure shows the estimated semiparametric cointegration coefficients for off-peak, base
and peak electricity prices. The parameters are a smooth function that depends on the residual
load in MW. The coefficients for natural gas are displayed in the left column and the parameters for
carbon emission allowances are in the right column.

12
residual load for the peak block. The position of the shifting coefficients reflects the

German generation portfolio. Nuclear, lignite and coal based electricity production has
a total capacity of approximately 57 GW. These technologies are generally assumed
to have lower marginal costs than gas based production. The model indicates that

the gas driven part of the merit order has a generation capacity of approximately 10
GW. This estimate is also highly consistent with the power plant portfolio, as there is

a total gas fueled capacity of around 13 GW in Germany.

One needs to be careful with an economic interpretation of pass-through rates in

this model. Gas and carbon prices are used as a proxy for input prices as a whole.
Thus, the direct effect of each variable itself might be misleading. Rickels et al. (2010)
find a positive effect of the coal and oil prices on the carbon price, which may be

caused by a common factor of general demand for energy. To measure a meaningful

pass-through rate, I determine how the electricity price increases when the input prices

as a whole increase by one percent. The mean of the sum of the parameter vectors is
0.745% for off-peak, 0.835% for base and 0.906% for peak. The first and third quartiles

are within bounds of 0.05 percentage points below and above the point estimates.
These values can be interpreted as the pass-through rate multiplied by the portion
that fuel costs contribute to the total marginal costs. Given this interpretation, it

makes sense that the estimate is higher for peak, because the fuel costs are relatively
more important. The results of this analysis suggest that fuel price changes are passed

through.
As a robustness test, the comparable parametric VECM estimates of the cointe-

gration vector are 0.51 for gas and 0.36 for carbon (see Table 4). These estimates are
also consistent with the results of Fezzi and Bunn (2009). Using a similar setup for
the English market, they find cointegration parameters of 0.66 for gas and 0.32 for

carbon. The differences might be driven by a higher ratio of gas production in the

13
UK.

The QQ-plots in Figure 2 show a good fit of the semiparametric model. It is

able to resemble the pricing behavior for normal price levels, but underestimates the
highest prices. This probably happens due to a scarce capacity effect that causes a

price premium that cannot be explained by fuel price changes.

QQ−Plot for base QQ−Plot for peak QQ−Plot for off−peak

Fitted log off−peak electricity price

Fitted log base electricity price

Fitted log peak electricity price

4.5
5.0
4.5

4.0
4.5
4.0

3.5
4.0
3.5

3.5

3.0
3.5 4.0 4.5 3.5 4.0 4.5 5.0 3.0 3.5 4.0 4.5

Realized log base electricity price Realized log peak electricity price Realized log off−peak electricity price

Figure 2: QQ-plot for the fit of the semiparametric model

The estimates of the semiparametric model can be used to predict the changes of
the merit order for different gas and carbon price scenarios. Load-varying beta param-

eters translate into flexible shifts of the merit order. Figure 3 shows the estimated base
electricity prices depending on load and input prices. The graph on the left illustrates

equal gas and carbon prices that vary from 10 Euro to 25 Euro, which is a realistic
scenario for the observed period. The right graphs show the merit order for varying

gas prices while holding the carbon price fixed. Due to the semiparametric estimates,
the gas price has a stronger impact on the electricity price if the load is high.
The model is capable of explaining the observed electricity prices with a flexible

and simple approach. The relationship between electricity, natural gas and carbon
prices is motivated by the underlying power plant portfolio. In the next section, the

model is used to analyze the impact of an unexpected and sudden change of the power
plant portfolio.

14
 Merit order for normal price scenarios Merit order for gas price variation
100

100
80

80
Base Electricity Price

Base Electricity Price

60

60
40

40
20

20
0

0
40000 45000 50000 55000 60000 65000 40000 45000 50000 55000 60000 65000

Residual Load Residual Load

Carbon and gas price vary identically between 10€ and 25€ Constant carbon price: 10€, Gas price varies between 10€ and 25€

Figure 3: Simulated merit order for different natural gas and carbon price scenarios

Notes: This figure illustrates the fitted merit order conditional on varying gas and carbon prices.
The fitted base electricity price (in Euro/MWh) is derived using the semiparametric cointegration
coefficients shown in Figure 1. The chart on the left shows the merit order for gas and carbon prices
varying identically between 10 and 25 Euro per MWh and per ton, respectively (in steps of 3 Euro).
For the chart on the right, the carbon price is fixed at 10 Euros per ton and the gas price varies
between 10 and 25 Euros per MWh.

4 Analysis of the German Nuclear Moratorium in

2011

On Friday, 11 March 2011, a heavy earthquake and tsunami hit Japan and severely
damaged the nuclear power plant in Fukushima. Following these disastrous events,
the German government surprisingly decided to put a nuclear suspension in place.

The decision for a moratorium of three months length was announced publically on
the evening of Monday, 14 March 2011. This policy intervention immediately removed

seven nuclear power plants from the market.

In this section, I conduct an event study in order to assess the impact of the nuclear

moratorium. According to Binder (1998), event studies are used to test if a market
efficiently incorporates information and to analyze the event’s price impact on some

15
securities. Classical event studies in finance focus on measuring the abnormal returns

around a firm specific or economy wide event of interest. MacKinlay (1997) gives an
overview about event study methods, which all start by defining the event of interest
and the event window, during which the impact of the event is measured. The event

of interest is the announcement of the moratorium and the event window is chosen to
be 10 trading days before and 25 trading days after the announcement.

The impact of the moratorium on the German electricity price can be assessed
either for the spot or the futures market. Looking at the German base electricity

day-ahead spot market first, the price fluctuated in a range between 45 and 60 Euro
per MWh without showing a clear trend in March 2011. However, due to the large
variation of spot prices, it is not possible to rule out an effect of the nuclear morato-

rium. Generally, the influence on the day-ahead market was less distinct because the
suspended capacity was comparable to the normal fluctuations of renewable electricity

production, which had a high availability during the time of the announcement of the
moratorium. Thus, the price effect on the day ahead-market was small and short-lived

as concluded by the European Commission (2011). Looking at cross-border electric-

ity trade flows, the immediate impact of the moratorium is obvious for the period
surrounding the commencement. According to the report of the Bundesnetzagentur

(2011), Germany turned from being a net-exporter of 4.1 GW to an net-importer of

1 GW, effectively compensating for the nuclear capacities by imports from the neigh-

boring European countries.

However, only analyzing the direct impact on the spot market has two distinct

drawbacks for a comprehensive event study of the moratorium. First, the real price
impact of the moratorium may be concealed by the high variance of spot prices or fac-
tors like the large fluctuation of electricity production from renewable energy sources.

Second, it is not possible to measure the expectations of the market regarding the

16
question whether the moratorium will be permanent.

Therefore, the following part of the event study focuses on the futures market,
which is well suited to analyze the impact of the moratorium because futures prices
reflect the expectations of all market participants. Furthermore, the derivatives mar-

kets of the EEX have a sufficiently high liquidity as the trading volumes are about two
to five times higher than at the spot market. Given the high importance of the fu-

tures market, most institutions such as the EEX (2011) and the European Commission
(2011) focus on futures prices to analyze the impact of the moratorium.

The electricity futures traded at the EEX reacted with a steep price increase,
which is shown in Figure 4. Given an instant daily price increase of roughly 15%, the
mere existence of a moratorium effect is obvious for the electricity futures. However,

also the gas and carbon futures prices rose, probably because the market expected
an increasing demand for fossil fuels, which are used to offset the suspended nuclear

capacities. As a consequence, the event study focuses on analyzing the impact of the
different influences that cause the electricity prices to rise. The method proposed in

this study allows to determine whether the market efficiently accounts for the new
information.
In theory, there are two separate shifts of the merit order for the according elec-

tricity futures with delivery between March 2011 and June 2011. First, the supply
curve is shifted left by about 6 GW, because nuclear generation capacity with low

marginal costs is removed from the system. This effect is called the capacity effect
of the moratorium. Second, the increased gas and carbon futures prices result in an

upwards shift of the merit order.

The event study is conducted in the following way. In order to isolate the capac-
ity effect for a certain electricity futures contract, I compare the observed electricity

futures price and the predicted merit order for the contract before and after the mora-

17
 70 Base Electricity Futures prices

Base Future April 2011

Base Future May 2011
Base Future June 2011
65

Base Future Q2 2011

Price in €/MWh

60
55
50
45

Mar 01 Mar 15 Apr 01 Apr 15

Figure 4: Base electricity futures prices at the time of the announcement of the nuclear
moratorium

Notes: This figure shows the EEX market reaction for base electricity futures that are directly affected
by the nuclhear moratorium. The moratorium was announced on the 14 March 2011.

torium. The semiparametric cointegration model, as discussed in Section 3, is used to

predict the merit order, i.e. the counterfactual electricity price function conditional on
residual load.10 Due to the varying beta coefficients, the observed natural gas and car-

bon emission allowance futures prices are sufficient to derive such a merit order curve
for an electricity futures contract.11 As the predicted merit order only accounts for the

change in gas and carbon futures prices, it is possible to derive the capacity effect of
the moratorium. First, the merit order of the electricity futures contract is predicted
using the observed settlement prices of the according natural gas and carbon futures

on a trading day before the moratorium. Then, the settlement price of the electricity
future on the same trading day is used to determine the implied expected demand,
10
Generally, cointegration is seen as a long-term framework, but in this context it is reasonable
to assume that the stable long-term relationship is relevant for the price expectations at the futures
market.
11
The event study uses unadjusted futures prices in levels. The model is calibrated with data from
the day-ahead market with seven days per week to match the delivery structure of the base futures.

18
which is defined as the residual load that is necessary to justify the observed electricity

futures price. This is achieved by calculating the intersection of the predicted merit
order and the actual observed electricity futures settlement price. In the second step,
the same procedure is repeated for electricity, gas and carbon futures prices observed

on a trading day after the moratorium. The difference of the implied expected demand
before and after the moratorium is the capacity effect.

This procedure is illustrated in Figure 5, which shows as an example how the

capacity effect is derived in this study. The capacity effect is determined for the Q2

2011 base electricity futures contract and the period between March 9 and March
24. The merit order shown in this figure is calculated with the semiparametric model
using the natural gas futures prices for delivery in Q2 2011 and the carbon emissions

allowance futures price for delivery in mid-December 2011 as inputs. The dashed bold
line is the predicted merit order derived from the futures settlement prices traded

on March 9. The implied expected residual load for the setup in Figure 5 can be
calculated by taking the intersection of the merit order and the observed electricity

futures settlement price on the same trading day. This expected residual load for Q2
2011 amounts to 47.5 GW on 9 March 2011, which is close to the 2008 - 2010 average
of 48.3 GW. Driven by the moratorium, the gas and carbon futures prices rise and

shift the Q2 2011 merit order upwards, as it is shown by the bold line representing
the merit order for the gas and carbon futures prices traded on March 24. This gas

and carbon price effect accounts for an electricity price increase of less than 3 Euro.
However, from March 9 to March 24, the Q2 2011 electricity futures settlement price

rose from 49.75 Euro to 56.90 Euro. This observed change of the electricity futures
price is used to determine the capacity effect of the nuclear moratorium by looking at
the merit order on each of both trading days and asking which residual load would

justify the actual electricity futures price. The difference between this implied residual

19
 Impact of the Nuclear Moratorium − Q2 2011 Future
65

Merit Order: 09 March 2011

Merit Order: 24 March 2011
60
Base Electricity Price in €/MWh

55
50

Capacity effect: 3911 MW

45
40

47479 51390

46000 48000 50000 52000 54000 56000

Residual Load in MW

Figure 5: Derivation of the capacity effect of the nuclear moratorium

Notes: The merit order is derived by using the previously calibrated semiparametric model as well as
gas and carbon futures settlement prices on a specific trading day. Due to the rising gas and carbon
futures prices, the merit order shifts upwards from March 9 to March 24. While the fuel price effect
accounts for an electricity price increase of less than 3 Euro, the actual futures price rose by 7.15
Euro. The capacity effect is determined by calculating the residual load that would justify the actual
electricity futures prices on each of both trading days and then taking the difference between the
implied residual load before and after the moratorium.

20
load before and after the moratorium is the capacity effect. It can be interpreted as

the additional demand that would be necessary to drive the electricity price to the
observed level if the nuclear moratorium had not been imposed. For the setup shown
in the graph of the Q2 2011 future, the capacity effect amounts to 3.9 GW.

The proposed approach allows to measure the capacity effect, but is based on the
assumption that futures prices mainly reflect expected spot prices and market condi-

tions. Analyzing only the futures market without the calibrated market model would
not allow to directly link the observed price developments to the capacity effect.12

The analysis in Figure 5 displays only the capacity effect for one single futures
contract and for the comparison of two trading days. Thus, in the next step, the
same procedure is used to calculate the capacity effect for different electricity futures

and the full range of trading days in the event study window. Figure 6 shows the
development of the capacity effect over time and for futures contracts with different

times to maturities. For example, in order to calculate the moratorium’s capacity effect
for the Q2 2011 futures contract traded on March 24, I compare the implied expected

demand for the contract on this trading day with the implied expected demand for the
same futures contract on all trading days before the announcement of the moratorium.
Finally, the average of these capacity effects between March 24 and each of the trading

days before the moratorium is displayed in Figure 6 as the capacity effect for the Q2
2011 contract on March 24.

The top panel of Figure 6 displays the capacity effect for directly affected futures.
12
The results derived with the approach in this study are consistent with the findings of Fritz
(2012) using a different setup in this working paper. Fritz (2012) estimates a VECM for European
energy futures prices, namely EEX base electricity prices, Title Transfer Exchange (TTF) natural gas
prices, Intercontinental Exchange (ICE) coal prices and EEX carbon emissions allowances. The results
regarding the cointegration properties are similar to Section 3, finding one cointegration relationship.
Within the VECM framework using futures prices only, the author finds an electricity price increase,
which cannot be explained by fuel prices, that is of the same magnitude as estimated in Figure 5
using the semiparametric approach. However, the setup in Fritz (2012) does not allow to express the
price increase as a capacity effect (in GW) due to the setup of the model using futures prices only.

21
On Monday, 14 March 2011, the first trading day after the Fukushima events, the prices

of the electricity, gas and carbon futures rise. However, the capacity effect, which
measures the abnormal price increase of electricity futures, shows no indication of
previous information about the moratorium. There is no evidence for a capacity effect

before 15 March 2011. Then, in direct response to the moratorium, all futures contracts
immediately account for the shut capacity of about 6 GW. The market efficiently

reacts to the moratorium by adding a capacity effect premium to the electricity price
in order to reflect the missing generation capacity. In the following days, the capacity

effect declines first, but remains at a rather stable level after this drop. This decline
might have been caused by the fact that the market agents did not anticipate a nuclear
moratorium and thus needed some time to develop sound forecasts. After a few trading

days, the market agents expect that a part of the capacity effect will be mitigated
by dynamic factors like the flexibility of the power plant portfolio or international

transmission.
The framework also allows measuring the market’s expectations for the time after

the end of the moratorium in June 2011. The middle and bottom panel of Figure 6
show the capacity effect for several futures with delivery after the moratorium. For
the quarterly future with delivery in Q3 2011, the development of the capacity effect

reveals an unsteady reaction, which is lasting for a few trading days, before sound
expectations have developed. Then, the market expects a capacity effect of roughly 3-

4 GW for the time after the moratorium. The capacity effect for the following quarter
is at a very similar level, but more stable over time. The yearly futures for 2012 and

2013 also reveal a more settled picture. There is no panic reaction and the markets
quickly adjust to a stable level of around 1 GW missing nuclear capacity.
Generally, the capacity effect for futures with delivery during and directly after

the moratorium is rather similar. Thus, there is an impact that is expected to be

22
 Capacity Effect for Base Electricity Futures with Delivery during the Moratorium
8000

Base Future April 2011

Base Future May 2011
Base Future June 2011
Base Future Q2 2011
6000
Capacity Effect in MW

4000
2000
0

Mar 01 Mar 15 Apr 01 Apr 15

Trading Days

Capacity Effect for Base Electricity Futures with Delivery after the Moratorium
5000

Base Future Q3 2011

Base Future Q4 2011
4000
3000
Capacity Effect in MW

2000
1000
0
−1000

Mar 01 Mar 15 Apr 01 Apr 15

Trading Days

Capacity Effect for long−term Base Electricity Futures with Delivery after the Moratorium
2500

Base Future 2012

Base Future 2013
2000
1500
Capacity Effect in MW

1000
500
0
−500

Mar 01 Mar 15 Apr 01 Apr 15

Trading Days

Figure 6: Implied capacity effect of the nuclear moratorium

Notes: This figure shows the implied capacity effect (in MW) that is caused by the nuclear mora-
torium. Only the futures illustrated in the top panel are directly affected by the moratorium. The
capacity effect is calculated with the same procedure that is depicted in Figure 5.

23
permanent. It is difficult to quantify the expected number of nuclear power plants

to remain closed down as there is some uncertainty introduced by dynamic effects.

These effects could be a change of the maintenance schedule, endogenously added new
generation capacity, changes of international transmission and demand responses. This

dynamic adjustment process mitigates some of the capacity effect. Second, weighted
expectations for different political scenarios might be reflected in the prices. If market

participants think that several scenarios are realistic, the estimated capacity effect will
reflect an average expectation that might not be a realistic scenario itself.

Given these considerations, there are two possible explanations for the decaying
capacity effect: (1) that the moratorium of 6 GW has an expected capacity effect of
only 1 GW in 2013 due to dynamic adjustment effects, or (2) that the market expects

that the probability of an extension of the moratorium decreases with the time to
maturity and is relatively low for 2012 or 2013.13

However, there is still consistent evidence for the existence of a capacity effect for
all futures with delivery after the end of the moratorium. Thus, one can conclude that

the market on average correctly expects an extension of the moratorium with several
nuclear power plants remaining closed down after the announced end in June 2011.

5 Conclusion

There are two main contributions of this paper. First, it shows that the relationship
between the input fuel prices and the electricity price varies with load and reflects

the underlying merit order. This result is potentially useful for other markets with
13
The finding that the capacity effect decays with the time until delivery might also be partially
driven by the well-known Samuelson (1965) effect that commodity futures with a longer time to
maturity are less volatile. In this case, both the electricity, gas and carbon futures for 2012 and
2013 would react less to new information than futures for 2011. However, this can also be explained
economically, as the long-term futures are not directly affected by the moratorium and additionally
would allow more time for dynamic adjustment effects.

24
different production technologies and inputs. One example are commodity markets,

where local conditions lead to different mining or extraction technologies.

Second, the paper provides a framework to assess the impact of the German nuclear
moratorium in 2011. The market incorporates the new information efficiently and

correctly expects that several power plants will remain shut off after the moratorium.
Furthermore, it anticipates that dynamic adjustment processes will mitigate some of

the capacity effect. However, these results are not necessarily applicable for additional
plant closures, which could affect the security of supply or lead to substantial capacity

premium effects.
The approach in this paper could be improved and extended in several ways. It
would be desirable to include other fuels to get a more granular picture of the nonlinear

fuel price effects. It would also be interesting to test and compare the fuel price effects
for various markets with different dominating technologies. Accounting for a possible

scarce capacity premium, which seems to exist, would also improve the model.
Due to the semiparametric approach, the demand elasticity is not included explic-

itly. However, Fezzi and Bunn (2010) show that it is preferable to model demand as an
endogenous variable. The analysis of the nuclear moratorium focuses on the German
futures market, but does not include the day-ahead market or indirect price effects on

other European markets. The impact on these markets and the response of input fuel
prices to the moratorium provide an interesting area for future research.

References

Armitage, S. (1995). Event study methods and evidence on their performance. Journal

of Economic Surveys, 9(1):25–52.

Asche, F., Misund, B., and Sikveland, M. (2013). The relationship between spot and

25
 contract gas prices in Europe. Energy Economics, 38:212 – 217.

Binder, J. (1998). The Event Study Methodology Since 1969. Review of Quantitative
Finance and Accounting, 11:111–137.

Bosco, B., Parisio, L., Pelagatti, M., and Baldi, F. (2010). Long-run relations in

European electricity prices. Journal of Applied Econometrics, 25(5):805–832.

Bundesnetzagentur (2011). Auswirkungen des Kernkraftwerk-Moratoriums auf die

Übertragungsnetze und die Versorgungssicherheit - Bericht der Bundesnetzagentur

an das Bundesministerium für Wirtschaft und Technologie. Report, Bundesnetza-
gentur.

Cai, Z., Li, Q., and Park, J. Y. (2009). Functional-coefficient models for nonstationary
time series data. Journal of Econometrics, 148(2):101–113.

Coulon, M. C. and Howison, S. (2009). Stochastic Behaviour of the Electricity Bid

Stack: From Fundamental Drivers to Power Prices. The Journal of Energy Markets,

2(1).

Dyckman, T., Philbrick, D., and Stephan, J. (1984). A comparison of event study
methodologies using daily stock returns: A simulation approach. Journal of Ac-

counting Research, 22:1–30.

EEX (2011). Market Monitor by the Market Surveillance of EEX Q1 & Q2 / 2011.

Market surveillance report, European Energy Exchange.

European Commission (2011). Quarterly Report on European Electricity Markets.

Market Observatory for Energy 4(1), European Commission.

Fell, H. (2010). EU-ETS and Nordic Electricity: A CVAR Analysis. The Energy
Journal, 31(2):1–26.

26
Ferkingstad, E., Loland, A., and Wilhelmsen, M. (2011). Causal modeling and infer-

ence for electricity markets. Energy Economics, 33(3):404–412.

Fezzi, C. and Bunn, D. W. (2009). Structural interactions of European carbon trading

and energy prices. The Journal of Energy Markets, 2(4):53–69.

Fezzi, C. and Bunn, D. W. (2010). Structural Analysis of Electricity Demand and

Supply Interactions. Oxford Bulletin of Economics and Statistics, 72(6):827–856.

Fritz, A. (2012). European electricity and interrelated futures markets: A cointegrated

vector autoregressive analysis. EWL Working Papers 1205, University of Duisburg-
Essen, Chair for Management Science and Energy Economics.

Furió, D. and Chuliá, H. (2012). Price and volatility dynamics between electricity and
fuel costs: Some evidence for Spain. Energy Economics, 34(6):2058–2065.

Hastie, T. and Tibshirani, R. (1993). Varying-Coefficient Models. Journal of the Royal

Statistical Society. Series B (Methodological), 55(4):757–796.

Hayfield, T. and Racine, J. S. (2008). Nonparametric Econometrics: The np Package.

Journal of Statistical Software, 27(5):1–32.

Kanamura, T. and Ohashi, K. (2007). A structural model for electricity prices with

spikes: Measurement of spike risk and optimal policies for hydropower plant opera-
tion. Energy Economics, 29(5):1010–1032.

Li, Q. and Racine, J. (2007). Nonparametric Econometrics: Theory and Practice.

Princeton University Press.

MacKinlay, A. C. (1997). Event Studies in Economics and Finance. Journal of Eco-

nomic Literature, 35(1):13–39.

27
Mjelde, J. W. and Bessler, D. A. (2009). Market integration among electricity markets

and their major fuel source markets. Energy Economics, 31(3):482–491.

Mohammadi, H. (2009). Electricity prices and fuel costs: Long-run relations and
short-run dynamics. Energy Economics, 31(3):503–509.

Mount, T. D., Ning, Y., and Cai, X. (2006). Predicting price spikes in electricity
markets using a regime-switching model with time-varying parameters. Energy Eco-

nomics, 28(1):62–80.

Renou-Maissant, P. (2012). Toward the integration of European natural gas markets:

A time-varying approach. Energy Policy, 51:779 – 790.

Rickels, W., Görlich, D., and Oberst, G. (2010). Explaining European Emission Al-
lowance Price Dynamics: Evidence from Phase II. Kiel Working Papers 1650, Kiel

Institute for the World Economy.

Samuelson, P. A. (1965). Proof that properly anticipated prices fluctuate randomly.

Industrial Management Review, 6(2):41–49.

Weron, R., Bierbrauer, M., and Trueck, S. (2004). Modeling electricity prices: jump
diffusion and regime switching. Physica A, 336(1-2):39–48.

Xiao, Z. (2009). Functional-coefficient cointegration models. Journal of Econometrics,

152(2):81–92.

Zachmann, G. and von Hirschhausen, C. (2008). First evidence of asymmetric cost

pass-through of EU emissions allowances: Examining wholesale electricity prices in

Germany. Economics Letters, 99(3):465–469.

28
A Appendix

Table 3: Augmented Dickey Fuller unit root tests

Variable Level 1st diff.

statistic p-value lags statistic p-value lags
Base Electricity -2.25 0.19 9 -11.73 0.00 8
Peak Electricity -2.17 0.22 9 -11.65 0.00 8
Off-peak Electricity -2.58 0.10 9 -12.35 0.00 8
NCG Gas -0.45 0.52 1 -19.63 0.00 1
EU Emission Allowance -0.65 0.43 0 -10.98 0.00 5
Base Residual Load -3.82 0.00 9
Peak Residual Load -3.12 0.03 15
Off-peak Residual Load -3.01 0.03 10
Notes: The null hypothesis of the ADF test is that there is a unit root in the considered time
series. Lag lengths are determined by the Akaike Information Criterion (AIC). Whether to include
a trend or constant was decided by checking the significance of the trend/constant parameters at a
5% significance threshold.

29
Table 4: Johansen cointegration analysis of electricity, gas and carbon prices

Panel A. Cointegration tests

Rank Trace test statistic p-value
Base electricity, gas, carbon
0 120.48 0.000
1 15.56 0.200
2 3.93 0.435

Peak electricity, gas, carbon

0 103.08 0.000
1 15.82 0.187
2 4.05 0.417

Off-peak electricity, gas, carbon

0 169.89 0.000
1 15.27 0.215
2 3.81 0.454
Notes: The Johansen test is used to test for the existence and rank of a
possible cointegration relationship between the three I(1) variables electricity,
gas and carbon. The constant is restricted to lie in the cointegration space, as
there is no indication for trends in the data. The lag length is determined by
the Schwarz Information Criterion (SIC). The trace statistic for rank j tests
the null hypothesis of rank r = j against r > j.

Panel B. Analysis of the cointegration parameters

α-Vector β-Vector
Parameter t-stat. Parameter t-stat.
Base -0.297 -10.58 1 -
Gas 0.012 1.06 -0.51 -9.29
Carbon -0.002 -0.27 -0.36 -4.50
Notes: The α-parameters indicate if and at which speed the variable of inter-
est reacts to a disequilibrium in the long-term relationship. In the equations
for gas and carbon, the α-parameters are not significant and thus, the gas
and carbon prices are treated to be weakly exogenous. The estimates of the
β-vector are significant, which shows that both gas and carbon prices are part
of the stable long-term relationship and important drivers of the electricity
price.

30