Market Power in the Spanish Wholesale Electricity Market.

Transcription

1 Market Power in the Spanish Wholesale Electricity Market. Aitor Ciarreta y and María Paz Espinosa z Preliminary and Incomplete January 14, 2004 Abstract In the context of the recent electricity market reforms in Europe and the US, we evaluate the performance of the Spanish pool. Our method is not based on cost estimates but rather on the di erent behavior of strategic generators as compared to the behavior of more competitive producers. Generators incentives in the Spanish pool are a ected by CTCs which have a price cap e ect. Our results indicate that, despite the price cap e ect, the two larger operators in the market are able to increase prices by a signi cant amount as compared to the situation in which each plant is run independently. JEL: L11, L13, L51 Keywords: market power, electricity market Financial support from Ministerio de Ciencia y Tecnología (BEC ), UPV (9/UPV /2001), BBVA and IVIE is gratefully acknowledged. We thank Jorge de la Cruz for his very helpful research assistance, and seminar participants at the IDEI Conference on Competition and Coordination in the Electricity Industry (Toulouse), Universidad Autónoma de Barcelona and Universidad de Alicante for their comments and suggestions. We are especially indebted to Eva Ferreira, M. Cruz Loyo and Enrique Pastor for their invaluable help. y Universidad Autónoma de Barcelona y Universidad del País Vasco. z (corresponding author) Universidad del País Vasco. Departamento de Fundamentos del Análisis Económico II. Avenida Lehendakari Aguirre 83, Bilbao, Spain. jepesalp@bs.ehu.es 1

2 1 Introduction In the context of European deregulation, the Spanish electricity market has undergone a process of reform with the objective of increasing e ciency and competition. The Spanish spot market was introduced in with rules similar to those guiding the English market at the time. Since then, centralized spot markets have been abolished in several places, including California and England and Wales. It has been argued that the problems in these electricity markets were due to market power coupled with a tight demand-supply balance (see Green, 2001). 2 Since the Spanish wholesale market shares the features of high concentration and tight demand-supply balance, it seems important to evaluate its performance in these years. In this paper we explore whether the two larger operators in the Spanish wholesale market exploit their market power and, if that is the case, to what extent market power raises price-cost margins. Our work is a rst step towards exploring the e ciency of the Spanish wholesale market. A high concentration index together with an inelastic demand suggest that rms will use their market power to set prices well above costs. However, depending on other market conditions or electricity auction rules, concentration may give rise to higher or lower margins. Wolfram (1999) found that for the British market prices were much closer to marginal cost than most theories predicted, although she also nds some evidence of strategic capacity withholding. Explanations for the restrained price levels were nancial contracts between the suppliers and their customers, 3 threat of entry and threat of regulatory intervention in the market. 4 In the industrial organization literature several methods have been used to measure market power in electricity markets. Mount (2001) associates systematic patterns of price spikes with market power use in the UK electricity market. Spear (2003) argues that horizontal market power explains price spikes in peak periods observed in the California generation market, as well as the reduction in additions to capacity. Several papers (Green (1994), von der Fehr and Harbord (1993), Borenstein, Bushnell and Wolak (2000) and Wolfram (1999), among others) have used direct measures of marginal cost to calculate price cost margins. Macatangay (2002) proposes a test of suspicious patterns of bidding behavior based on the slopes of the supply curves; he shows that suspects behave di erently from the rest and checks whether the strategies of the suspect rms a ect one another. Bushnell and Saravia (2002) measure the competitiveness of the New England electricity market by comparing equilibrium prices with a competitive benchmark: an estimate of the price that would result if no rm exerted market power. They obtain a demand-weighted markup from 4% to 1 Regulated by Act 2019 of December 26, See Fabra (2001) for an overview of the literature on electricity markets and empirical evidence. 3 See Green (1999) on contracts for di erences. 4 However, Newbury (2002) argues that many European countries lack the necessary regulatory power to mitigate generator market power. 2

3 12% depending on whether equilibrium prices include operating constraints or not. 5 Our approach is di erent from previous papers measuring the impact of market power in that we do not use cost estimates. Rather, we study the optimal behavior at the electricity auction of rms with high market share and compare it to that of small rms. We model the outcome of the pool as a supply function equilibrium. The supply function of a large operator 6 at the pool is obtained by aggregating the supply functions of each generating plant under its control. In the absence of any market power, a generating plant would bid at the pool independently of whether it belongs to a large operator or to a small rm, and thus the supply function of a larger operator would coincide with the supply curve obtained as the sum of the supply functions of similar plants under the control of small rms. Of course, in real auctions production units will take into account their e ect on other production plants under the same ownership and will respond to their incentive to restrict output and raise prices (i.e. to bid a supply curve more to the left). Larger generators are very often marginal bidders at the auction, determining the price that is paid to all plants for all units sold. This impact on equilibrium prices creates an incentive to o er supply curves which are to the left of the equivalent supply curves of small generators. Our measure of market power is based on this di erence on supply curves between larger and small operators at the pool. More precisely, to measure market power we compare the behavior at the pool of technologically similar plants, ones under the ownership of larger generators and the others under the ownership of smaller rms. We choose for the comparison plants with the same technology and similar capacity and compare their bids for the same auction (same day and same time) so that demand and cost conditions coincide. Thus, any systematic di erence in their supply functions can only be attributed to the market power of larger generators. In this paper we observe this di erent behavior in terms of supply curves at the pool and measure the impact on equilibrium prices. It is worth noting that, compared to previous works based on price-cost margin estimates, our method provides a lower bound for that margin. In other words, our competitive benchmark is a situation in which each plant is run independently (and the equilibrium price that would be determined in that case) but, as long as the number of plants is nite, each plant would bid above marginal cost. Our main ndings for the Spanish pool are that the two larger operators do exploit their market power and consistently submit supply curves which are to the left (higher prices) of the competitive benchmark. We also estimate the increase in price-cost margins for peak and o -peak hours and for each technology (hydro, thermal). These results are somewhat consistent with those 5 In the I.O. literature there is a long tradition of price-cost measurement. See for example Nevo (2001), who estimates price-cost margins in the cereal industry and separates these margins into three sources of market power: product di erentiation, multiproduct rm pricing and collusion. 6 In what follows we consider that the size of a generator is its capacity. 3

4 of Wolfram (1998) who nds evidence that in the British market the larger supplier submitted higher bids for similar plants. Besides market concentration and the electricity auction rules, there are other feautures of the market which could potentially a ect rms incentives for price setting. The market is vertically integrated, so that larger generators are also large buyers in this market. This feature might raise some doubts on the real incentives for rms to keep pool prices high. However, the nal price to consumers is regulated and, thus, it is the outcome of a game between the regulator and the rms. As long as the nal price level xed by the regulator takes into account the pool prices, as the main component of the distributors costs, there is a strong incentive to raise these prices. This question, however, deserves further analysis. 7 The paper is organized as follows. Section 2 gives a very brief description of the Spanish pool. Section 3 presents the supply function equilibrium model, where we show that in equilibrium a plant under the ownership of a larger generator has a supply function which is to the left of the supply curve of a plant under the ownership of a smaller generator; results for Cournot competition are also provided. In Section 4 we de ne a measure of the market power of a generator, based on the impact that its bidding has on the equilibrium price: if all the plants of a generator were run independently we would obtain an equilibrium price; when these plants coordinate their bids the equilibrium price is higher. This price di erence yields a measure of market power. The rest of the paper presents our empirical results for the Spanish pool. In Section 5 we describe our competitive benchmark and the procedure for measuring each rm s market power and, in Section 6, the statistical analysis. Section 7 examines the evolution of market power from 2001 to Section 8 concludes. 2 The Spanish wholesale electricity market The Spanish pool for electricity (day-ahead market) started its operations in January Two companies, Endesa (EN) and Iberdrola (IB), own the majority of generating capacity, while Unión Fenosa (UF) and Hidrocantábrico (HC) are smaller competitors; all are private companies and each owns nuclear, thermal plants and hydroelectric units. At the beginning of 2002, EN sold a small part of its capacity (Viesgo) to the Italian company ENEL, which has become the fth competitor in this market. During 2002 and 2003 there has been entry in small scale (Repsol, Gas Natural,...). The pool works as follows. Before 11:00 a.m., quali ed buyers and sellers of electricity present their o ers for the following day. Each day is divided into 24 hourly periods. Sellers in the pool present bids consisting of up to 25 di erent prices and 7 See Kühn and Machado (2003) for an analysis of vertical integration in the Spanish wholesale market. 8 After Act 54/1997 liberalizing the market was approved in November 1997 and Act 2019/1997 established the rules of the production market. 4

5 the corresponding energy quantities for each of the 24 periods and for each generating unit they own; the prices must be increasing. 9 If no restriction is included in the o er this is called a simple o er. A seller may also present a complex o er which may include indivisibility conditions, a minimum revenue condition, production capacity variation (load gradient conditions) and scheduled stop conditions. The pool administrator consolidates the sales bids for each hourly period to generate an aggregate supply curve. Quali ed buyers in the pool present o ers. 10 Purchase bids state a quantity and a price of a power block and there can be as many as 25 power purchasing blocks for the same purchasing unit, with di erent prices for each block; the prices must be decreasing. The pool administrator constructs an aggregate demand with these o ers. In a session of the daily market the pool administrator combines these offers matching demand and supply for each of the 24 hourly periods and determines the equilibrium price for each period (the system marginal price) and the amount traded. 11 This matching is called the base daily operating schedule (PBF). After the base daily operating schedule is settled, the pool administrator evaluates the technical feasibility of the assignment; if the required technical restrictions are met then the program is feasible; if not, some previously accepted o ers are eliminated and others included to obtain the provisional feasible daily schedule (PVP). This reassignment ends at 14:00. By 16:00 the nal feasible daily schedule (PVD) is obtained taking into account the ancillary services assignment procedure. There is also an intra-day market to make any necessary adjustments between demand and supply. 12 The result is called the nal hourly schedule (PHF). 3 Theory Our main interest is the extent of market power in the Spanish day-ahead market. In the presence of market concentration, most models would predict prices above marginal cost. This is the case when the spot market is modeled as supply function competition (Green and Newbery, 1992) or as Cournot competition. However, it has been pointed out that this market is better analyzed 9 According to the Electricity Market Activity Rules, p. 6, generators shall be required to submit electric power sale bids to the market operator for each of the production units they own for each and every one of the hourly scheduling periods. There is an exception to this rule when the production unit has a bilateral contract which, due to its characteristics, is excluded from the bidding system. 10 From January 1st 2003, all buyers of electricity are considered quali ed buyers. Before that date quali ed buyers were those with consumption greater or equal to 1 GWh per year. The required consumption has decreased over time from 5GWh (December 1998) to 3GWh (April 1999), to 2GWh (July 1999) and to 1 GWh (October 1999). 11 Appendix 1 describes the procedure for calculating the system marginal price when demand and supply intersect in a vertical or horizontal section of either the aggregate demand or the aggregate supply curves. 12 The intra-day market started working in April In the rst three months it had 2 sessions per day. From July 1998 it had 4 sessions per day and from September 1998 it had 5 sessions. Now it has at least 6 sessions. 5

6 as a multi-unit auction rather than as a supply function competition (Fabra, von der Fehr and Harbord, 2002, von der Fehr and Harbord, 1993, Fabra, )). The results with discrete bid functions do not tend to the supply function equilibrium results when the number of steps in the bid function tends to in nity. Thus, it has become apparent that the speci c rules of the auction become crucial for existence and characterization of equilibrium. Market power may be also present in equilibrium in the multiunit auction models even though results are more Bertrand-like. 13 García-Castro and Marín (2003) have modeled the Spanish pool (decreasing demand and short-lived bids) using auction theory and obtained equilibria in pure strategies with prices above marginal cost. Fist, we derive an index of market power (a lower bound for the Lerner index) for a supply function equilibrium model. Then we extend the index for Cournot competition and for the multiunit auction model in García-Castro and Marín (2003). Then we will measure such an index with Spanish pool data and test for market power. In this paper we will not try to distinguish empirically between models. 3.1 A supply function equilibrium model In this section we represent strategic interaction in the electricity market through a supply function equilibrium model, 14 where each generator decides a supply curve for each of the plants it owns. The supply curve of a generator is then obtained as the sum of the supply curves of all its individual plants. We analyze the equilibrium behavior of generators with di erent sizes and hence di erent market power in the pool. In our model the participants in the generation market will have di erent numbers of production units: a generator with m plants (generator 1), a generator with k plants (generator 2) and a third generator with only one plant (generator 3). We assume m k > 1. The cost function for each plant is quadratic: C(q ij ) = 1 2 cq2 ij (1) where q ij denotes electricity produced by plant j owned by generator i. The choice of a linear marginal cost is frequent because it allows the solution to the system of di erential equations to be found more easily. Green and Newbury (1992) use quadratic marginal cost, which requires numerical solution of di erential equations. 15 All plants are identical. This is obviously a simplifying assumption, useful to explain di erences in bidding behavior due to di erences in market power, 13 See for example von der Ferh and Harbor, 1993, p See Klemperer and Meyer (1989), Green and Newbury (1992), Green (1996), Baldick, Grant and Kahn (2000), and Bolle (1992). Other authors have analyzed the electricity market as Cournot competition (see Borenstein and Bushnel, 1997) or as a sealed-bid, multiple-unit, private-value auction (see Wolfram, 1999, von der Fehr and Harbord, 1993, and García-Díaz and Marín, 2003). For a further discussion of the advantages of the supply function equilibrium model over Cournot, see Baldick, Grant and Kahn (2000). 15 See also Laussel (1992). 6

7 not di erences based on cost asymmetries. Demand function is linear: D t = a t bp t + u t (2) where u t is a random error with zero mean and p t denotes price at period t. Note that the slope of the demand function b is assumed independent of time, while the intercept a t may vary over time. The electricity auction is a uniform price auction; thus, all buyers (sellers) whose o er has been accepted pay (receive) the marginal price for the electricity required (supplied) in their o er. Firms are assumed to be risk neutral and therefore they maximize their expected payo. Each plant s bid at the auction will be represented here as a continuous supply function. The problem for generator 1, with m plants, is to decide the supply curve for each plant j at each period t, q t 1j (p t), such that it maximizes the expected value of pro ts: max q t 1j (pt) mx q1j(p t t ) p t j=1 mx C(q1j(p t t )) j=1 for j = 1; :::m Since all plants owned by the same generator are identical, q t i (p t) will denote the supply curve of any plant belonging to generator i. Substituting (1) and (2) in the pro ts expression we have: max p t at bp t kq2(p t t ) q3(p t t ) m 1 q1j t (pt) 2 c X at bp t kq2(p t t ) q3(p t t ) (m 1)q1j(p t t ) 2 j=1 for j = 1; :::m The rst order conditions for generator 1 are: p t b k dqt 2(p t ) dq t 3(p t ) +mq t dp t dp 1(p t ) cmq1(p t t ) t b k dqt 2(p t ) dq t 3(p t ) (m 1) dqt 1(p t ) = 0 dp t dp t dp t We look for solutions of the form: 16 and obtain: q t i(p t ) = A i + B i p t i = 1; 2; 3 16 When the support of the demand uncertainty is unbounded the linear equilibrium is unique, but in general there will be multiple equilibria. See Klemperer and Meyer (1989) and Laussel (1992). 7

8 A 1 = A 2 = A 3 = 0 B 1 = b + kb 2 + B 3 m + cm [b + kb 2 + B 3 + (m 1)B 1 ] B 2 = b + mb 1 + B 3 k + ck [b + mb 1 + B 3 + (k 1)B 2 ] (3) B 3 = b + mb 1 + kb c [b + mb 1 + kb2] Solving the system we get the equilibrium values B 1 (b; c; m; k), B 2 (b; c; m; k), B 3 (b; c; m; k), as functions of the parameters of the model, and thus, the equilibrium supply curve for each plant. 17 Since we are assuming that b and c are constant over time, the slope of the supply function is also constant over time. The main result of this section, which will be tested later on, is the following: Proposition 1 Large generators submit plant supply curves which are to the left of the plant supply curves of small generators: B 1 (b; c; m; k) B 2 (b; c; m; k) < B 3 (b; c; m; k) The result can be checked from the expressions for B 1, B 2 and B 3 in (3). A generator with a large number of plants has to take into account the e ect of a plant s bid on the price received by its other plants. Therefore, to maximize total pro ts, each plant restricts output, that is, it o ers a lower amount at each price, or asks for a higher price for each energy volume. Since all production units have the same technology, the di erent positions of the supply curves are due only to the di erent market power of the generators. Increasing output (moving the supply curve to the right) has a negative e ect on all the other plants pro ts. A larger generator would internalize these e ects and therefore choose for each plant a supply curve with a higher slope (lower B). In fact this result is not speci c to a a supply function competition model. As we will see later, the same e ect appears under Cournot competition and also in a multiunit auction. 17 When the cost function in (1) includes a term d i q i, the supply function has a non-zero intercept. 8

9 Bids bid (plant of large generator) bid (independent plant) Mg C kwh The aggregate supply function is: S t = Bp t + " t (4) where B = (mb 1 + kb 2 + B 3 ) and " t is an error collecting random breakdowns in production, etc. Matching aggregate demand and aggregate supply (equations 2 and 4) we obtain the equilibrium price and energy traded p t = at ut "t B+b + B+b and q t = Bat B+b + But+b"t B+b. Note that q t and p t are higher in high demand periods (high a t ) and are a ected by demand and supply errors (u t and " t, respectively). 3.2 A measure of market power under supply function competition The standard measure of market power is the Lerner index (Lerner, 1934): p c p, where c is marginal cost. In this section we propose a measure of market power which is a lower bound for the Lerner index. Thus, if we nd that market power is signi cant according to our index, we can be sure that p c p is also signi cant. The measure is based on the comparison of a generator s behavior, referred to a particular production unit, to the behavior of a generator who owns only one production unit. If a plant in a larger generator were to bid the same supply curve as a plant from a generator with only one plant, then it would not be using its market power associated with size. However, from Proposition 1 we would expect a larger generator to instruct its plants to restrict output, submitting supply curves to the left. Any di erence between the two supply curves will be attributed to market power and the impact on equilibrium prices will be used to construct a measure of individual market power. More precisely, we de ne a synthetic generator with m plants as a generator which does not maximize joint pro ts for the m plants, rather it instructs each plant to present a supply curve at the pool to maximize the plant s pro ts. In other words, a synthetic generator does not internalize the e ects of its plants on each other s pro ts, i.e. it does not fully exploit its market power. 9

10 First, we construct a measure of generator 1 s market power. In our analysis of the pool equilibrium, we replace generator 1 by a synthetic generator 1. Each plant in synthetic generator 1 maximizes pro ts individually; as a result, the supply function equilibrium is given by: B s 1 = b + kb 2 + B 3 + (m 1)B1 s 1 + c [b + kb 2 + B 3 + (m 1)B1 s] B 2 = b + mb s 1 + B 3 k + ck [b + mb s 1 + B 3 + (k 1)B 2 ] (5) B 3 = b + mb s 1 + kb c [b + mb 1 + kb2] where superscript s denotes that the rm is synthetic. Solving this system we obtain B1 s = B 3 > B 2. The equality between B1 s and B 3 is not surprising: since synthetic generator 1 s plants are maximizing individual pro ts, they behave exactly as the single-plant generator 3 does. It is worth noting that the equilibrium values for B 3 and B 2 are di erent from before, since generators 2 and 3 react to the behavior of generator 1. Denote by p B (a t ) the expected equilibrium price at the pool when rms submit supply curves given by system (3) and p B 1 s (a t ) the expected equilibrium price with supply curves given by system (7), i.e. when the slope of the aggregate supply function is B 1s = (mb1 s + kb 2 + B 3 ). We de ne a measure of market power of generator 1 as: In the linear model, and MP 1 (a t ) = p B(a t ) p B 1 s (a t ) : (6) p B (a t ) p B (a t ) = a t B + b (7) p B 1 s (a t ) = a t B 1s + b In other words, we measure a rm s market power as the percentage increase in price obtained by joint pro t maximization as compared to individual plant pro t maximization. Similarly, a measure of market power for generator 2 is: (8) 10

11 MP 2 (a t ) = p B(a t ) p B 2 s (a t ) : (9) p B (a t ) where B 2s = (mb 1 + kb s 2 + B 3 ) : Market power for rm 3 is zero by de nition since B s 3 = B 3. Our measure of market power is a measure of market power with respect to minimum size (one plant): The market power of a generator with one plant is set at zero and we measure the di erential market power of larger generators. We can also measure the impact of joint market power as follows. De ne p B 1 s +2 s (a t ) as the competitive benchmark, that is the price that would be determined were each plant to behave independently at the pool. The competitive benchmark is the equilibrium price in the least concentrated market structure, given the existence of (m + k + 1) plants and no entry. When the number of plants tends to in nity then B1 s tends to 1 c, that is, each rm submits its marginal cost function at the pool. 18 This is what is usually called competitive benchmark in the relevant literature and if the number of plants is high (as is usually the case) the two de nitions will be similar. We can de ne the joint market power of generators 1 and 2 as: MP (a t ) = p B(a t ) p B 1 s +2 s (a t ) : (10) p B (a t ) where B 1s +2 s = (mb1 s + kb2 s + B 3 ), and p B 1 s +2 s (a t ) = a t B 1s +2 s + b : (11) Our measure is a lower bound for the standard index of market power, p c p, since p B 1 s +2 s (a t ) > c, and will be interpreted as such, rather than as the true measure of market power. An advantage of our procedure for measuring market power (expressions (8) and (11)) is that it shows the contribution of asymmetric rms to the price-cost margin. 3.3 An extended market power index Our measure of market power is based on the comparison of a generator s equilibrium behavior, referred to a particular production unit, to the equilibrium behavior of a generator who owns only one production unit. In the previous section this comparison is performed using the supply function equilibrium model. We can extend the analysis to account for other models of the spot market. If rms compete in the level of output, denoting by q i the output of each plant owned by generator i, we obtain in equilibrium: q 1 q 2 < q 3 (see Appendix 2). This result is equivalent to that of Proposition 1: larger generators supply lower output per plant than smaller generators and the reason behind the result is basically the same. Our measure of market power could be de ned in this 18 See Appendix 3. 11

12 case as follows: p B (a t ) is simply the Cournot equilibrium price and p B 1 s +2 s (a t ) is the price that would be determined if plants owned by rms 1 and 2 behaved as independent rms. The electricity wholesale market has also been modeled as a multiunit auction by von der Ferh and Harbord (1993). In their model demand is determined as a random variable independent of price, and marginal costs are constant. Pure-strategy equilibrium do not always exist in this model, but for the case when all generators have the same marginal cost they show there exists a unique symmetric mixed-strategy equilibrium, in which generators bid prices above marginal costs.they look at the question of what will happen to the equilibrium bids if generators are split up into smaller units, and obtain that prices tend to be lower the more fragmented the industry. Again, this result is similar to that of Proposition 1. The Spanish day-ahead market has been modeled by García-Castro and Marín (2003) as a multi-unit rst price auction. Their model di ers from von der Fehr and Harbord s (1993) in that they allow for a downward-sloping demand function (deterministic) and asymmetric rms with increasing step cost functions. In this framework, they obtain existence of equilibria in pure strategies. They model demand as deterministic on the basis that bids are short lived and there is low uncertainty. Here we present a simple example with symmetric unit cost to show how market fragmentation a ects the equilibrium prices. There is a generator with m plants (generator 1), a generator with k plants (generator 2) and a third generator with only one plant (generator 3). Each plant has unit cost c and capacity '. Denote O j (p) the amount that can be supplied by rm j at a cost below p, and i (p) pro ts of generator i. García-Castro and Marín (2003) show that in an equilibrium in pure stategies all non-marginal rms sell O j (p ), i. e. they behave as price-takers (given p they sell the quantity that maximizes their pro ts). The marginal rm however behaves as a price maker and maximizes its pro ts, given the strategy of the other rms, by setting a price above marginal cost when residual demand D(p) O j (p) is positive. When several rms are marginal, then equilibrium price equals marginal cost. An equilibrium in pure startegies always exists and in general there is multiplicity of equilibria. They show that p = max i fp : p 2 arg max p i (p) and D(p) O i (p) > 0g is always an equilibrium. In this example it is easy to see that the greater the market fragmentation, the lower is p. There is always an equilibrium in which the largest rm is marginal since marginal cost is the same for all rms and residual demand is higher for the larger rm so that this rm is able to set the highest marginal price. Thus, the smaller is the largest rm in the market, the smaller the residual demand and the lower p. With cost asymmetries we could have a smaller rm setting the marginal price p. Residual demand may be lower but since marginal cost could be higher, the price maximizing pro ts for the marginal rm could also be higher. This is not possible however for a very small rm: D(p) O i (p) would be too small 12

13 or even negative. From the previous discussion it follows that we could use market power index () in the case of a multiunit auction, rede ning p B (a t ) as the auction equilibrium price and p B 1 s +2 s (a t ) as the auction equilibrium price when the industry is fragmented and each generator owns only one production unit. However, there is an important di erence. In a multi-unit auction, all generators who are not the marginal unit bid so as to sell all capacity having marginal cost below the marginal price; 19 a consequence of this result is that only the marginal generator could have any market power. 4 Measuring market power in the Spanish pool We want to examine the e ect of each rm s market power. The reference point is the bidding behavior at the pool of a generator who did not exercise any market power. Larger generators present bids for each unit that maximize joint pro ts for the rm. At the same auction, there are small generators with units of similar characteristics. We approximate the competitive behavior for a larger generator using the bids at the same auction of small generators. The two larger generators in the Spanish wholesale market are Endesa (EN) and Iberdrola (IB). Therefore, we rst build a so-called Synthetic Endesa (EN s ) and a Synthetic Iberdrola (IB s ). Then, we compare the auction outcome to the outcome obtained after replacing the supply functions of the rms by the supply curves of the synthetic rms. More precisely, 1. First, we build the empirical supply functions of Endesa (EN), Iberdrola (IB), Unión Fenosa (U F ), Hidrocantábrico (HC) and the rest of the rms, using hourly data. Then, we aggregate all of them to get the aggregate supply function, S t, for each day and hour. Finally, we intersect the aggregate supply curve with the demand schedule, D t, and compute the equilibrium price p t ignoring technical restrictions. The result is a time series of prices p t. In the linear model, the observed price depends on demand and supply parameters as well as on demand and supply random errors (see (9)): " t p t = p B (a t ) + u t B + b : The computed prices do not take technical restrictions into account. Technical restrictions should not represent a signi cant bias for prices since they involve only a very small fraction of the total volume traded in the daily market See Theorem 1 in García-Díaz and Marín (2003). 20 In June 2002, technical restrictions a ected 1.08% of the volume in the daily market, and implied an increase on the average price in the daily market of just 0.065cEur/KWh. 13

14 2. Second, we use UF, HC s and other small generators production units to build the synthetic Endesa (EN s ) and the synthetic Iberdrola (IB s ) in the following way: For each plant under the ownership of a large generator, plant L, we choose a technologically similar plant under the ownership of a small generator, plant S, and with a capacity as close as possible. We compute a capacity coe cient dividing the capacity of plant L over the capacity of plant S: K L K S : We multiply the quantities in the bid by rm S by the coe cient K L K S, so as to get a scaled bid. We replace the bid by plant L by the scaled bid by plant S. The bid so obtained is called the synthetic bid for plant L. Repeating this procedure for all the plants under the ownership of a large generator, and aggreggating for all the rms we get the synthetic supply curves for the larger generators. Then, in the aggreggate supply we replace the original supply functions of the large rms by the synthetic ones, and obtain a new aggregate supply denoted S EN s +IB s t. Intersecting this aggregate supply S EN s +IB s t with the demand schedule D t, the result is a time series of synthetic equilibrium prices, as they would have been if Endesa and Iberdrola had followed their synthetic supply curves. This series of synthetic equilibrium prices is denoted p EN s +IB s t. In the linear model, that equilibrium price p EN s +IB s t depends on demand and supply parameters as well as on demand and supply random errors (see (13)): p EN s +IB s t = p B EN s +IB s (a t ) + u t t B EN s +IB s + b Note that the error t, with zero mean, is di erent from " t. This is so because productive plants owned by EN and IB have been replaced by similar plants owned by HC, UF and other small generators. 3. We can also repeat the procedure for each rm individually to nd the proportion of the expected price variation which is due to each rm s market power. In this case we only replace the supply curve by one of the large rms by its synthetic supply curve, keeping all the rest constant. We obtain the equilibrium price when Iberdrola s supply is replaced by its synthetic supply: p IBs t the linear model these synthetic prices are:: and the corresponding price for Endesa: p EN s t p IBs t = p B IB s (a t ) + u t t B IBs + b p EN s t = p B EN s (a t ) + u t t B EN s + b 14. In

15 where t and t are random error terms with zero mean. 4. If neither of the large operators had any market power, then B EN s +IB s = B and therefore p EN s +IB s t p t = "t t B+b, that is, the time series s pen +IB s t and p t only di er in the realization of a random term with zero mean, "t t B+b. Under the alternative hypothesis, if the large generators had market power, then B EN s +IB s < B, which implies that we should expect positive values for the di erence: p t p EN s +IB s ut " t u t t = p B (a t ) p B EN s +IB s (a t ) + t B + b B EN : s +IB s + b (12) which is the numerator of our meausre of market power MP (a t ). Our empirical test is based on that implication of the model, although it does not depend on the speci c functional form of demand and supply schedules. Under the null hypothesis p t and p EN s +IB s t will only di er in the realization of a random error, while under the alternative, p t and p EN s +IB s t will show a systematic di erence which is a function of a t. 21 We extend this analysis to individual market power of the two larger generators and check whether p EN s t and p IBs t are di erent from p t. First we test for di erences on the unconditional means of the two series. If the di erence in means is signi cant, then we have detected market power. If the means are not di erent, we could still have that the series are di erent (although they have the same mean). In that case we would perform a test for di erences between the conditional means E [p t na t ] and E hp EN s +IB s t na t i, or, in other words, we test whether the functions p B (a t ) and p B EN s +IB s (a t ) are identical or not (Ferreira and Stute, 2003, see appendix) 4.1 The data The data consists of hourly demand and supply bids for each agent and for each production and demand unit, in the day-ahead electricity wholesale market. 22 There are a total of hours, 5881 corresponding to the period May 2001 to December 2001, and 8760 hours in 2002, classi ed in peak, o -peak 1 and o -peak 2 hours (high, low and intermediate demand, respectively) See Delgado (1993), Cabus (1998), Koul and Schick (1997, 2001), Hall and Hart (1990) and Ferreira and Stute (2002). 22 We do not consider the energy traded in the intra-day market, which amounts to less than 5% of the energy traded in the day-ahead market. 23 Data are available from May Following a pool administrator classi cation, data are divided into three categories: Peak demand hours: From 16:00 to 22:00 week days (excluding holidays) in November, December, January, and February. From 9:00 to 15:00 week days in March, April, July, and October. 15

16 From these data, we have generated the following time series: the market clearing prices, the synthetic prices obtained by replacing EN s bids by its synthetic rm bids, the synthetic prices obtained by replacing IB s bids by the synthetic bids, and nally the synthetic prices obtained by replacing both EN s bids and IB s bids by their respective synthetic bids: (IB s +EN s ) prices. Since the bids from nuclear plants are never the marginal bids at the auction, it seems that plants with this technology are not being used strategically. We checked whether the synthetic supply curves coming form nuclear energy plants and the real ones are somewhat di erent. The data indicate that they are almost identical, so that in the case of nuclear plants we have kept the real bid instead of replacing it by a synthetic one. The supply bids at the auction sometimes include restrictions that may be binding. 24 When that is the case, those bids are not included in the nal assignment by the market operator, OMEL. Since these restrictions cannot be replicated for the synthetic bids we have decided to ignore them. Ignoring those restrictions sometimes causes our market clearing prices to be lower than the price made public by the market operator. Since the complex conditions on the supply bids are ignored both for the real and the synthetic plants, there is no reason to think that this procedure is introducing any bias in the measurement of market power. On the other hand, the market operator sometimes rejects demand bids at a high price because they are unfeasible given the capacity restrictions of the interconnections with the neighbor countries. In those cases, there is a rationing procedure to assign the interconnection capacity among bidders. This reduction on demand sometimes causes our market clearing price to be higher than the price published by OMEL. Again, these capacity limits are ignored both for the synthetic rms and for the real ones so that no bias is introduced. 4.2 Testing We present a test of unconditional means. The null is that market power is zero. We test this hypothesis for each of the larger rms and we also test whether joint market power is zero. Results are reported in Table 1. We run the test considering all the observations (column two), peak demand hours, o -peak 2 demand hours (intermediate demand levels), and o -peak 1 hours (low demand), for each of the hypotheses to be tested. O -peak 1 demand hours: From 0:00 to 8:00 every day of the year, plus Saturdays, Sundays, and holidays. August is also included. O -peak 2 demand hours: From 6:00 to 16:00 and from 22:00 to 00:00, week days in November, December, January, and February. From 8:00 to 9:00, and from 15:00 to 00:00, week days in March, April, July, and October. From 8:00 to 00:00 week days in May, June, and September. 24 A complex o er may include indivisibility conditions (for the rst block in the bid), a minimum revenue condition, load gradient conditions and scheduled stop conditions. 16

17 Table 1: Test of Means 2001&2002 c /kwh Base Peak Off-Peak 2 Off-Peak 1 p-p ENS (0.006) (0.023) (0.011) (0.006) p-p IBS (0.015) (0.076) (0.033) (0.012) p-p ENS+IBS (0.017) (0.081) (0.038) (0.015) Table observations (1518 peak, 4546 o -peak 2, and 8575 o -peak 1 observations). On average, the di erences between the observed prices and the synthetic prices are positive. The average market power indexes are p p p p EN s p p IBs p p EN s +IB s p = 0:03; (13) = 0:131; (14) = 0:152; (15) And using weighted average prices: p w p w p w p EN s w p w = 0:034; (16) p IBs w p w = 0:142; (17) p EN s +IB s w p w = 0:168; (18) According to these results, the two larger rms have increased prices by almost 17%. One result stands out: while IB is responsible for a 14% increase, only a 3 % can be attributed to EN. Market power is greater for IB than it is for EN, even though EN has a higher market share and higher capacity than IB. Int he appendix we also show the results for the test of conditional means. There is a di erence between the two larger rms. EN seems to be exerting a lower impact on equilibrium prices than IB. This may be either a consequence of a more restrictive bidding behavior on the part of IB, or re ect di erences in the energy mix. The theoretical model in Section 3 does not distinguish between the di erent production technologies and thus it traces market power 17

18 only to the level of capacity. However, hydro and thermal generating units have di erent technical characteristics and they may imply di erent capabilities to exploit the market power associated with size. In o -peak 1 hours, conventional thermal generation units set the marginal price in almost 60% of the auctions, while in peak hours and o -peak 2 hours hydro units set the price in around 80% of the auctions. Hydro units are very exible, they allow energy storage and quick output adjustment. For that reason these resources can be used strategically. According to Borenstein, Bushnell and Wolak (2000), a price-taking rm with hydro resources would allocate these resources to peak hours, while a rm with market power is likely to allocate more hydro resources to o -peak periods than to peak periods. 25 This greater exibility of hydro resources might be behind the greater market power of the company IB as shown in the data. IB has a 52% of its capacity in hydro generating units and 28% in thermal units, while for EN the percentages are reversed (27% and 56%, respectively). Regulation may also make a di erence for generators incentives. In Spain there are high subsidies to the use of domestic coal (di erent subsidies for each coal plant). The right to collect the subsidy is lost (or partially lost) if the nal weighted average price perceived by these plants goes above a price cap plus the subsidy (partially lost when it is above the price cap but below the price cap plus the subsidy). This implies that a rm with a high percentage of its productive capacity in coal plants has a lower incentive to increase the pool price: when the market clearing price approaches the price cap, by restricting output it does not get a higher efective price for the energy produced. 26 We have measured the market power that IB and EN would be able to exert only with one type of technology. To isolate this market power coming from the strategic use of hydro units we replace the bids from hydroelectric production plants for the synthetic bids, we compute the synthetic prices and compare these to the market clearing prices. We also repeat this procedure to isolate market power coming from the use of thermal units. Results are presented in Tables 3 for coal. Table 3. Test of means. Coal c /kwh Base Peak Off-Peak 2 Off-Peak 1 p-p ENS (0.004) (0.011) (0.006) (0.005) p-p IBS (0.001) (0.006) (0.003) (0.002) p-p ENS+IBS (0.004) (0.011) (0.006) (0.006) These negative values can be explained by the di erential subsidies to the 25 See also Bushnell (1998). 26 The price cap is around 3.6 cents/kwh, not far from average equilibrium prices. Subsidies range from 0.3 cents/kwh to 1.3 cents/kwh (April 25, 2001). 18

19 use of domestic coal. IB exerts higher market power than EN (Table 1). This may be explained by the fact that, although EN has a bigger size, most of its plants do not bene t from increasing the pool s price (they would lose the subsidies) so the rm has the same incentives for output reduction as a much smaller company. 5 Concluding Comments In this paper we have modeled the outcome of the pool as a supply function equilibrium.there has been some discussion concerning the model that best suits this type of market. Some authors have considered the pool as a continuous share auction (see Klemperer, 2001; Wang and Zender, 2002), and in these models the outcomes are prices above costs: Participants submit supply functions with slopes higher than the marginal cost function so that the residual demand for each participant is also steep and no one has incentives to undercut other competitors; in this manner, high prices may be sustained in equilibrium. However, other authors have used a discrete, multiple-unit model, where prices can be any real number but suppliers must submit a nite number of price-quantity bids; von der Fehr and Harbord (1993) have shown that there is an incentive to undercut the rival s price slightly and increase output (see also Fabra, 2001, and García-Díaz and Marín, 2003). We extend our market power index also for this model. An important issue which has been neglected here is the fact that producers may own some of the companies who buy electricity in the pool and this may change their incentives to raise prices. Actually, market share of IB and EN on the demand side is not far from their market share on generation. However, this vertical structure does not eliminate the incentives to raise prices on the wholesale market if we take into account that the nal price to consumers is regulated and its level is likely to depend on the pool prices. The outcome of the bargaining game between the regulator and the rms to set the nal price to consumers is a ected by the distributors costs, so in ating these costs may be a good idea for the vertically integrated rms. Entry provides another argument in favor of the incentives to keep the pool prices high. Although barriers to entry are high in generation, they are relatively low in the downstream market and keeping the pool price high may lower the margins of potential entrants downstream. The explicit modelling of this vertical structure and its impact on market power in the electricity market are left for future research. 27 The possibility of collusion has been ignored. It is possible that part of the market power that we measure in this paper is due to the repetition of the auction, which would allow rms to sustain outcomes which are more cooperative than the one-shot outcome. In this case the supply curve that we observe would be to the left of the one-shot supply curve predicted by the model. It is di cult to empirically distinguish between the impact of collusion and the e ect of static market power. But there are some arguments in favor of market 27 On the importance of vertical integration for market power see Kühn and Machado (2003). 19

20 power: collusion would not necessarily imply a di erence in behavior between large and small generators if collusion is market-wide (although a cartel formed by the two larger generators would give rise to such a di erence). The analysis of collusion would require further work and is left for future research. Our conclusions suggests that, apart from market concentration and demand elasticity, other factors are a ecting rms incentives to set prices. One of them is the fact that rms are earning CTC s (payments for transition to competition) and future payments are calculated so that they are lower when the annual average pool price is higher than a given amount. This rule could establish incentives not to rise prices (a price cap e ect). Nevertheless, we have identi ed signi cant market power. Other important issues omitted include capacity choice (see Castro, Marín and Siotis (2001). Finally, an interesting issue will be the analysis of the impact of an Iberian market. 20

21 6 References Bolle, F. (1992) Supply Function Equilibria and the Danger of Tacit Collusion: The Case of Spot Markets of Electricity, Energy Economics, 14(2), p Baldick, R., R. Grant and E. Kahn, 2000, Linear Supply Function Equilibrium: Generalizations, Applications, and Limitations, UCEI Working Paper PWP-078. Borenstein, S., J. B. Bushnell, and F. A. Wolak, 2002, Measuring Market Ine ciencies in California s Restructured Wholesale Electricity Market, The American Economic Review 92, 5, p Borenstein, S., J. Bushnell, and F. Wolak, 2000, Diagnosing Market Power in California s Deregulated Wholesale Electricity Market, UCEI Working Paper PWP-064. Bushnell, J., 1998, Water and Power: Hydroelectric Resources in the Era of Competition in the Western U.S. POWER Working Paper PWP-056, University of California Energy Institute. Bushnell, J and C. Saravia, 2002, An Empirical Assessment of the Competitiveness of the New England Electricity Market, University of California Energy Institute, CSEM WP-101. Cabus, P., 1998, Un test de type Kolmogorov-Smirnov dans le cadre de comparison de fonctions de règression, C.R. Acad. Sci. Paris, 327, p Castro-Rodriguez, F., P. Marín and G. Siotis, 2001, Capacity Choices in Liberalized Electricity Markets, CEPR Discussion Paper # Delgado, M. A., 1993, Testing for the equality of nonparametric regression curves, Statistics and Probability Letters 17, p Fabra, N., 2001, Market Power in Electricity Markets, Ph. D. thesis, Department of Economics, European University Institute, Florence. Fabra, N., 2003, Tacit Collusion in Repeated Auctions: Uniform versus Discriminatory, The Journal of Industrial Economics 51, p Fabra, N., von der Fehr and Harbord, 2002, Modeling Electricity Auctions, The Electricity Journal. Ferreira, E. and W. Stute, 2003, Testing for di erences between conditional means in a time series context, Journal of the American Statistical Association. García-Díaz, A. and Marín, P., 2003, Strategic Bidding in Electricity Pools with Short-Lived Bids: An Application to the Spanish Market, International Journal of Industrial Organization 21, 2, p Green, R. and D. Newbury, 1992, Competition in the British Electricity Spot Market, Journal of Political Economy 100, 5, p Green, R., 1994, Britain s Unregulated Electricity Pool, From Regulation to Competition: New Frontiers in Electricity Markets. M. Einhorn ed., Boston, Ma: Kluwer, p Green, R., 1996, Increasing Competition in the British Electricity Market, Journal of Industrial Economics 44, p Green, R., 1999, The Electricity contract market in England and Wales, The Journal of Industrial Economics 47, 1, p