Market power in the German wholesale electricity market

Transcription

1 The Journal of Energy Markets (47 74) Volume 2/Number 2, Summer 2009 Market power in the German wholesale electricity market Dominik Möst Institute for Industrial Production, Universität Karlsruhe (TH), Karlsruhe Institute for Technology (KIT), Hertzstrasse 16, Karlsruhe, Germany; Massimo Genoese Institute for Industrial Production, Universität Karlsruhe (TH), Karlsruhe Institute for Technology (KIT), Hertzstrasse 16, Karlsruhe, Germany; Since 2001, wholesale electricity prices have increased dramatically in Europe and especially in Germany. It has been argued that utilities have been exercising market power by withholding available power plant capacity. In this paper we investigate the exercise of market power in the German wholesale electricity market with an agent-based simulation model that uses detailed German wholesale power market data. The analysis was carried out for the years 2001, 2004, 2005 and We start with 2001 as it is seen as a year with well-functioning competition that validates this model. The year 2004 was chosen because it was the last year without emissions trading. In 2005 the EU emissions trading scheme started; this was accompanied by rising prices and windfall profits for electricity generating companies; 2006 was chosen because it supposedly suffered from bad competition. We test our results with the Lerner Index, but find that they do not necessarily confirm the exertion of market power. 1 INTRODUCTION The liberalization of electricity markets in Europe started in the early 1990s and has since led to claims that energy suppliers have been exercising market power, for example in the UK, Scandinavia and Germany. Shortly after liberalization, German wholesale electricity prices dropped sharply but soon began to rise again, on average from 24 per megawatt hour ( 24/MWh) in 2001 to 51/MWh in 2006, representing an increase of 113%. In the same period, gas prices rose by 74% and the European Emission Trading Scheme was introduced. We would like to thank Dr Frank Sensfuß from the Fraunhofer Institute for Systems and Innovation Research in Karlsruhe for his contribution to the joint model development and the Volkswagen Foundation (VW-Stiftung) for financing this research work. 47

2 48 D. Möst and M. Genoese Generally speaking, regulated electricity markets have been characterized by considerable overcapacities. Efforts to reduce these overcapacities began in the early years of the liberalization process. At the same time, a concentration of energy utilities took place in Germany (ie, the merger of Preussag and Bayernwerk to form E.ON). Four companies were responsible for about 80% of Germany s electricity generation capacity in 2007, which was deemed an oligopolistic structure and harshly criticized in public debate. Several studies reported rising prices and argued that the rise was not merely caused by higher fuel prices and the introduction of emissions trading, but was also due to the exercise of market power by the energy companies. In this paper we analyze market power potential and its exercise in the German wholesale power market. The analysis is conducted with the detailed agent-based electricity market simulation platform PowerACE (power agent-based computational economics), which is able to simulate relevant market participants and hourly spot market prices. The main difference between our work and prior studies is the broad time interval used in our analysis and the choice of modeling approach, which might be more appropriate than previous approaches for testing the exertion of market power. An overview of indicators for measuring market power is provided in Section 2; the relevant literature on market power in the German wholesale electricity market is reviewed in Section 3. A general overview of agent-based simulation and a description of the developed model are given in Section 4. In Section 5, we discuss our results. We summarize our main findings and draw conclusions in Section 6. 2 APPROACHES FOR MEASURING MARKET POWER IN LIBERALIZED ELECTRICITY MARKETS There are a number of reports dealing with market power in liberalized energy markets in various countries: the UK (see, for example, Green and Newbery (1992); Von der Fehr and Harbord (1993); Newbery (1998); Wolfram (1999); Bunn and Martoccia (2005)), the US (especially California) (see, for example, Borenstein et al (1995); Sweetser (1999); Surratt (1998)), Australia (see, for example, Brennan and Melanie (1998); Tamaschke et al (2005a)) and Germany (see, for example, Bower et al (2001); Müsgens (2006); Schwarz and Lang (2006); Von Hirschhausen et al (2007)). Three parameters can be used to analyze the functioning of markets: market structure, market behavior and market performance/result. Several indicators exist for determining these three elements. In the following, selected indicators will be briefly discussed. The concentration ratio (CR) is used by several regulators, eg, the German Federal Cartel Office (Bundeskartellamt; hereafter abbreviated as GFCO ), for the determination of structural market power. The Residual Supply Index (RSI) was developed especially for measuring market power in the electric- The Journal of Energy Markets Volume 2/Number 2, Summer 2009

3 Market power in the German wholesale electricity market 49 TABLE 1 Concentration ratio for n D 1; 2; 3; 4 according to (1). Critical values of the n GFCO ity industry. The Lerner Index measures exercised market power based on market results. In general, tests for structural market power are applied by authorities in order to analyze market power. Nevertheless, it first has to be determined whether testing the market structure is useful for analyzing the exercise of market power. In Germany, the aggregated market share of the largest companies is taken to calculate the concentration ratio as follows: CR n D P n id1 S i P m id1 S i (1) with: S i being the capacity of generator i; S i > S j 8i<j; and m being the number of generators. The Herfindal Hirschman Index (HHI) is quite similar to this ratio, but it is mainly used in US literature. The GFCO defines critical threshold values for the concentration ratio. A market-dominating position is assumed when the calculated values are higher than the critical threshold values. The GFCO assumes a dominating position for CR 1 > 33:3% (one company), CR 3 >50% (three companies) and CR 5 > 66:7% (five companies) (see Bundeskartellamt (2001)). Similar critical threshold values exist for the HHI. For the case of Germany, CR 1 is smaller than 0.33, but CR 3 is higher than 0.5 (see Table 1). On the basis of the concentration ratio, the GFCO makes decisions concerning company mergers. It is important to note that the threshold values are not adapted to market characteristics such as high capital requirements, etc. Extensions of these ratios take the temporal variations of supply and demand into account. The RSI measures the influence of a supplier on the demand in a specific hour as follows: RSI i;h D S tot;h S i;h (2) D h Research Paper

4 50 D. Möst and M. Genoese with: S tot;h being the total available capacity in hour h; S i;h being the available capacity of company i in hour h; and D h being the demand in hour h. The RSI thus takes different levels of the demand into account. Depending on the demand level, the suppliers might be able to exercise market power. The RSI should not be less than 110% in 5% of the hours of a year (approximately 438 hours). If the RSI is below 100% for a supplier, the capacity of this company is pivotal for satisfying demand. Nevertheless, it has to be kept in mind that the RSI can only provide the necessary condition for market power; it is not a sufficient condition for the exercise of market power. The results of the RSI for the German market are discussed in Section 5.4. To measure whether market power is being exercised or not, the market results have to be analyzed. This is realized by estimating the supply curves and marginal costs of power production. The estimated marginal costs of power production are then compared with the market results. The Lerner Index, defined above, is widely used to calculate the difference between the market price and marginal price in hour h (see Tamaschke et al (2005)): with: L p h D p h MC h or L MC h D p h MC h (3) p h MC h p h being the market price in hour h; and MC h being the competitive price in hour h. The bid-up can be based on real prices.l p / or marginal costs.lmc/. If the index h h is based on marginal costs, it is sometimes also referred to as the price cost margin index (PCMI). 1 The closer the index is to zero, the more competitive the market is. In contrast, the higher the value, the greater the degree of exercised market power is. One criticism of the Lerner Index is that with increasing electricity prices (ie, caused by rising production costs after the introduction of emissions trading), the index decreases even if the gap between prices and competitive prices remains the same. However, another major problem is the lack of data, which is rarely available for estimating supply curves and calculating marginal costs. Different simulation techniques are generally used to calculate marginal costs. Therefore, the different assumptions of the model calculations have to be taken 1 As in general p h > MC h, then L p h <LMC, consequently we choose the larger value LMC for h h discussion in Section 5.3. The Journal of Energy Markets Volume 2/Number 2, Summer 2009

5 Market power in the German wholesale electricity market 51 into account. 2 In the following, a sophisticated agent-based simulation model with extensive market data is used to calculate the values of the Lerner Index for several years in the German wholesale market. It should be noted that researchers as well as energy supply companies use such programming techniques to make decisions about their generation bidding. Thus, the application of such a model for analyzing market power appears to be suitable. Before the model and its assumptions are explained in detail, the current debate on market power in Germany will be briefly discussed; several research papers and studies will be cited in this discussion. 3 REVIEW OF RELEVANT WORK ANALYZING MARKET POWER IN THE GERMAN ELECTRICITY MARKET In the last three to five years, wholesale electricity prices have increased dramatically in Europe and especially in Germany. From 2001 to 2006, the average yearly increase was approximately 13.4%. In Germany, doubts have been voiced as to whether electricity markets function competitively. In this context, it is supposed that electricity producers use market power to increase prices and thus boost their profits. It has been argued by some authors that suppliers exercise market power by withholding available power plant capacity. It is supposed that this practice would create an additional burden for end consumers and would result in welfare losses for society. The current debate is primarily based on empirical studies of price formation on the German wholesale electricity market. Most of the studies used fundamental market models to calculate the marginal costs of power production. On the basis of the difference between wholesale electricity prices and the calculated marginal costs of power production, the authors concluded that electricity markets are not competitive. The first study dealing with market power in Germany was published by Bower et al (2001). After the dramatic reduction in electricity prices in 1999 and 2000 due to the liberalization of the markets, Bower et al (2001) concluded that the price decrease created the circumstances for a benign endorsement of industry consolidation as well as a prolonged period of high prices. As prices had previously been quite low, the authors demonstrate using an agent-based simulation model that the process of strategic consolidation could have resulted in average annual peak prices rising by 87% and average annual off-peak prices increasing by 50%. They concluded that the creation of four dominant firms, which have the potential to strategically withdraw capacity, appears to have resulted in a significant increase in prices above competitive levels. Five years after the significant price increases for electricity, Müsgens (2006) started a new debate about market power in the German electricity market. For the 2 An overview of the different models deriving electricity prices can be found in Ventosa (2004). Research Paper

6 52 D. Möst and M. Genoese years , he calculated the hourly marginal costs for power production on the basis of a linear programming model. These calculations are considered to be a benchmark result for a completely competitive and well-functioning market. The marginal costs were then subtracted from real market results from the European Energy Exchange (EEX) to obtain the mark-up. On the basis of this difference, Müsgens concluded that until August 2001, prices were based on marginal costs and that the market was therefore functioning well. From September 2001 onward, he observed a shift toward an average increase of 50% and up to 77% in peak periods on top of the calculated marginal costs. Müsgens regarded this as strong evidence of market power. Schwarz and Lang (2006) calculated the marginal costs of electricity generation for the period from June 2000 to December 2005 on the basis of a linear programming model combined with a mixed-integer linear programming model. As emissions trading started in 2005, they considered the CO 2 prices as an additional production factor. Their results indicated that in 2000 and 2001, prices were close to marginal costs. From 2002 onward, they observed substantial deviations of market prices from marginal costs in peak times. Prices in 2003 were on average 30% higher than the calculated marginal generation costs. In 2004 and 2005, this difference was about 15%, but varied significantly in 2005 from month to month. Schwarz and Lang (2006) traced this variation back to uncertainties, especially political uncertainties, in that year. It was expected that emission levels in this year would form the basis for the allocated emission allowances for the second trading period. This resulted in a kind of prisoner s dilemma for energy utilities: if a producer could expand its production, it would receive additional free emission allowances for the next period. Finally, Schwarz and Lang (2006) concluded that the price increase could mainly be traced back to fundamental factors, such as rising fuel prices, and to the additional production factor of CO 2 allowances. The additional increase due to market power was considered to be quite small. Schwarz and Lang stated that significant increases due to market power could only be observed in 2003 at peak times, when demand was very high and free generation capacities were scarce. Von Hirschhausen et al (2007) analyzed market prices and market power with a linear programming model for the period from January 2004 to June Again, the authors argued that a difference between the marginal costs of generation and wholesale electricity prices could be considered an indication of market power exertion. On the basis of their model calculations, they concluded that in all periods, prices were much higher than marginal costs. Particularly in peak times, they observed prices significantly above their calculated marginal costs. They concluded that market power was being exercised on electricity markets. The main results of the four papers analyzing market power on the German wholesale electricity market are summarized in Table 2 on the facing page. Bower et al (2001) already presumed increasing electricity prices for future periods with a The Journal of Energy Markets Volume 2/Number 2, Summer 2009

7 Market power in the German wholesale electricity market 53 TABLE 2 Summary of the analysis concerning market power on the German market. Von Bower et al Musgens Schwarz and Hirschhausen (2001) (2006) Lang (2006) et al (2007) Interval /2000 to 06/2000 to 01/2004 to 06/ / /2006 Model Germany, Europe, linear Germany, Germany, horizon and agent-based program mixed linear methodology simulation integer program linear program Results The process Until 08/2001 Until 2001, Increasing of strategic prices near prices near mark-ups in consolidation marginal costs; marginal costs; every year, could result from 2002, from 2002, especially in significant significant significant during peak price increases mark-up on deviations from hours (87% peak, marginal costs marginal costs 50% off-peak) especially in peak hours strategic consolidation. The three later studies directly analyzed the increased prices. Although the authors applied slightly different model approaches, they all concluded that from 2002 onward, wholesale electricity prices were above marginal costs. In time intervals with a high electricity demand, they calculated a particularly significant difference between the electricity prices and the calculated marginal costs of generation. Thus, they concluded that market power had been exercised by means of withholding available capacities from the market. The last three papers cited above sparked further studies that criticized their general approach. Weber and Vogel (2007) and Swider et al (2007) observed that uncertainties concerning model data, especially concerning the efficiency of the generation portfolio as well as the availability of power plants, were not considered in an adequate manner. Furthermore, they stated that the applied models were too simple; they claimed that the models main components, such as start-up costs, electricity exchange and adequate reserve capacities, were not represented in detail. They also observed that the authors had failed to critique their own analysis. Thus, Weber and Vogel (2007) and Swider et al (2007) concluded that the earlier authors assertion, ie, that prices had increased due to market power, was not tenable on the basis of the results achieved with the simplified models. Our own calculations, which are based on a detailed agent-based model, will hopefully make a valuable contribution to the debate on market power in the German wholesale market. Research Paper

8 54 D. Möst and M. Genoese 4 OVERVIEW OF THE AGENT-BASED SIMULATION MODEL POWERACE 4.1 Requirements for a model analyzing market power in liberalized electricity markets Analyzing the possible exercise of market power requires a detailed model of the wholesale electricity market that is capable of reflecting the complexity of the market s structure (ie, electricity as a non-storable, homogeneous product). Moreover, many of the power plants technical restrictions have to be considered, such as startup costs, reserve requirements, availability and efficiency as well as the fluctuating character of wind energy. Additionally, a lot of data has to be included for a realistic simulation, including load profiles and electricity exchange as well as daily fuel and CO 2 certificate prices. Furthermore, cross-border transmission interconnections between Germany and neighboring countries play a significant role in the analysis of market power because the import capacity can be as high as 18 GW. The agent-based PowerACE model meets all of these requirements and was therefore adopted for this study. The model simulates the hourly plant dispatch and hourly prices. It has already been used for the analysis of CO 2 emissions trading (Genoese et al (2007)), the merit-order effect (Sensfuß et al (2008)) and the impact of emission allocation schemes on power plant investments (Genoese et al (2008)). 4.2 Agent-based modeling of electricity markets There are three major bottom-up approaches to modeling electricity markets: optimization models, equilibrium models and simulation models (see Ventosa et al (2005)). The agent-based simulation of the German electricity market presented in this paper can be classified as a simulation model. The concept of agent-based simulation allows a simulation from a player s perspective to be built that enables the integration of different players with individual strategies. The approach of agent-based simulation is based on several disciplines, such as economics, game theory, social sciences and software engineering (see Wooldridge (2002)). The variety of approaches to agent-based simulation has led to a variety of definitions for the term agent. A review of agent-based simulation platforms shows that the agents used in these simulations in many cases conform to weaker definitions of the term (see Drogoul et al (2002)). The agent-based analysis of economic systems is called agent-based computational economics. According to Tesfatsion (2006) the term agent refers to a computational entity that can be characterized by bundled data and behavioral methods. Agents can thus represent individuals, institutions or physical entities. Thanks to its high degree of flexibility, our approach allows us to build a very detailed model of the German electricity sector. Furthermore, the perspective of single actors can be included in the simulation. As shown in Sensfuß et al (2007), The Journal of Energy Markets Volume 2/Number 2, Summer 2009

9 Market power in the German wholesale electricity market 55 many agent-based simulation models show promising results but have not been validated. The model presented here is subjected to a validation process and thus might contribute to greater acceptance of agent-based models. 4.3 PowerACE model description Relevant players in the electricity system, such as consumers, utilities, renewable agents, grid operators, government agents and market operators, are modeled as one or several computational agents. More complex players, like utilities, are modeled using several agents representing different functions, such as trading or generation (see the agents Generator and Seller in Figure 1 on the next page). The following markets are included in our model: a spot market for electricity, markets for balancing power (primary, secondary and minute reserve), a market for CO 2 emissions and a forward market for electricity. The objective of this paper is to analyze whether spot market prices are influenced by market power. Thus, only the spot and reserve markets are activated for the simulation runs. An overview of the entire model and the main agents involved in the simulation is given in Figure 1 on the next page. The four main modules consist of markets, electricity demand, conventional electricity supply and renewable electricity generation. A detailed mathematical description of the PowerACE model can be found in Genoese et al (2007). The realistic simulation of spot market prices requires extensive data on electricity demand, renewable electricity generation and German power plants. This data is either provided by soft-links to various models (eg, the PERSEUS model (Möst et al (2005)) provides load profiles for electricity exchange) or derived from published data. Monthly or daily fuel data and daily market prices of CO 2 emissions are usually provided by energy exchanges and statistical agencies. Figure 2 on the next page shows the most important prices on a monthly basis. Because power exchange plays an important role, it is integrated in the model with fixed exchange curves for each hour of the year derived from Union for the Coordination of the Transmission of Electricity (UCTE) values and the European electricity market model PERSEUS (see Möst (2006)). The power exchange of a winter week is shown in Figure 3 on page 57. Similar profiles are given for spring, summer and fall. In the year 2004 net electricity imports in Germany amounted to TWh and exports amounted to TWh. This corresponds to approximately 9% of total electricity consumption, which amounts to 552 TWh. The daily auctions (spot market, minute reserve) are the main driver for the simulation. The entire German electricity demand is traded on the spot market. Although in reality the demand is only partly traded on the spot market, the spot price is seen as the most relevant indicator and forms the basis for prices of other contracts. Hence, the assumption of trading the complete volume on the spot market is justifiable. All participating agents submit their bids for selling and buying electricity to the spot market. The primary and secondary reserve markets take place annually. Research Paper

10 56 D. Möst and M. Genoese FIGURE 1 Structure of PowerACE (authors visualization). Certificate Distributor allocation RenAgents REGDB support GridOperator-Trader bid support LoadDB InvestmentPlanner Investment decision results Spot Market ask Supplier Contracts Household Transport Services Load profiles new plant plantdb Forward Market Industry DemandDB plant data ask Future Demand Demand growth Generator Merit Order Seller bid/ask CO 2 - Market SavingPotential Industry DB Unit Commitment, Emissions bid /ask CO 2 -Trader Legend Demand Markets Renewables Reserve Markets Utilities ask GridOperator Agents Database Data/Information flow Primary Reserve Secondary Reserve Tertiary Reserve FIGURE 2 Development of selected fuel prices and of the CO 2 certificate price. Gas and coal price ( /MWh) Coal Gas CO CO 2 price ( /t) 0 Jan 01 Jan 02 Jan 03 Jan 04 Jan 05 Jan 06 Time (months) 0 Sources: Federal Office of Economics and Export Control (2008); European Energy Exchange (2009). The Journal of Energy Markets Volume 2/Number 2, Summer 2009

11 Market power in the German wholesale electricity market 57 FIGURE 3 Power exchange in a winter week (authors visualization). 4,000 3,000 Electricity exchange (MW) 2,000 1, ,000 2,000 3,000 4, Hour The electricity supply side is simulated by generator and seller agents. Generators hold a power plant portfolio based on a detailed power plant database containing the most important technical, economic and ecological parameters of power plants (capacity, costs, availability, technology, efficiency, start-up time and fuel) and check the availability of power plants daily using uniform distributed random numbers. The sellers receive a daily updated list of available power plants based on the generators information. The electricity output of the generation capacities is offered for sale on the spot electricity market. Price bids are based on variable electricity generation costs and start-up costs. Start-up costs are derived by each agent on the basis of an estimated day-ahead price and the resulting plant dispatch. Two strategies are possible: 1) either the bid price is lowered if the plant is running and the agent assumes that the plant is not scheduled according to the price projection; or 2) start-up costs are added to the bid price during hours when the plants are expected to be switched on. Additionally, in times of scarce capacity, a mark-up based on fixed costs is included. Basically, this mark-up function can be specified freely for each agent. A possible mark-up (which was also used in the simulation runs) is shown in Figure 4 on the next page, which was derived on the basis of the peak load concept and corresponding models. The mark-up function is based on various publications in the field of Research Paper

12 58 D. Möst and M. Genoese FIGURE 4 Mark-up used in the model (see Grobbel (1999)) Fixed costs share Scarcity factor electricity prices. In peak load intervals, capacities may become scarce when all units of production are in use. If this is the case, prices can exceed the marginal costs of production. If so, an incentive is given to invest in additional capacities. Borenstein (2000) confirmed that: in the absence of market power exercised by any seller in the market, price may still exceed the marginal production costs of all facilities producing output in the market at that time. This corresponds to the concept of peak load pricing in the deterministic case, which was applied to energy markets by Boiteux (1964). According to this concept, electricity generation needs two production factors: capacity and energy. The provision of energy at any time is limited by the available capacity. A simple example with two time segments, ie, one with a peak load and one with a base load, helps to illustrate the basic idea. For both periods, the same generation capacity is available and the demand is deterministic. The price of the energy output is based on the marginal production costs, and in the peak load period, a capacity mark-up is added to cover the fixed costs (including investment) of a plant. This concept was developed in the context of regulated monopolies. Its objective is to ensure that production costs are completely covered. Later developments of the peak load pricing concept advanced the methodology by taking a stochastic demand and various generation technologies into account (see Chao (1983); Murphy et al (1982); Brown and Johnson (1969); Visscher (1973); Crew and Kleindorfer (1976); Carlton (1977)). If uncertainty is taken into account, the full investment is not only assigned to the peak load period, but is also spread over several periods, as the distinction between peak The Journal of Energy Markets Volume 2/Number 2, Summer 2009

13 Market power in the German wholesale electricity market 59 TABLE 3 Investment for hard coal and combined cycle power plants (see Bagemiehl (2002); Jopp (2008)). Hard coal Combined cycle 2001 investment ( /kw) investment ( /kw) 1, and non-peak load depends on the probability. The investment costs for the peak load plant will be earned by the sum of the probability weighted capacity payments of the corresponding periods. Concerning the peak load concept, it is important to distinguish between the short run and the long run. Long-run adjustments will lead to a long-term competitive equilibrium. It is therefore important to remove barriers for new market entrants, because this will curtail the exercise of market power. In the short run, prices may exceed marginal costs in order to set a necessary price signal for additional capacities. As the justification of scarcity rents as investment incentives is a rather delicate argument, we also compare model runs with a mark-up function with runs without the mark-up function (see Section 5.5). Mathematically, the mark-up function is defined stepwise: 8 ˆ< 0; sf <b l Mark-up D c f f i ; b i 1 6 sf 6 b i (4) ˆ: c f ; sf >b u where: sf is the scarcity factor; f i is the share of fix costs (depending on the scarcity factor, cf Figure 4 on the facing page); b l is the lower bound; b i is the bound; b u is the upper bound; and c f is the fixed costs. The fixed costs c f are based on the annuities of the values in Table 3. The increase in investment costs in past years is considered (see, for example, Jopp (2008)). The role of renewable electricity generation is growing in the German electricity sector, and it therefore has to be given due consideration. Electricity grid operators in Germany are obliged to take the feed-in of electricity from renewable energy sources whenever the renewable sources (eg, wind, photovoltaic, etc) are available Research Paper

14 60 D. Möst and M. Genoese (according to the priority rule of the Renewable Energy Act passed in 2004). In the model, the feed-in of renewable energy is represented by detailed hourly profiles that reduce the remaining system load to be covered by conventional power plants. A more detailed description of the representation of renewables in this model can be found in Sensfuß et al (2008). Electricity demand is represented by supplier agents. These supplier agents are modeled as price takers with inelastic electricity demand. The load profiles are based on demand data derived from UCTE values and the vertical grid load of the German system operators. 5 SIMULATING THE GERMAN ELECTRICITY MARKET WITH AN AGENT-BASED MODEL 5.1 General overview The described agent-based simulation model is applied to simulate the German wholesale electricity market in order to analyze the existence of market power (structural market power with the residual supply index) and the possible exercise of market power (with the Lerner Index). The years 2001, 2004, 2005 and 2006 were simulated in our analysis. These years were chosen for the following reasons: 2001 was seen as a year with well-functioning competition (see Section 3); 2004 was the last year without emissions trading; in 2005 the EU emissions trading scheme started, and this was accompanied by rising prices and windfall profits for electricity generating companies; and 2006 was chosen because it supposedly suffered from poor competition. As explained in the brief description of the model, electricity prices were derived from fundamental costs based on detailed power plant dispatches. First, the overall simulation results are validated to demonstrate the ability of the model to answer the research questions raised in this paper. In the subsequent sections, the Lerner Index and the RSI are computed and discussed. 5.2 Comparison of simulated prices and spot market prices The results of the model runs are evaluated on the basis of several indicators. The mean absolute error (MAE) and the root mean squared error (RMSE) 3 are calculated from the difference of the market price and computed marginal costs. These two indicators are commonly used for evaluating the quality of fundamental models. Furthermore, the correlation is measured between the daily EEX price curves and the 3 The MAE and the RMSE are defined as follows: v MAE D 1 nx u jx i Ox i j and RMSE D t 1 nx.x i Ox i / n n 2 id1 id1 The Journal of Energy Markets Volume 2/Number 2, Summer 2009

15 Market power in the German wholesale electricity market 61 daily model results. It has to be mentioned that the correlation is not very meaningful for fundamental models because only the relative trend of the two price curves can be compared. Very high prices above marginal costs cannot be explained with fundamental models, but these prices occur only for a few hours a year, thus we propose to filter them out.a filter is applied if real market prices are too far above fundamental prices. In this case, other effects such as lower electricity imports (eg, caused by a strike in a neighboring country), insufficient amounts of cooling water during summer periods or even mistakes made by market participants (as was probably the case in 2001, when the market was very young) can cause extremely high market prices, which the model is not able to simulate. Both filtered and original data are reported to keep results interpretable. The sorted hourly price curves (EEX and model) as well as the unsorted daily average curves are compared graphically for every year. Figure 5 on the next page shows the model results for the year 2001 in comparison with real EEX prices on (a) a sorted hourly and (b) an unsorted daily basis. Both figures show that the model is able to generate realistic electricity prices, although it does not attain some of the extreme price peaks. The main indicators are summarized in Table 4 on the next page. The average price in the simulation run is 23.59/MWh and thus is a little bit lower than the average price for 2001 ( 24.07/MWh). The standard deviation is 9.06/MWh, which is about 37% of the EEX value ( 24.68/MWh). The MAE amounts to about 2.73/MWh and the RMSE to about 20.13/MWh. In 2001 we observe some extremely high prices (ie, /MWh), which have a significant influence on the indicators. If the 46 hours above 100/MWh are filtered out, we have an almost equivalent average model price of 23.47/MWh compared with the 22.83/MWh filtered EEX average, a correlation of 0.68, an MAE of 1.64/MWh and an RMSE of 3.02/MWh. The standard deviation is now 8.89/MWh compared with 11.06/MWh for the filtered EEX data. If 0.5% of all data points are removed, a very good validation of the model is achieved. Other validated fundamental models achieve a significantly lower model quality, for example, Weber s (2004) electricity price model. In his validation, the MAE is 4.26/MWh and the RMSE is 8.91/MWh; he judges these results to be quite good for a fundamental model. Applying a filter of 50/MWh, the MAE comes to 3.48/MWh and the RMSE to 5.08/MWh. Möst (2005) achieves an MAE of 6.3/MWh and an RMSE of 8.42/MWh, admittedly for a model with a longer time horizon. As the reproduction of 2001 is quite accurate, especially in comparison with other fundamental models, the model and its quality can be considered to be very good. Because the review of relevant work analyzing market power in the German electricity market (see Section 3) confirms that prices are competitive and our model reproduces these prices very well, we keep the settings equal for the other runs. For Research Paper

16 62 D. Möst and M. Genoese FIGURE 5 Results for 2001 for both the model (PowerACE, dashed gray line) and EEX (solid black line): (a) price duration curve; (b) unsorted daily averages. 140 (a) ( /MWh) ( /MWh) ,000 4,000 6,000 8,000 Hours (b) Days TABLE 4 Key figures for 2001 for filtered and unfiltered EEX and PowerACE. EEX PowerACE EEX (filter 100) PowerACE (filter 100) price ( /MWh) Min. price ( /MWh) Max. price ( /MWh) Std. dev. ( /MWh) Correlation average MAE (sorted) ( /MWh) RMSE (sorted) ( /MWh) The Journal of Energy Markets Volume 2/Number 2, Summer 2009

17 Market power in the German wholesale electricity market 63 TABLE 5 Key figures for 2004 for unfiltered EEX and PowerACE. EEX PowerACE price ( /MWh) Min. price ( /MWh) Max. price ( /MWh) Std. dev. ( /MWh) Correlation average 0.67 MAE (sorted) ( /MWh) 1.91 RMSE (sorted) ( /MWh) 2.81 the other years, the power plant portfolio, fuel and CO 2 emission permit prices are adjusted to the market data of the corresponding years. Table 5 shows the main indicators, comparing the model results with real market data for the year The average model price is 27.31/MWh and is slightly lower than the EEX average price of 28.55/MWh. The maximum price generated by the model is significantly lower than the maximum EEX price. But very high prices, which are not explainable by fundamental data, are very rare and thus have nearly no effect on the calculated indicators. Hence, a filter is not applied for the prices in The MAE amounts to 1.91/MWh and the RMSE is 2.81/MWh; these values also indicate an excellent fit. The correlation of the two price curves comes to The high correlation and the good fit of the price duration curve as well as the daily averages can also be observed in Figure 6 on the next page. The model does not derive as many low prices as observed on the market and generally does not explain the extremely high price spikes. The daily averages (which can be seen toward the bottom right of Figure 6 on the next page) show a good match too in terms of both the occurrence of the spikes and their height. The European emissions trading scheme started in The model results for 2005 show a slight underestimation of prices. Since 2005, prices for emission allowances have had to be taken into account as a new production factor for electricity. Nevertheless, due to the different allocation rules (eg, the adaptation rule, the option rule, etc), prices for emission allowances are not being completely included in the bid offers, as shown in Genoese et al (2007). We can observe that the maximum price of the EEX is far above the model s results. Applying a filter for prices above 200/MWh, which corresponds to the elimination of 47 hours of the year (or 0.45% of 8,760 prices), significantly ameliorates all indicators. We observe a maximum price of /MWh compared with /MWh (EEX), a correlation of 0.67, an MAE of 3.56 /MWh and an RMSE of 5.72/MWh. In part (a) of Figure 7 on page 65, the price duration curves are compared. Both curves are nearly identical up to the hour 4,200, at which point the model prices dip Research Paper

18 64 D. Möst and M. Genoese FIGURE 6 Results for 2004 for both the model (PowerACE, dashed gray line) and EEX (solid black line): (a) price duration curve; (b) unsorted daily averages. 120 (a) 100 ( /MWh) ,000 4,000 6,000 8,000 Hours 50 (b) 40 ( /MWh) Days TABLE 6 Key figures for 2005 for filtered and unfiltered EEX and PowerACE. EEX PowerACE EEX (filter 200) PowerACE (filter 200) price ( /MWh) Min. price ( /MWh) Max. price ( /MWh) Std. dev. ( /MWh) Correlation average MAE ( /MWh) RMSE ( /MWh) The Journal of Energy Markets Volume 2/Number 2, Summer 2009

19 Market power in the German wholesale electricity market 65 FIGURE 7 Results for 2005, for both the model (PowerACE, dashed gray line) and EEX (solid black line): (a) price duration curve; (b) unsorted daily averages. 200 (a) 150 ( /MWh) ,000 4,000 6,000 8,000 Hours 120 (b) 100 ( /MWh) Days slightly below the market prices. A high correlation of the unsorted daily averages curves can be observed in part (b) of Figure 7, except for the two peaks in spring and summer, which cannot be explained by the model. According to Jopp (2008), investment costs for new power plants dramatically increased by 90% between 2005 and The mark-up function, which is a function of investment costs, was therefore also influenced by the increase. Thus, we assume about 50% higher costs for 2006 compared with 2001 (see Table 3 on page 59). This increase in investment costs has been confirmed in several discussions with experts from energy utilities and power plant manufacturers. The model s average prices are a little bit lower than the market prices. This is caused by some extreme price spikes (the maximum price is 2,436.63/MWh), especially in the summer. These price spikes are assumed to be caused mainly by hot temperatures and too little cooling water, along with diminishing electricity imports Research Paper

20 66 D. Möst and M. Genoese TABLE 7 Key figures for 2006 for filtered and unfiltered EEX and PowerACE. EEX PowerACE EEX (filter 200) PowerACE (filter 200) price ( /MWh) Min. price ( /MWh) Max. price ( /MWh) 2, Std. dev. ( /MWh) MAE ( /MWh) RMSE ( /MWh) Correlation is not computed because no real data on wind is available. from French nuclear power plants. Cold temperatures cause a significantly higher demand and thus a higher electricity price should be expected. Nevertheless, cold temperatures also cause higher efficiencies in nuclear and conventional plants and thus a higher availability. Thus the elevated prices were offset by the good availability of plants as well as by a good water and wind supply (see Rahn (2007)). Because detailed data on power plant dispatches (eg, about the lack of cooling water) is not available, we have to base our assumptions on generally available information. If we apply a filter of 200/MWh, 4 as we did for the model year 2005, the validation of the results summarized in Table 7 improves significantly. An average price that is about 12% lower than market prices is achieved. Part (b) of Figure 8 on the facing page shows that the model simulates the development of the prices quite well, with the exception of the abovementioned higher prices in winter and the summer price spikes. Part (a) of Figure 8 on the facing page shows that prices from hour 3,000 to hour 8,000 are slightly underestimated. At this point it should be mentioned that model quality greatly depends on the quality of the data gathered; indeed, this is the main challenge with respect to fundamental models. The data used in our model comes from different sources and was gathered with extreme care. Nevertheless, there are some inputs, like the daily availability of power plants, which are unknown and have to be estimated. In general, the uncertainty about these model parameters for example, the missing data about the non-availability of plants (eg, due to hot temperatures and low water levels) can have a significant influence on the simulated prices. 5.3 Results from using the Lerner Index to measure the exercise of market power in Germany After the model results for 2001 were validated and the results for the other years were examined, the Lerner Index was computed according to the second equality in (3). Table 8 on page 68 shows the index for the simulated years. The Lerner Index 4 Forty-three prices were removed (0.45%), which equals 0.37% of the total yearly profit obtained. The Journal of Energy Markets Volume 2/Number 2, Summer 2009

21 Market power in the German wholesale electricity market 67 FIGURE 8 Results for 2006, for both the model (PowerACE, dashed gray line) and EEX (solid black line): (a) price duration curve; (b) unsorted daily averages. 200 (a) 150 ( /MWh) ,000 4,000 6,000 8,000 Hours 120 (b) 100 ( /MWh) Days can be computed in different manners; we chose the following four possibilities to calculate the index based on marginal costs, using: the yearly average 5 price, L D. Np MC/= MC; the average of the unsorted single hours, L D 1 n nx pi MC i id1 the average of the price duration curve (sorted), L D 1 n the average of peak prices 7 from the duration curve. ; MC i nx pi MC i id1 MC i ; 6 5 The operator Nx is defined as the arithmetic mean. 6 p i <p j for i<j,mc i ; MC j for i<j. 7 The 3,150 highest prices of one year were taken as peak prices. Research Paper

22 68 D. Möst and M. Genoese TABLE 8 Lerner Index for the simulated years (%). Price Price Yearly duration duration Year average (base) peak hours filter filter filter FIGURE 9 Development of the Lerner Index. 15 Average price Price duration Price duration peak 10 (%) Year In general, we observe an increasing index over time (see Figure 9). The Lerner Index of the price duration curve for 2001 and 2004 is very low, so we cannot assume the existence of market power for this period. This also holds for the other computation methods (see Table 8). For 2005 and 2006 the Lerner Index rises, but does not exceed 8.81% in the case of the price duration curve or rise above 12.43% during the peak hours. The daily electricity demand is based on values from the UCTE grid, which is only available for selected days of the month, and on an extrapolation of the vertical grid load, which is published on a daily basis by the grid operators.as the exact daily grid The Journal of Energy Markets Volume 2/Number 2, Summer 2009

23 Market power in the German wholesale electricity market 69 load is unknown, we presume that the Lerner Index based on the price duration values is most appropriate. We have already observed that for 2006, the model prices were lower than the market prices. The computed Lerner Indexes confirm these results. For peak hours we observe slightly increased values for 2006 (12.43%) and 2005 (12.4%), with 4.24% (2004) and 1.77% (2001) as acceptable values for the other years with respect to competitive prices. Figure 9 on the facing page shows the development of the Lerner Index from 2001 to 2006 for the computation methods of yearly average, price duration and peak. As already mentioned, the index increases with time. It is difficult to decide whether the increase in the Lerner Index can be interpreted as market power. Keeping in mind the aforementioned uncertainty of some input data, like power plant availability and the weakness of the Lerner Index, the height of the index might be acceptable when prices are still competitive. It should be noted that the values for the Lerner Index (based on filtered data) are generally less than 12.5%; therefore, we cannot confirm the existence of market power. A better explanation can be found in the rise of scarce capacities in the market: the high prices are necessary to stimulate new investments before capacities become too scarce to satisfy demand. Given that several investments in new power plants have already been announced for the next few years, the investment incentive seems to be working. 5.4 Results from using the residual supply index to measure structural market power in Germany Detailed information about available power plants and the daily grid load was not available for the computation of the RSI. The PowerACE model includes the simulation of the outage of power plants for each player, so only an estimation of the RSI is possible. Table 9 on the next page shows the hours in which the RSI is below 1 and below 1.1 for the four largest electricity suppliers in Germany. Obviously, according to the definition of the RSI, structural market power exists for the years under consideration (see (2)). The capacity of the largest player is needed to satisfy the demand for more than a third of the entire year for each year considered. The results do not necessarily indicate that market power was executed, however, because structural market power is a necessary but not sufficient condition for the exertion of market power. Furthermore, additional electricity imports have to be considered (see Ockenfels (2007)). According to the UCTE, the import capacity for Germany amounts to approximately 18 GW. We consider this as additional, potentially available capacity and compute the RSI again, increasing the total available capacity by the import potential of 18 GW minus the effective model import. Table 10 on the next page shows the results. The number of hours in which the RSI is less than 1, as well as less than 1.1, is significantly lower than in the case where import potential was disregarded. However, the RSI is still less than 1.1 for more than 5% of each year for Research Paper