Professional Documents
Culture Documents
www.elsevier.com/locate/ijepes
Abstract
A single-period auction game model for analyzing strategic behavior in pool-based electricity markets is introduced in the paper. We study
the Nash equilibrium in a pure strategy sense of such games. First an equilibrium existence lemma is proved. Equilibrium characterization
under tight capacity constraints is provided. Then it is demonstrated that an auction game does not possess a pure strategy Nash equilibrium
under a wide range of market conditions. The paper provides a characterization of equilibrium under weak capacity constraints. We apply the
introduced results to analyze market power indices presented in our earlier work and in related reports. Applications to actual market
analysis, as well as limitations of the introduced model are provided.
q 2005 Elsevier Ltd. All rights reserved.
The related work, to the best of our knowledge, is The Lagrangian function of the above optimization
presented in reference [11]. Our results distinguish those of problem is as follows:
[11] in several aspects. First, we provide equilibrium
analysis results which are computable, not just of G Z sT ,P C leT ðP K LÞ C gT ½TðP K LÞ K F
‘existence’ nature as those in [11]. Second, we explicitly
model the impact of capacity constraints on market behavior C aT ðP K PÞ
K bT P (2)
and present market power indices that can be conveniently
used in applications. where the bold Greek characters l, g, a, and b represent
Lagrangian multipliers. These Lagrangian multipliers and
the dispatch P must satisfy the so-called Kuhn-Tucker
optimality conditions.
2. Electricity auction game Exactly four alternatives of the auction problem (1) can
be seen. They are: (i) the problem is unbounded; (ii) the
Like most of the game models reported in the literature, problem is infeasible; (iii) the problem has a unique optimal
the proposed model is intended for a single-period auction solution; and, (iv) the problem has multiple optimal
game. Unit commitment is not considered in this paper. In solutions. Under very weak conditions, we will show that
this section we will first describe the auction and pricing the first two alternatives can be excluded from consideration
structure, then define the auction game. throughout this paper. We are then left with the third and
fourth alternatives.
2.1. The single-period auction in electricity markets When the auction problem (1) has a unique optimal
solution, the spot price of electricity in a zone is equal to the
Let s be the bid prices of generators and P be the marginal cost of supplying one unit of electricity to that
power dispatch vector. The power dispatch and pricing zone. Following the Envelope Theorem [7], the spot prices
solution is then obtained from the following optimization in zones are computed by differentiating with respect to load
problem [1]: as follows:
Min sT ,P (1.1) vG
P rZ Z Kle K gT T (3)
vL
S:T: eT ðP K LÞ Z 0;
TðP K LÞ% F; 0% P% P If the problem (1) has multiple optimal solutions, it implies
(1.2) that the bid prices of some units are the same and these
generators set clearing price. Under such circumstances,
where P is the high operating limit vector, e is a vector these generators are dispatched in proportion to their bid-in
with all ones, and L is the load vector. The vector F quantities. This widely accepted method is important in this
contains certain transmission limits. Eq. (1.2) describe the paper, and therefore is presented mathematically (for
feasible set of the dispatch problem. The above power illustration, in the following equation we assume that the
dispatch and pricing model is assumed to be loss-less. bid prices of two generators are the same):
In Eq. (1.2), matrix T contains the configuration data of a
transmission network that is represented by a zonal model P P
[3]. Each zone is subject to an import limit. For example, P1 Z 1 L;^ P2 Z 2 L^ (4)
P1 C P2 P 1 C P2
suppose there are three generators in a zone, the zonal
import constraint is represented as follows: where L^ is the residual load that the two generators supply
(see Fig. 1).
KP1 K P2 K P3 C L% F
The variations of this zonal model, which can accommodate Price
common transmission constraints [25,26], are being
followed in several real world markets (e.g., Australia,
L̂
Nordic Pool). Zones do not intersect with each other but can
CP
be nested, therefore T has a simple structure. For example, a
small zone consists of generator A, B, and C. This small
zone can be included in a larger zone, which consists of
generator A, B, C, and D. The information of flows between Load
two zones is not explicit in the model. In many markets Quantity
where metering is insufficient to support nodal pricing (e.g.,
Fig. 1. Residual load when two generators bid the same price (CP: Clearing
New York and New England), using the above zonal trans- ^ residual load as defined in the figure). For interpretation of the
Price; L:
mission model is the only choice for settlement purposes. references to colour in this figure legend, the reader is referred to the web
The elements of the matrix are equal to 0, 1 or K1. version of this article.
482 D. Gan et al. / Electrical Power and Energy Systems 27 (2005) 480–487
2.2. The equilibrium model If the k-th supplier’s best response sk is a single-valued
function of his rival’s strategies sKk, then the above
The following assumptions are made in this paper. relationship reduces to:
First, we assume that eT ðPK LÞR 0 and TðPK LÞ% F.
This ensures there are feasible solutions to the auction sk Z rk ðsKk Þ; k Z 1; 2; .; Nsupplier (7.2)
problem.
It is convenient to re-write (7.1) into compact matrix form as
Second, each unit has constant marginal cost, which is
s2r(s) or sZr(s). The intersection(s) of the graph of these
publicly known. This assumption is only approximately
correspondences rk($), kZ1,2,.,Nsuppliers, is Nash equili-
satisfied in the real world since the generator types and fuel
brium [7]. In the case of rk($) being single-valued functions,
costs are public information. Since the purpose of the game
this is to say, the solution of the simultaneous Eq. (7.2) is the
model is to study qualitative pricing behavior, with an
Nash equilibrium of the game. In other words, the reaction
emphasis on analyzing the impact of capacity constraints,
correspondence or function, r(s), possesses a fixed point.
this assumption is believed to be acceptable.
Having constructed the auction game model, the first
Third, all generators are subject to a price cap, s.
question concerns whether a game possesses an equilibrium.
Fourth, a generator is allowed to bid one quantity block
The following result provides a partial answer to this
only. This assumption is not critical to the results of this
question.
work. It is needed only for ease of explanation. If a
generator is allowed to bid two blocks with the same width, Lemma 1. (Existence of Equilibrium) In a two-generator
one can obtain the same results described in the paper. This market, if the payoff functions pk(sk), kZ1,2, in Eq. (6) are
is because the generator can be viewed as two generators continuous in sk, then the auction game has an equilibrium.
owned by the same supplier; as will become clear, the main
Proof The graph of payoff function pk(sk) is illustrated in
results of the paper accommodate the situation where a
Fig. 3. Notice that, as sk increases, pk(s) either jumps down
supplier owns multiple generators.
at a point, or is non-decreasing.
Fifth, there are at least two suppliers, each of which owns
By the additional hypothesis pk(sk) being continuous, the
generators with non-zero capability.
payoff function can only be non-decreasing. This indicates
Sixth, the marginal costs of generators are not identical.
that pk(s), kZ1,2, is quasi-concave in sk [7, page 933].
In subsequent work, we study the same auction game
The quasi-concavity of pk(sk) and the fact that the
under the assumption that generator marginal costs are
feasible set of (6), that is, the set fs : c% s% sg is convex
identical [33].
imply that the reaction correspondence, r(s), is convex
The payoff function of the game is assumed to be the
valued [34, page 402].
profit of the supplier. If ck represents the costs of generators
Because the set fs : c% s% sg is closed and the payoff
of the k-th supplier, the profit of the k-th supplier is:
function pk(sk) is continuous in sk, by the Theorem of
pk ðsk ; sKk Þ Z ðr K ck ÞT Pk Z ðKle K gT T K ck ÞT Pk (5) Maximum [7, page 963], the reaction correspondence, r(s),
is non-empty and upper semi-continuous.
The problem that the k-th player faces is to choose a price Apparently, r(s) maps the compact, convex set fs :
vector, sk, so as to maximize his profit, assuming that the c% s% sg into itself.
price vectors of the other suppliers sKk are given. Notice that All the conditions required by the Kakutani Fixed Point
P and r are single-valued implicit functions of s. In other Theorem [7, page 953] are satisfied, so r(s) has a fixed point.
words, if s is given, P and r can be determined from The proof is complete. ,
Eqs. (1)–(4). Let us denote this function as P($) and r($),
respectively. Now the profit maximization problem can be As an example, suppose in a two-generator system the
condition eT ðPK LÞZ 0 holds, this means that the
re-formulated as follows:
generators are all needed in order to meet the load. There
Max pk ðsk ; sKk Þ Z ½rðsk ; sKk Þ K ck T Pk ðsk ; sKk Þ (6.1) is then no discontinuity introduced by the tie situation
sk
illustrated in Fig. 1. As a result, the payoff functions pk(sk),
kZ1,2, in Eq. (6) are continuous in sk. Therefore, the
S:T: ck % sk % sk (6.2) conditions required by Lemma 1 are satisfied, and the game
possesses an equilibrium.
In the above optimization problem, sKk is viewed as a
parameter. If this parameter changes, then the optimization In general, conditions of Lemma 1 are excessively
solution changes. Therefore the k-th supplier’s best restrictive. As an example, the result does not extend to a
response, s k, is in general a multi-valued function two-supplier multi-generator market since quasi-concavity
(correspondence, or point-to-set map) of his rival’s of pk(s) would not be guaranteed.
strategies, sKk. Let rk($) denote this multi-valued function; It is apparent that a complete study of equilibrium
it follows that: existence of the introduced auction game is analytically
difficult, not to mention a complete characterization. The
sk 2rk ðsKk Þ; k Z 1; 2; :::; Nsupplier (7.1) strategy of this paper is to characterize equilibrium in
D. Gan et al. / Electrical Power and Energy Systems 27 (2005) 480–487 483
Price
Must-run Gen Load
π 'B
cB cA s sB
Load Quantity Fig. 3. Payoff function in the 2-generator example (cA,cB: cost of suppliers
A and B, respectively; sB: bid price of supplier B; s: price cap; pB: profit of
Fig. 2. Competition under tight capacity constraints. supplier B).
484 D. Gan et al. / Electrical Power and Energy Systems 27 (2005) 480–487
sB Price
A’s quasi reaction correspondence
cA
B’s reaction correspondence
Load Qualtity
set CP. Let L^ be the residual load (refer to Fig. 1). The then it is obvious that the conditions required by Proposition
profit of supplier C would be: 2 are met because (11) means the following:
X X
X P ðcl K cj ÞPj0 O ðs K cj ÞPj00i Z 0
p0 Z ðr K cj ÞP j C ðr K c1 ÞL^ 1 j2Ai0 j2Ai00
j2C;s !r
P 1 C P2
j P
One can therefore assert that the smaller j2Ai00 Pj00i is, the
The first term is the profit earned by C’s infra-marginal more likely the conditions of Proposition 2 are met thus
generators. This term could be zero. Now suppose the price the more likely the market clearing price at the equilibrium
of generator #1 is lowered to r*Kh, here h is a small is low.
positive number, then the profit of C would become: P P
The terms j2A00 Pj00 and j2Ai00 Pj00i were named as must-
X
p 00 Z ðr K h K cj ÞP j C ðr K h K c1 Þ L^ run generation in our previous work [28]. This is depicted as
j2C;sj!r the must-run ratio (MRR):
P 00 P P
After simple manipulations, we have: j2A 00 Pj j Lj K
j;A Pj
0 1 MRR Z P Z P
(12)
j2A Pj j2A Pj
X P
p 00 K p 0 Z Kh@L^ C P j A C ðr K c1 ÞL^ 2 Zonal must-run generation and zonal must-run generation
j2C;s !r
P1 C P2
j ratio can be defined likewise. A salient feature of these
Supplier C can always find an (possibly small) hO0, such indices is that they better reflect the price-controlling
that p 00 Kp 0 O0. This indicates that supplier C has an potential of individual suppliers. The conventional capacity
incentive to deviate from bidding r* for generator #1. The margin is less informative. In a market composed of only
same is true for supplier D. This proves that configuration relatively small suppliers, even if the capacity margin is
(c) is not at equilibrium. low, no supplier may have incentive to bid high prices; the
The above configurations that we have considered MRR would reflect that.
constitute all the possible configurations. This completes An alternative form of must-run-ratio is the Residual
the proof. , Supply Index (RSI). This market power index is frequently
used in California [31,32]. For supplier A, the definition of
RSI is as follows:
P P
j Pj K P
6. Applications RSI Z P j2A j (13)
j Lj
ISO New England was recently required by FERC to when MRRO0, or RSI!100%, supplier A has the ability to
develop ‘bright-line test’ bid mitigation rules to monitor control the market clearing price; when MRRZ0, or
and possibly mitigate pricing behavior when the capacity RSIO100%, supplier A has little ability to control clearing
margin is low [30]. The results presented in the paper price. We have also calculated must-run ratios using
were utilized to establish numerical indices that serve NEPOOL market operation data. The results reveal that
such purposes. the must-run ratio is a better index than the commonly used
Let us study two extreme cases in Propositions 1 and 2. market power index HHI (the abbreviation for Herfindahl-
Suppose that: Hirschmann Index).
X X Since the market data of NEPOOL is not publishable for
Pj00 Z P j (10)
some reason, let us demonstrate the superiority of MRR
j2A 00 j2A
using a simple example. First notice that the HHI of a
then it is obvious that the conditions required by Proposition market is equal to the sum of square of market share in
1 are met because (10) means the following: percentage times 10,000 [27]. Therefore if the market has
X X X only a monopoly then its HHI is equal to 10,000, if the
ðs K cj ÞPj00 Z ðs K cj ÞP j O ðcR K cj ÞP j market has two suppliers with identical market share, then
j2A 00 j2A j2A
the HHI is equal to 5000 under certain conditions. In the
P 00
One can therefore assert that the larger j2A 00 Pj is, the example, there are two identical suppliers in an electricity
more likely the conditions of Proposition 1 are met thus the market, the HHI is therefore equal to 5000, regardless of
more likely the market clearing price at the equilibrium is how tight the capacity constraint is. It is reasonable to
high. deduce that, however, this market can be rather competitive
Similarly suppose the following holds for every i, during off-peak hours and less competitive during on-peak
iZ1,2,.,m: hours. This example demonstrates that HHI fails to reveal
X 00i the impact of capacity constraints on market performance
Pj Z 0 (11) while MMR introduced in this paper serves the purpose
j2Ai00 quite well. It is worth mentioning that the empirical work
486 D. Gan et al. / Electrical Power and Energy Systems 27 (2005) 480–487
1000 7. Conclusions
900
800 It is a common perception that capacity margin has
700 significant impact on market behavior. In this paper we
600 show that must-run-ratio, a market power index recently
500
developed by the authors, is a superior index compared with
400
capacity margin index. Another important finding is that the
300
200 degree of cost asymmetry among suppliers is an important
100 market index. These findings are justified using a game-
0 theoretic approach. We are able to identify conditions under
1
11
13
15
17
19
21
23
which market-clearing price at equilibrium is high or low.
price mrr load_divide_100 We also show that an electricity market game needs not
possess a Nash equilibrium in a pure strategy sense. Under
Fig. 6. Actual Zhejiang electricity market data on July 8, 2002. For such situations, the introduced notion of quasi-equilibrium
interpretation of the references to colour in this figure legend, the reader is
offers an alternative for market studies. Despite the
referred to the web version of this article.
limitations of the suggested game model, the main results
of this work provide an analytical foundation for construct-
ing effective market power indices.
performed in California [31,32] also observed that RSI is a
better market power index compared with HHI.
The introduced results have been recently applied
in Zhejiang Provincial Electricity Market (in People’s Acknowledgements
Republic China) to predict short-term, mid/long-term
market prices. In this application, a correlation analysis The opinions described in the paper do not necessarily
was performed, the result justifies that MRR, among other reflect those of ISO New England, Inc. The authors remain
factors, is the most important factor determining market solely responsible for errors.
prices. Fig. 6 shows that the market prices are closely
related to MRR of the market. Two price forcasting models,
a neural network model for short-term price forecasting, and References
a linear regression model for mid/long term price
[1] Singh H, Hao S, Papalexopoulos A. Transmission congestion
forecasting, have thus been developed. Both of the models management in competitive electricity markets. IEEE Trans Power
utilize MRR as the main input variables. The details of this Syst 1998;13(2):672–80.
study, which are provided in reference [36], can not be [2] Wu FF, Varaiya P. Coordinated multilateral trades for electric power
presented here because of space limitation. networks: Theory and implementation. Int J Elect Power Energy Syst
The cost structure of suppliers also has an impact on 1999;21(2):75–102.
[3] Gan D, Chen Q. Locational marginal pricing: New England
market performance based on Propositions 1 and 2. A prospective, presentation summary for panel session flow-gate and
somewhat surprising finding is that a large supplier may not location based pricing approaches and their impacts, in Proceedings of
necessarily have the incentive to raise bid prices. This is IEEE PES winter meeting, Columbus, Ohio, January 2001.
because the supplier may have very efficient generators and [4] Ethier R, Zimmerman R, Mount T, Schulze W, Thomas R. A Uniform
price auction with locational price adjustments for competitive
would not risk losing higher profits. The results presented in
electricity markets. Int J Elect Power Energy Syst 1999;21(2).
Sections 3 and 5 also assert that the higher the price cap s is, [5] Richter CW, Sheble G. Genetic algorithm evolution of utility bidding
the more likely the suppliers would bid aggressively. strategies for the competitive marketplace. IEEE Trans Power Syst
No model is without its limitations and the equilibrium 1998;13(1):256–61.
model presented herein is no exception. The described [6] Zimmerman R, Thomas R, Gan D, Murillo-Sanchez C. A Web-Based
Platform for Experimental Investigation of Electric Power Auctions,
model is a simultaneous, complete information, infinite, and Decision Support Systems, 24(3and4), January 1999, p. 193–205.
discontinuous, model. Complete characterization on equili- [7] Mas-Colell A, Whinston MD, Green JR. Microeconomic theory.
brium is not provided but rather only a characterization for Oxford, UK: Oxford University Press; 1995.
two polar cases—those under tight capacity constraints and [8] Gross G, Finlay DJ. Optimal bidding strategies in competitive
electricity markets, in Proceedings of 12th power systems compu-
those under weak capacity constraints. Another limitation of
tation conference, Dresden, Germany, August 1996, p. 815–23.
the proposed model is that it is a static model, while in [9] Andersson B, Bergman L. Market structure and the price of
reality the electricity market auction is dynamic, and electricity: An Ex Ante analysis of the deregulated swedish electricity
repeated. Finally, many important market features, such as market. Energy J 1995;16(2):97–109.
transmission rights, reserves, unit characteristics, etc., were [10] Oren S. Economic inefficiency of passive transmission rights in
congested electricity systems with competitive generation. Energy J
not considered in this work. Further insight to the 1997;18(1):63–84.
characteristics of auction games could be gained by relaxing [11] von der Fehr N-HM, Harbord D. Spot market competition in the UK
these limitations. electricity industry. Econ J 1993;103:531–46.
D. Gan et al. / Electrical Power and Energy Systems 27 (2005) 480–487 487
[12] Baldick R, Grant R, Kahn E. Linear Supply Function Equilibrium: [24] Weber JD, Overbye TJ. A two-level optimization problem for analysis
Generalizations, Applications, and Limitations, Working Paper, of market bidding strategies, IEEE Power Engineering Society
University of California Energy Institute, http://www.ucei.berkeley. Summer Meeting, July 1999, p 682–687.
edu/uceiw. [25] Ma X, Sun D, Cheung K. Energy and reserve dispatch in a multi-zone
[13] Green RJ, Newbery DM. Competition in British electricity spot electricity market. IEEE Trans Power Syst 1999;14(3):913–9.
market. J Political Econ 1992;100(5):929–53. [26] Chattopadhyay D, Gan D. Market dispatch incorporating stability
[14] Bai X, Shahidehpour SM, Ramesh VC, Yu E. Transmission constraints. Int J Elect Power Energy Syst 2001;23(6):459–69.
analysis by nash game method. IEEE Trans Power Syst 1997;12(3): [27] Tirole J. The theory of industrial organization. Cambridge, MA: The
1046–52. MIT Press; 1988.
[15] Guan X, Ho Y-C, Pepyne DL. Gaming and price spikes in electric [28] Gan D, Bourcier D. Locational market power screen and congestion
power markets. IEEE Trans Power Syst 2001;16(3):402–8. management: Experience and suggestions. IEEE Trans Power Syst
[16] Bialek JW. Gaming the uniform-price spot market—quantitative 2002;17(1):180–5.
analysis, submitted to IEEE Trans On Power Systems, April 26, 2001. [29] Kelman R, Barroso LAN, Pereira MVF. Market power assessment and
mitigation in hydrothermal systems. IEEE Trans Power Syst 2001;
[17] Yu Z, Preckel PV, Nderitu G, Sparrow FT. A locational gaming model
16(3):354–9.
with CO2 emission tax and limits. Int J Elect Power Energy Syst 2001;
[30] NEPOOL Document, Market Rules and Procedures—Chapter 17,
23(6):451–7.
November 2000, available at http://www.iso-ne.com.
[18] Ilic M, Galiana F, Fink L, editors. Power systems restructuring:
[31] California ISO, Determining the Existence or Absence of Workable
Engineering and economics. Boston: Kluwer Academic Publishers;
Competition, July 1999, available at http://www.caiso.com.
1998.
[32] Sheffrin A. California electricity market crisis: Viewpoint of the
[19] Wen F, David AK. Optimal bidding strategies and modeling of market
system operator, IEEE PES Summer Meeting, Vancouver, B.C.
imperfection among competitive generators. IEEE Trans Power Syst Canada, July 2001.
2001;16(1):15–21. [33] Gan D, Shen C, price A. competition model for power and reserve
[20] Hao S. A study of bidding strategy in clearing price auctions. IEEE market auctions. Elect Power Syst Res 2004;70(3):187–93.
Trans Power Syst 2000;15(3):975–80. [34] Zangwill WI, Garcia CB. Pathways to solutions, fixed points, and
[21] Bard JF. Practical bilevel optimization.: Kluwer Academic Publish- equilibria. Englewood Cliffs, NJ: Prentice-Hall, Inc.; 1981.
ers; 1998. [35] Podczeck K. Quasi-equilibrium and equilibrium in a large production
[22] Cardell JB, Hitt CC, Hogan WW. Market power and strategic economy with differentiated commodities, available at http://ideas.
interaction in electricity networks. Recourse Energy Econ 1997;19: repec.org/p/vie/viennp/vie9811.html.
109–37. [36] Hu Z, Yu Y, Wang Z, Sun W, Gan D, Han Z. Price Prediction
[23] Hobbs BF, Metzler C, Pang J. Strategic gaming analysis for electric Using an Integrated Approach, The International Conference on
power networks: An MPEC approach. IEEE Trans Power Syst 2000; Electric Utility Deregulation, Restructuring and Power Technologies,
15(2):638–45. Hong Kong, April 2004 (PANEL SESSION).