ARTICLE IN PRESS

Energy Policy 33 (2005) 897913

Electricity market modeling trends

!
Mariano Ventosa*, Alvaro Ba!llo, Andre! s Ramos, Michel Rivier
! Tecnologica,
Instituto de Investigacion ! Universidad Pontificia Comillas, Alberto Aguilera 23, 28015 Madrid, Spain

Abstract

The trend towards competition in the electricity sector has led to efforts by the research community to develop decision and
analysis support models adapted to the new market context. This paper focuses on electricity generation market modeling. Its aim is
to help to identify, classify and characterize the somewhat confusing diversity of approaches that can be found in the technical
literature on the subject. The paper presents a survey of the most relevant publications regarding electricity market modeling,
identifying three major trends: optimization models, equilibrium models and simulation models. It introduces a classication
according to their most relevant attributes. Finally, it identies the most suitable approaches for conducting various types of
planning studies or market analysis in this new context.
r 2003 Elsevier Ltd. All rights reserved.

Keywords: Deregulated electric power systems; Power generation scheduling; Market behavior

1. Introduction research for the electrical industry has opened up.

Numerous publications give evidence of extensive effort
In the last decade, the electricity industry has by the research community to develop electricity market
experienced signicant changes towards deregulation models adapted to the new competitive context.
and competition with the aim of improving economic This paper focuses on electricity generation market
efciency. In many places, these changes have culmi- modeling. Two main technical features determine the
nated in the appearance of a wholesale electricity complexity of such models: the product electricity
market. In this new context, the actual operation of cannot be stored and its transportation requires a
the generating units no longer depends on state- or physical link (transmission lines).
utility-based centralized procedures, but rather on On the one hand, these features explain why
decentralized decisions of generation rms whose goals electricity market modeling usually requires the
are to maximize their own prots. All rms compete to representation of the underlying technical character-
provide generation services at a price set by the market, istics and limitations of the production assets. Pure
as a result of the interaction of all of them and the economic or nancial models used in other kind
demand. of activities do a poor job of explaining electrical
Therefore, electricity rms are exposed to signicantly market behavior. This paper deals specically with those
higher risks and their need for suitable decision-support models that combine a detailed representation of the
models has greatly increased. On the other hand, physical system with rational modeling of the rms
regulatory agencies also require analysis-support models behavior.
in order to monitor and supervise market behavior. On the other hand, unless a high interregional or
Traditional electrical operation models are a poor t international capacity interconnection exists or a very
to the new circumstances since market behavior, the new proactive divestiture program is prompted (and this is
driving force for any operation decision, was not true for very few countries), only a handful of rms are
modeled in. Hence, a new area of highly interesting expected to participate in each wholesale electricity
market. Proper market models, in most cases, must deal
*Corresponding author. Tel.: +34-91-542-28-00; fax: +34-91-542- with imperfectly competitive markets, which are much
31-76. more complex to represent. This paper focuses on these
E-mail address: [email protected] (M. Ventosa). kinds of models.

0301-4215/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enpol.2003.10.013
 ARTICLE IN PRESS
898 M. Ventosa et al. / Energy Policy 33 (2005) 897913

The aim of this paper is to help to identify, classify tions devoted to simulation models. Section 6 details the
and characterize the somewhat confusing diversity of proposed taxonomy for electricity market models. Section
approaches that can be found in the technical literature 7 points out the major uses of each modeling approach
on the subject. The paper presents a survey of the most and, nally, Section 8 provides some conclusions.
relevant publications regarding electricity market mod-
eling, identifying three major trends: optimization
models, equilibrium models and simulation models. 2. Electricity market modeling trends
Although there is a large number of papers devoted to
modeling the operation of deregulated power systems, in From a structural point of view, the different
this survey only a selection of the most relevant has been approaches that have been proposed in the technical
considered for brevitys sake. An original taxonomy of literature can be classied according to the scheme
these models is also introduced in order to classify them shown in Fig. 1.
according to specic attributes: degree of competition, Research developments follow three main trends:
time scope, uncertainty modeling, interperiod links, optimization models, equilibrium models and simula-
transmission constraints and market representation. tion models. Optimization models focus on the prot
These specic characteristics are helpful to understand maximization problem for one of the rms competing in
the advantages and limits of each model surveyed in this the market, while equilibrium models represent the
paper. Finally, the paper identies which approaches are overall market behavior taking into consideration
most suitable for each purpose (i.e., planning studies or competition among all participants. Simulation models
market analysis), including risk management, which is are an alternative to equilibrium models when the
an increasingly important market issue. problem under consideration is too complex to be
Four articles, Smeers (1997), Kahn (1998), Hobbs addressed within a formal equilibrium framework.
(2001) and Day et al. (2002), have already addressed the Although there are many other possible classications
classication of these approaches. The rst points out based on more specic attributes (see Section 6), the
how game theory-based models can be used to explore different mathematical structures of these three modeling
relevant aspects of the design and regulation of liberal- trends establish a clearer division. Their various purposes
ized energy markets. It also introduces the application of and scopes also imply distinctions related to market
multistage-equilibrium models in the context of invest- modeling, computational tractability and main uses.
ment in deregulated electricity markets. Kahn (1998)
surveys numerical techniques for analyzing market 2.1. Mathematical structure
power in electricity focusing on equilibrium models,
based on prot maximization of participants, which Optimization-based models are formulated as a single
assume oligopolistic competition. Two kinds of equili- optimization program in which one rm pursues its
bria are mentioned in this survey. The commonest one is maximum prot. There is a single objective function to
based on Cournot competition, where rms compete in
quantity. In contrast, in the supply function equilibrium
approach (SFE), rms compete both in quantity and
Exogenous
price. The conclusion is that Cournot is more exible Price
Optimization
and tractable, and for this reason it has attracted more
Problem for
interest. More recently, Hobbs (2001) presents a brief One Firm Demand-price
overview of the related literature, concentrating on Function
Cournot-based models. Finally, Day et al. (2002)
perform a more detailed survey of the power market
modeling literature with emphasis on equilibrium Market
Cournot
Electricity Equilibrium
models. The new survey presented in this paper does Equilibrium
not offer a signicantly different vision of the existing Market
Considering
Modeling Supply Function
electricity market modeling trends, but rather a com- All Firms
Equilibrium
plementary and unifying one. It constitutes an effort to
organize and characterize the existing proposals so as to
clarify their main contributions and shortfalls and pave Equilibrium
Models
the way toward future developments. Simulation
The paper is organized as follows. Section 2 Models Agent-based
summarizes the classication scheme used in the survey. Models
Section 3 describes the publications related to optimiza-
tion models, whereas Section 4 focuses on those related Fig. 1. Schematic representation of the electricity market modeling
to equilibrium models. Section 5 presents the publica- trends.
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 899

Single-firm
Optimization Model Equilibrium Model

Optimization Program Optimization Program Optimization Program Optimization Program

of firm f of Firm 1 of Firm f of Firm F

maximize : f
( x) maximize : 1 ( x1 ) maximize : f
( xf ) maximize : F
( xF )
subject to : h f ( x ) = 0 subject to : h1 ( x1) = 0 subject to : h f ( x f ) = 0 subject to : h F ( x F ) = 0
g (x ) 0
f
g (x ) 0
1 1
g f
(x ) 0
f
g F (x F ) 0

Supply = Demand Supply = Demand

Electricity Market Electricity Market

Fig 2. Mathematical structure of single-rm optimization models and equilibrium-based models.

be optimized subject to a set of technical and economic short-term. On the contrary, equilibrium models are
constraints. In contrast, both equilibrium and simula- more suitable to long-term planning and market power
tion-based models consider the simultaneous prot analysis since they consider all participants. The
maximization program of each rm competing in the modeling exibility of simulation models allows for a
market. Both types of models are schematically repre- wide range of purposes although there is still some
sented in Fig. 2, where Pf represents the prot of each controversy as to the appropriate uses of agent-based
rm f Af1; y; F g; xf are rm fs decision variables; and models. The major uses of existing electricity models are
hf x and gf x represent rm fs constraints. presented in more detail in Section 7.

2.2. Market modeling

3. Single-rm optimization models
Equilibrium and simulation-based models represent
market behavior considering competition among all
In this paper, approaches based on the prot
participants. On the contrary, optimization models only
maximization problem of one rm are grouped together
represent one rm. Consequently, in the latter models,
into the single-rm optimization category. These models
the market is synthesized in the representation of the
take into account relevant operational constraints of the
price clearing process, which can be modeled as
generation system owned by the rm of interest as well
exogenous to the optimization program or as dependent
as the price clearing process. According to the manner in
of the quantity supplied by the rm of interest.
which this process is represented, these models can be
classied into two types: price modeled as an exogenous
2.3. Computational tractability
variable and price modeled as a function of the demand
supplied by the rm of study.
While complex mathematical programming methods
are required to deal with equilibrium-based models,
powerful and well-known optimization algorithms 3.1. Exogenous price
bestowing a more detailed modeling capability can be
applied to solve optimization-based models. Simulation The lowest level of market modeling represents the
models provide a more exible way to address the price clearing process as exogenous to the rms
market problem than equilibrium models although, in optimization program, i.e., the system marginal price
general, they are based on assumptions that are is an input parameter for the optimization program.
particular to each study. Consequently, as the price is xed, the market revenue
price times the rms productionbecomes a linear
2.4. Major uses function of the rms production, which is the main
decision variable in this approach. In view of that,
The previously mentioned differences in mathematical traditional Linear Programming (LP) and Mixed Integer
structure, market modeling and computational tract- Linear Programming (MILP) techniques can be em-
ability provide useful information in order to identify ployed to obtain the solution of the model. Unfortu-
the major uses of each modeling trend. For example, the nately, this type of optimization model can only
better computational tractability of optimization models properly represent markets under quasi-perfect competi-
enables them to deal with difcult and detailed tion conditions because it neglects the inuence of the
problems, such as building daily bid curves in the rms decisions on the market clearing price.
 ARTICLE IN PRESS
900 M. Ventosa et al. / Energy Policy 33 (2005) 897913

These models can again be classied into two sub- Fleten approach by explicitly measuring the risk as
groups, depending on whether they use a deterministic conditional value at risk (CVaR). Similar to the models
or probabilistic price representation. proposed by Fleten and Unger, another stochastic
approach, which focuses on the solution method, is
presented in Pereira (1999). The resulting large-scale
3.1.1. Deterministic models
A good example is the model proposed in Gross and optimization program is solved using the Benders
decomposition technique, in which the entire rms
Finlay (1996).1 In this model, since the price is
maximization problem is decomposed into a nancial
considered to be exogenous, it is shown that the rms
master-problem and an operation sub-problem. While
optimization problem can be decomposed into a set of
the nancial master-problem produces nancial deci-
sub-problemsone per generatorresembling the
sions related to the purchase of nancial assets
Lagrangian Relaxation approach.2 As expected in a case
(forwards, options, futures and so forth), the operation
of perfect competition, deterministic price and convex
sub-problems decide both the dispatch of the physical
costs, the simple comparison between each generators
marginal cost and the market price decides the produc- generation system and the exercise of nancial assets
providing feedback to the nancial problem. The
tion of each generator; therefore, the best offer of each
master-problem and sub-problems are solved using LP.
generation unit consists of bidding its marginal cost.

3.2. Price as a function of the firms decisions

3.1.2. Stochastic models
The previous approach can be improved if price In contrast to the former approaches in which the
uncertainty is explicitly considered. For instance, price clearing process is assumed to be independent of
Rajamaran et al. (2001) describe and solve the self- the rms decisions, there exists another family of
commitment problem of a generation rm in the models that explicitly considers the inuence of a rms
presence of exogenous price uncertainty. The objective production on price. In the context of microeconomic
function to be maximized is the rms prot, based on theory, the behavior of one rm that pursues its
the prices of energy and reserve at the nodes where the maximum prot taking as given the demand curve and
rms units are located, which are assumed to be both the supply curve of the rest of competitors is described
exogenously determined and uncertain. Similar to the by the so-called leader-in-price model (Varian, 1992). In
Gross and Finlay approach, the authors correctly such a model the amount of electricity that the rm of
interpret that, in this setting, the scheduling problem interest is able to sell at each price is given by its
for each generating unit can be treated independently, residual-demand function.3 Electricity market models of
which signicantly simplies the process of obtaining a this type can also be classied in two sub-groups
solution, thus permitting a detailed representation of depending on whether a probabilistic representation of
each unit. The problem is solved using backward the residual-demand function is used.
Dynamic Programming and several numerical examples
illustrate the possibilities of this approach.
3.2.1. Deterministic models
A number of recent models represent the price of
The rst publication on electricity markets based on
electricity as an uncertain exogenous variable in the
the leader-in-price model is Garc!a et al. (1999). They
context of deciding the operation of the generating units
address the unit commitment4 problem of a specic rm
and at the same time adopting risk-hedging measures.
facing a linear residual-demand function. Given that the
Fleten et al. (1997, 2002) address the medium-term risk
market revenue is a quadratic function of the rms total
management problem of electricity producers that
output, in order to allow for the use of powerful MILP
participate in the Nord Pool. These rms face signicant
solvers, a piecewise linearization procedure of the
uncertainty in hydraulic inows and prices of spot
market revenue is proposed. Likewise, Ba!llo et al.
market and contract markets. Considering that prices
(2001) develop a MILP-based model focusing on the
and inows are highly correlated, they propose a
problem of one rm with signicant hydroresources.
stochastic programming model coordinating physical
The Ba!llo model is more advanced in that it allows
generation resources and hedging through the forward
non-concave market revenue functions by means of
market. They model risk aversion by means of penaliz-
ing risk through a piecewise linear target shortfall cost 3
From the point of view of one rm, its residual-demand function is
function. More recently, Unger (2002) improves the obtained by subtracting the aggregation of all competitors selling
offers from the demand-sides buy bids. The term residual-demand
1
Many later models are based on the same assumptions, thus function is also known as effective demand function.
4
leading to similar conclusions. The Unit Commitment Problem deals with the short-term schedule
2
A large-scale problem with complicating constraints is amenable of thermal units in order to supply the electricity demand in an efcient
for a dual decomposition solution strategy, commonly known as manner. In this type of model, the main decision variables are
Lagrangian Relaxation approach. generators start-ups and shut-downs.
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 901

additional binary variables. This approach is included in be a set of algebraic equations, while the mathematical
a recent monograph on new developments in unit structure of SFE models turns out to be a set of
commitment models (Hobbs et al., 2001). differential equations. As a result, most equilibrium-
based models stem from the Cournot solution concept.
3.2.2. Stochastic models The publications devoted to these models concentrate
Unlike previous approaches, Anderson and Philpott on four areas: market power analysis, hydrothermal
(2002) do not formulate the problem of optimal coordination,5 inuence of the transmission network
production but rather the problem of constructing the and risk assessment.
optimal offer curve of a generation rm. In order to
obtain the optimal shape of that offer curve, the 4.1.1. Market power analysis
uncertain behavior of both competitors and consumers Market power measurement was the earliest applica-
must be taken into account. For this reason, they tion to electricity markets of a Cournot-based model.
represent uncertainty in the residual-demand function Borenstein et al. (1995) employed this theoretical market
by a probability distribution. This approach constitutes model to analyze Californian electricity market power
an interesting starting point for the development of new instead of using the more traditional Hirschman
models that convert the offer curve into a protable risk Herndahl Index (HHI) and Lerner Index, which
hedging mechanism against short-term uncertainties in measure market shares and price-cost margins,
the marketplace. The thesis of Ba!llo (2002) advances the respectively. Later, Borenstein and Bushnell (1999)
Anderson and Philpott approach by incorporating a have extended this approach by developing an em-
detailed modeling of the generating system which pirical simulation model that calculates the Cournot
implies that offer curves of different hours are not equilibrium iteratively: the prot-maximizing output
independent. of each rm is obtained assuming that the production
of the remaining rms is xed. This is repeated for
each supplier until no rm can improve its prot.
4. Equilibrium models Although this model has been successfully applied
to the Californian market, it shows some algorithmic
Approaches which explicitly consider market equili- deciencies regarding convergence properties as
bria within a traditional mathematical programming well as a simplistic representation of the hydroelectric
framework are grouped together into the equilibrium plants operation. Finally, a collection of modelsmost
models category. As mentioned earlier, there are two of them based on Cournot competitionfor mea-
main types of equilibrium models. The commonest type suring market power in electricity can be found in
is based on Cournot competition, in which rms Bushnell et al. (1999). This paper summarizes in
compete in quantity strategies, whereas the most tabular format these models, which have been
complex type is based on SFE, where rms compete in applied to the analysis of some of the most relevant
offer curve strategies. Although both approaches differ deregulated power markets: California, New England,
in regard to the strategic variable (quantities vs. offer England and Wales, Norway, Ontario, and New
curves), both are based on the concept of Nash Zealand.
equilibriumthe market reaches equilibrium when each
rms strategy is the best response to the strategies
actually employed by its opponents. 4.1.2. Hydrothermal coordination
Apart from market power analysis, Cournot competi-
4.1. Cournot equilibrium tion has also been considered in hydrothermal models.
The rst publication on this subject is by Scott and Read
Although the theoretical support of applying Cournot (1996), in the context of New Zealands electricity
equilibrium model to electricity markets is controversial, market. Their model utilizes Dual Dynamic Program-
the economic research community tends to agree that, in ming (DDP), whereby at each stage the hydrooptimiza-
the case of imperfect competition, this is a suitable tion problem is superimposed on a Cournot market
market model. In addition, it has frequently been used equilibrium. In this dual version of the dynamic
to support market power studies. A thoughtful collec- programming algorithm, the state space is dened by
tion of essays regarding Cournot competition, which the marginal water value (value of water) instead of the
links this approach with other later modelsincluding storage level of the reservoir. Bushnell (1998) proposes a
the SFE mentioned abovecan be found in (Daughety, similar model for studying the California market. Its
1988). 5
The Hydrothermal Coordination Problem provides the optimal
Cournot equilibrium, where rms choose their opti- allocation of hydraulic and thermal generation resources for a specic
mal output, is easier to compute than SFE because the planning horizon by explicitly considering the fuel cost savings that
mathematical structure of Cournot models turns out to can be obtained due to an intelligent use of hydroreserves.
 ARTICLE IN PRESS
902 M. Ventosa et al. / Energy Policy 33 (2005) 897913

most signicant contribution is its discussion about the Complementarity Problem (LCP).9 His model includes
meaning of the rms marginal water value in a a congestion-pricing scheme for transmission in which
deregulated framework. Bushnell points out that the load ows are modeled considering both the rst and the
rms water value is related to the rms marginal second Kirchhoff laws by means of a linearized
revenue instead of the traditional systems marginal formulation. This type of electric network model is
cost. Although Bushnells analytical formulation of the known as DC model. In contrast to previous models, the
market equilibrium conditions is more elegant, the Scott VI and LCP approaches are able to cope with large
and Read model contains a more detailed representation problems. In all these models, it is assumed that the
of the physical system. Similar to the Bushnell approach, generation units of each rm are located at only one
Rivier et al. (2001) state the market equilibrium using node of the networkwhich is, obviously, a particular
the equations that express the optimal behavior of case. Unfortunately, since in the general case in which
generation companies, i.e., by means of the rms each rm is allowed to own generation units in more
optimality conditions. Unlike both the Scott and Read than one node, a pure-strategy equilibrium does not
model and the Bushnell model, the Rivier et al. (2001) exist, as it is pointed out by Neuhoff (2003).
approach takes advantage of the fact that the optimality
conditions can be directly solved due to its Mixed 4.1.4. Risk analysis
Complementarity Problem6 (MCP) structure, which Finally, because of the difculty in applying tradi-
allows for the use of special complementarity methods tional risk management techniques to electricity mar-
to solve realistically sized problems. Kelman et al. (2001) kets, only one publication has been identied that
combine the Cournot concept with the Stochastic explicitly addresses the risk management problem for
Dynamic Programming technique in order to cope with generation rms under imperfect competition condi-
hydraulic inow uncertainty problems. However, they tions. Batlle et al. (2000) present a procedure capable of
do not mention how they deal with the fact that the taking into account some risk factors, such as hydraulic
recourse function7 of the Dynamic Programming algo- inows, demand growth and fuel costs. Cournot market
rithm is non-convex in equilibrium problems. Barqu!n behavior is considered using the simulation model
et al. (2003) introduce an original approach to compute described in Otero-Novas et al. (2000), which computes
market equilibrium, by solving an equivalent minimiza- market prices under a wide range of scenarios. The
tion problem. This approach is oriented to the medium- Batlle model provides risk measures such as value-at-
term planning of large-size hydrothermal systems, risk (VaR) or prot-at-risk (PaR).
including the determination of water value and other
sensitivity results obtained as dual variables of the 4.2. Extensions of cournot equilibrium
optimization problem.
The assumption of generation companies behaving as
4.1.3. Electric power network Cournot players has been extensively used to conduct a
Congestion pricing in transmission networks is diversity of analysis concerning the medium-term out-
another area in which Cournot-based models have also come of a variety of electricity market designs. The
played a signicant role. Both Hogan (1997) and Oren possibility of formulating these models under the MCP/
(1997) formulate a spatial electricity model wherein VI framework and beneting from specic commercial
rms compete in a Cournot manner. Wei and Smeers solvers capable of tackling large-scale problems has
(1999) use a variational inequality8 (VI) approach for signicantly contributed to the popularity of this
computing the spatial market equilibrium including approach.
generation capacity expansion decisions. They model However, a number of drawbacks seem to question
the electrical network considering only power-ow the applicability of the Cournot model. The most
conservation-equations since they omit Kirchhoffs important one stems from the fact that under the
voltage law. This type of electric network model is Cournot approach, generators strategies are expressed
known as transshipment model. in the terms of quantities and not in the terms of offer
More recently, Hobbs (2001) models imperfect curves. Hence, equilibrium prices are determined only
competition among electricity producers in bilateral by the demand function being therefore highly sensitive
and POOLCO-based power markets as a Linear to demand representation and usually higher to those
observed in reality.10 This shortcoming seems to
6
The KarushKuhnTucker (KKT) optimality conditions of any
9
optimization problem can be formulated making use of a special A LCP is obtained when the KKT conditions are derived from an
mathematical structure known as Complementarity Problem. A MCP optimization problem with quadratic objective function and linear
is a mixture of equations with a Complementarity Problem. constraints.
7 10
In the Hydrothermal Coordination Problem, the recourse function In some respects, the models predicted prices are too high because
is known as the future water value. they do not take into account some of the external circumstances such
8
KKT conditions can also be formulated as a VI problem. as stranded cost payments, new entry aversion or regulatory threats.
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 903

reinforce the idea that the SFE approach is a better in this particular power system increased the relevance
alternative to represent competition in electricity mar- of these studies. Green and Newbery (1992) analyze the
kets (Rudkevich et al., 1998). Incorporating the behavior of the duopoly that characterized the E&W
conjectural variations (CV) approach described in electricity market during its rst years of operation
traditional microeconomics theory (Vives, 1999) is under the SFE approach. It is assumed that each
another way to overcome this limitation. The CV company submits a daily smooth supply function. The
approach is easy to introduce into Cournot-based demand curve faced by generation companies is
models. This approach changes the conjectures that extremely inelasticdemand-side bidding was almost
generators are expected to assume about their compe- non-existentand varies over time since in the E&W
titors strategic decisions, in terms of the possibility of Pool offers were required to be kept unchanged
future reactions (CV). Two recent publications (Garc!a- throughout the day. Interesting conclusions were
Alcalde et al., 2002; Day et al. 2002) suggest considering reached. For instance, in the case of an asymmetric
this approach in order to improve Cournot pricing in duopoly, it is shown that the large rm nds price
electricity markets. Garc!a-Alcalde et al. (2002) assume increases more protable and therefore has a greater
that rms make conjectures about their residual demand incentive to submit a steeper supply function. The small
elasticities, as in the general CV approach, whereas Day rm then faces a less elastic residual demand curve and
et al. (2002) assume that rms make conjectures about also tends to deviate from its marginal costs. This was
their rivals supply functions. In the context of electricity previously pointed out by Bolle (1992), where the large
markets, this approach is already labeled as the generation company suffers the consequences of the
Conjectured Supply Function (CSF) approach. curse of market power and indirectly causes an increase
of its rivals prots.
4.3. Supply function equilibrium
4.3.2. Electricity pricing
Klemperer and Meyer (1989) showed that, in the The possibility of obtaining reasonable medium-term
absence of uncertainty and given the competitors price estimations with the SFE approach is considerably
strategic variables (quantities or prices), each rm has attractive, particularly when conventional equilibrium
no preference between expressing its decisions in terms models based on the Cournot conjecture have proven to
of a quantity or a price, because it faces a unique be unreliable in this aspect mainly due to their strong
residual demand. On the contrary, when a rm faces a dependence on the elasticity assumed for the demand
range of possible residual demand curves, it expects, in curve. Indeed, the SFE framework does not require the
general, a bigger prot expressing its decisions in terms residual demand curve either to be elastic or to be
of a supply function that indicates the price at which it known in advance. Based on the assumption of inelastic
offers different quantities to the market. This is the SFE demand, further advances on the SFE theory have been
approach which, originally developed by Klemperer and reported which increase its applicability. Rudkevich et al.
Meyer (1989), has proven to be an extremely attractive (1998) has obtained a closed-form expression that
line of research for the analysis of equilibrium in provides the price for a SFE given a demand realization
wholesale electricity markets. under the assumption of an n-rm symmetric oligopoly
Calculating an SFE requires solving a set of with inelastic demand and uniform pricing. Conver-
differential equations, instead of the typical set of gence problems due to the numerical integration of the
algebraic equations that arises in traditional equilibrium SFE system of differential equations are thus overcome.
models, where strategic variables take the form of This approach also allows to consider stepwise marginal
quantities or prices. SFE models have thus considerable cost functions, which is more realistic than the convex
limitations concerning their numerical tractability. In and differentiable cost functions typical of previous SFE
particular, they rarely include a detailed representation models.
of the generation system under consideration. The
publications devoted to these models concentrate on
4.3.3. Linear supply function equilibrium models
four topics: market power analysis, representation of
For numerical tractability reasons, researchers have
electricity pricing, linearization of the SFE model and
recently focused on the linear SFE model, in which
evaluation of the impact of the electric power network.
demand is linear,11 marginal costs are linear or affine
and SFE can be obtained in terms of linear or afne
4.3.1. Market power analysis
supply functions. Green (1996) considers the case of an
The SFE approach was extensively used to predict the
asymmetric n-rm oligopoly with linear marginal costs
performance of the pioneering England & Wales (E&W)
Pool, whose revolutionary design did not seem to t into 11
According to Baldick (2000), the precise description would be
more conventional oligopoly models. The relatively afne demand, whereas the term linear should be restricted to
unimportant role played by the transmission network afne functions with zero intercept.
 ARTICLE IN PRESS
904 M. Ventosa et al. / Energy Policy 33 (2005) 897913

facing a linear demand curve whose slope remains system (e.g., capacity constraints) or the transmission
invariable over time. An SFE expressed in terms of network (e.g., transmission constraints). Finally, the
afne supply functions is obtained. Baldick et al. (2000) solution method used by each author and the numerical
extend previous results to the case of afne marginal cases addressed are also two relevant features. In order
cost functions and capacity constraints. Solutions for to illustrate the evolution of this line of research, Table 1
the SFE are provided in the form of piecewise afne presents a summary of the works that have been
non-decreasing supply functions. They use this method reviewed in this section.
to predict the extent to which structural changes in the In conclusion, the SFE approach presents certain
E&W electricity industry may affect wholesale electricity advantages with respect to more traditional models of
spot prices. Baldick and Hogan (2001) perform a imperfect competition. In particular, it appears to be an
comprehensive review of the SFE approach. The appropriate model to predict medium-term prices of
authors rst revisit the general SFE problem of an electricity, given that it does not rely on the demand
asymmetric n-rm oligopoly facing a linear demand function,12 as the Cournot model, but on the shape of
curve (no explicit assumption is made concerning the the equilibrium supply functions decided by the rms. In
rms marginal costs) and show the extraordinary addition to this, rms strategies do not need to be
complexity of obtaining solutions for the system of modied as demand evolves over time. Quite the
differential equations that results. In particular, they opposite, supply functions are specically conceived to
highlight the difculty of discarding infeasible solutions represent the rms behavior under a variety of demand
(e.g., equilibria with decreasing supply functions). An scenarios. This exibility, however, is accompanied by
iterative procedure to calculate feasible SFE solutions is signicant practical limitations concerning numerical
proposed and extensively used to analyze the inuence tractability. To date, only under very strong assump-
of a variety of factors such as capacity constraints, price tions have SFE problems been solved when applied to
caps, bid caps or the time horizon over which offers are real cases. Given that SFE shortcomings are well
required to remain unchanged. documented, only the main disadvantages will be cited
here. Firstly, in general, multiple SFE may exist and it is
4.3.4. Electric power network not clear which of them is more qualied to represent
In Ferrero et al. (1997), generation companies are rms strategic behavior. Secondly, except for very
assumed to offer one afne supply curve at each of the simple versions of the SFE model, existence and
nodes in which their units are located. Transaction costs uniqueness of a solution are very hard to prove. Thirdly,
are calculated based on Schweppes spot pricing theory, closed-form expressions of a solution are very rarely
including the inuence of transmission constraints. A obtained. Consequently, numerical methods are needed
nite number of offering strategies are dened for each to solve the system of differential equations, thus
generation company and an exhaustive enumeration increasing the computational requirements of this
solution process is proposed. Berry et al. (1999) use an approach. Moreover, some of this systems solutions
SFE model to predict the outcome of a given market may violate the non-decreasing constraint that supply
structure including an explicit representation of the functions must observe. This leads to ad hoc solution
transmission network. Forcing supply functions to be procedures that usually present convergence problems.
afne typically alleviates the complexity of searching for Needless to say, transmission constraints are only
a solution. Different conceptual approaches have been considered in extremely simplied versions of the SFE
adopted to obtain numerical solutions for this family of model. Nevertheless, research efforts have recently
models. In general, no existence or uniqueness condi- produced encouraging results that may ultimately
tions are derived. Hobbs et al. (2000) propose a model in increase the applicability of this approach.
which the strategy of each rm takes the form of a set of
nodal afne supply functions. The problem is structured
in two optimization levels and therefore the solution 5. Simulation models
procedure is based on Mathematical Programming with
Equilibrium Constraints (MPEC). As indicated above, equilibrium models are based on
In spite of the variety of modeling proposals, it is a formal denition of equilibrium, which is mathema-
possible to identify a number of attributes that can be tically expressed in the form of a system of algebraic
used to establish a comparison between different SFE and/or differential equations. This imposes limitations
approaches. Some of these attributes refer to the market on the representation of competition between partici-
representation adopted by each author, such as the pants. In addition, the resulting set of equations, if it has
possibility of considering asymmetric rms and the a solution, is frequently too hard to solve. The fact that
assumptions made about the shape of the marginal cost
curves, the supply functions or the demand curve. 12
In general, SFE-based approaches model the demand function as
Others attributes refer to the model of the generation inelastic, which is the most suitable hypothesis in the case of electricity.
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 905

IEEE 30-bus system

power systems are based on the operation of generation
units with complex constraints only contributes to

Four-node case
complicate the situation.

Pennsylvania

30-node case
application

E&W Pool

E&W Pool
E&W Pool

E&W Pool
Numerical
Simulation models are an alternative to equilibrium
models when the problem under consideration is too

No
complex to be addressed within a formal equilibrium
framework. Simulation models typically represent each
Transmission

agents strategic decision dynamics by a set of sequential

rules that can range from scheduling generation units to
network

constructing offer curves that include a reaction to

Yes
Yes

Yes
No

No
No

No
No
No

previous offers submitted by competitors. The great

advantage of a simulation approach lies in the exibility
Exhaustive enumeration

it provides to implement almost any kind of strategic

Closed-form expression

Closed-form expression
Numerical integration
Necessary conditions

behavior. However, this freedom also requires that the

assumptions embedded in the simulation be theoreti-
cally justied.
Heuristics

Heuristics
Heuristics
Solution
method

MPEC

5.1. Simulation models related to equilibrium models

In many cases, simulation models are closely related

constraints

to one of the families of equilibrium models. For

Capacity

example, when in a simulation model rms are assumed

to take their decisions in the form of quantities, the
Yes

Yes
Yes

Yes
Yes

Yes

Yes
No
No

authors will typically refer to the Cournot equilibrium

model in order to support the adequacy of their
Twice continuously differentiable
Twice continuously differentiable

Piecewise linear non-decreasing

approach.
Otero-Novas et al. (2000) present a simulation model
that considers the prot maximization objective of each
generation rm while accounting for the technical
Supply functions

constraints that affect thermal and hydrogenerating

Piecewise linear
Differentiable

units. The decisions taken by the generation rms are

derived with an iterative procedure. In each iteration,
given the results obtained in the previous one, every rm
Afne
Afne

Afne
Afne

modies its strategic position with a two-level decision

process. First, each rm updates its output for each
Concave
Demand

planning period by means of a prot maximization

Inelastic
Inelastic
Linear
Linear

Linear
Linear
Linear
Linear
curve

problem in which market clearing prices are held xed

and a Cournot constraint is included limiting the
companys output. Subsequently, the price at which
Quadratic
Marginal

the company offers the output of each generating unit in

Stepwise
Convex

Linear
Afne

Afne
Afne
Afne
Afne

each planning period is modied, according to a

costs

descending rule. New market clearing prices are

calculated based on these offers and on the evolution
Asymmetric

of demand, which is assumed to be inelastic.

Day and Bunn (2001) propose a simulation model,
rms

which constructs optimal supply functions, to analyze

Yes
Yes

Yes
Yes
Yes
Yes
No
No

No
A characterization of SFE models

the potential for Market Power in the E&W Pool. This

approach is similar to the SFE scheme, but it provides a
Klemperer and Meyer (1989)
Green and Newbery (1992)

Baldick and Hogan (2001)

more exible framework that enables us to consider

Rudkevich et al. (1998)

actual marginal cost data and asymmetric rms. In this

Ferrero et al. (1997)

Baldick et al. (2000)

model, each generation company assumes that its

Hobbs et al. (2000)
Green et al. (1996)

Berry et al. (1999)

competitors will keep the same supply functions that

they submitted in the previous day. Uncertainty about
the residual demand curve is due to demand variation
Table 1

Author

throughout the day. The optimization process to

construct nearly optimal supply functions is based on
 ARTICLE IN PRESS
906 M. Ventosa et al. / Energy Policy 33 (2005) 897913

an exhaustive search, rather than on the solution of a Competition

formal mathematical programming problem. The
authors compare the results of their model for a Market Model Based on the Profit
Monopoly
symmetric case with linear marginal costs to those Maximization of the Monopolist Firm
obtained under the SFE framework, which turns out to
be extraordinarily similar. Nash Nash
Leader in Equilibrium Equilibrium
Oligopoly (Cournot and (Cournot and
Price
5.2. Agent-based models SFE) Stackelberg)

Simulation provides a more exible framework to Market Model Based on the Cost
Perfect
explore the inuence that the repetitive interaction of Competition Minimization of the Whole System Time
participants exerts on the evolution of wholesale Scope
electricity markets. Static models seem to neglect the Short Term Medium Term Long Term
fact that agents base their decisions on the historic (Days) (Months) (Years)

information accumulated due to the daily operation of Fig 3. Theoretical electricity market models depending on competition
market mechanisms. In other words, agents learn from and time scope.
past experience, improve their decision-making and
adapt to changes in the environment (e.g., competitors considering more specic attributes. These character-
moves, demand variations or uncertain hydroinows). istics are useful in understanding the advantages and
This suggests that adaptive agent-based simulation limits of each model surveyed in previous sections. The
techniques can shed light on features of electricity taxonomy presented here considers the following issues:
markets that static models ignore. degree of competition, time scope of the model,
Bower and Bunn (2000) present an agent-based uncertainty modeling, interperiod links, transmission
simulation model in which generation companies are constraints, generating system representation and mar-
represented as autonomous adaptive agents that parti- ket modeling.
cipate in a repetitive daily market and search for
strategies that maximize their prot based on the results 6.1. Degree of competition
obtained in the previous session. Each company
expresses its strategic decisions by means of the prices Markets can be classied into three broad categories
at which it offers the output of its plants. Every day, according to their degree of competition: perfect
companies are assumed to pursue two main objectives: a competition, oligopoly and monopoly.
minimum rate of utilization for their generation Since microeconomic theory proves that a perfectly
portfolio and a higher prot than that of the previous competitive market can be modeled as a cost minimiza-
day. The only information available to each generation tion or net benet maximization problem, optimization-
company consists of its own prots and the hourly based models are usually the best way to model this type
output of its generating units. As usual in these models, of market. Similarly, a monopoly can be modeled by the
demand side is simply represented by a linear demand prot maximization program of the monopolistic rm
curve. This setting allows the authors to test a number of (see Fig. 3). In these models the price is derived from the
market designs relevant for the changes that have demand function. In contrast, under imperfect competi-
recently taken place in E&W wholesale electricity tion conditionsthe most common situationthe prot
market. In particular, they compare the market outcome maximization problem of each participant must be
that results under the pay-as-bid rule to that obtained solved simultaneously. In addition, as discussed in the
when uniform pricing is assumed. Additionally, they next subsection, the suitability of each oligopolistic
evaluate the inuence of allowing companies to submit model depends on the time scope of the study.
different offers for each hour, instead of keeping them
unchanged for the whole day. The conclusion is that 6.2. Time scope
daily bidding together with uniform pricing yields the
lowest prices, whereas hourly bidding under the pay-as- The time scope is a basic attribute for classifying
bid rule leads to the highest prices. electricity models since each time scope involves both
different decision variables and different modeling
approaches. For example, when long-term planning
6. Taxonomy of electricity market models studies are conducted, capacity-investment decisions are
the main decision variables while unit-commitment
In addition to the classication presented in Sections decisions are usually neglected. On the contrary, in
25, which is based on the mathematical structure of short-term scheduling studies, start-ups and shut-downs
each model, electricity market models can be categorized become signicant decision variables, while the
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 907

maximum capacity of each generator is considered to be outcome under a wide range of scenarios since prices
xed. depend on random variables such as generators forced
As previously mentioned, under imperfect competi- outages, hydraulic inows and levels of demand. More-
tion conditions, the time scope of the model denes over, in a competitive context, new sources of un-
different market modeling approaches. To be specic, in certainty must be considered due to both strategic
the case of short-term operation (one day to one week), behavior of competitors and fuel price volatility.
the experience drawn from the literature surveyed in this According to the manner in which uncertainty is
paper suggests that the best way to represent the market represented, models can be classied into probabilistic
is the leader-in-price model from microeconomics theory when the uncertain nature of random variables is
(Garc!a et al., 1999; Ba!llo et al., 2001; Anderson and incorporated using probabilistic distributionsand de-
Philpott, 2002; Ba!llo, 2002). In the leader-in-price terministicwhen only the expected value of such
model, the incumbent rm pursues its maximum prot variables is considered. Needless to say, probabilistic
taking into account its residual demand function that models result in large-scale stochastic problems that
relates the price to its energy output. The most require complex solution techniques.
controversial assumption of this theoretical model lies In regard to the representation of the stochasticity of
on the static perspective that the residual demand demand within the context of electricity markets, the
function provides about other agents. An intuitive best examples are those models based on the SFE
explanation about the suitability of this conjecture in (Fig. 4) (Green and Newbery, 1992; Bolle, 1992;
short-term models is that the shorter the time scope Rudkevich et al. 1998) since they all consider uncer-
considered, the more consistent this conjecture becomes. tainty in demand. Based on a probabilistic version of the
In the medium-term case (1 month to 1 year), the vast price-leadership model, the Ba!llo model (2002) not only
majority of the models are based on both Cournot considers the uncertainty in demand but also in
equilibrium (Scott and Read, 1996; Bushnell, 1998; competitors behavior. Finally, Fleten et al. (2002) and
Rivier et al., 2001; Kelman et al., 2001; Barqu!n et al., Unger (2002) models focus on uncertainty in prices and
2003; Otero-Novas et al., 2000) and SFE (Green and hydraulic inows under pure competition assumptions,
Newbery, 1992; Bolle, 1992; Rudkevich et al., 1998; while Kelman et al. (2001) considers a Cournot frame-
Baldick and Hogan, 2001). work.
Finally, microeconomics suggests that the Stackelberg
equilibrium may t better than other oligopolistic 6.4. Interperiod links
models with the long-term investment-decision problem
due to its sequential decision-making process. There is a The time scope considered in planning studies is
leader rm that rst decides its optimal capacity; the typically split into intervals commonly known as
follower rms then make their optimal decisions periods. In electricity generation, there are many costs
knowing the capacity of the leader rm (Varian, 1992). and decisions that, when addressed within a certain time
Up to now, there are only a few articles (Ventosa et al., scope, involve the scheduling of resources in the multiple
2002; Murphy and Smeers, 2002) devoted to represent intermediate periods. For example, long-term studies
investment in imperfect electricity markets. In both are typically oriented to derive optimal annual manage-
publications, a comparison between the Cournot equili- ment policies for hydroreserves that must consider the
brium and Stackelberg equilibrium for modeling invest- dynamic process of inows and thus take the form of a
ment decisions is conducted. One conclusion is that set of monthly or weekly operation decisions. Similarly,
although from a theoretical point of view both models short-term models must take into account the inter-
are based on different assumptions, from a practical temporal constraints implicit in thermal unit commit-
point of view there are minor differences in most results. ment decisions.
The Stackelberg model of Ventosa et al. turns out to SFE-based models do not usually consider these
have the structure of a MPEC due to the fact that there interperiod effects. In contrast, almost all the rest of
is only one leader rm. In contrast, the Stackelberg- models reviewed in this paper, such as those devoted to
based model of Murphy and Smeers has the structure of optimal offer curve construction, hydrothermal coordi-
an Equilibrium Problem with Equilibrium Constraints nation and capacity expansion problems, focus on the
(EPEC) because more that one leader rm may exist. tradeoff of scheduling resources across time.
The EPEC model is more general although it is also
more difcult to manage. 6.5. Transmission constraints

6.3. Uncertainty modeling As in the case of previous attributes, the consideration

of transmission constraints divides electricity market
One of the most common applications of electricity models into two main types: single-node models and
market models is in the eld of forecasting the market transmission network models.
 ARTICLE IN PRESS
908 M. Ventosa et al. / Energy Policy 33 (2005) 897913

Uncertainty
Intraperiod Interperiod
constraints constraints
Single
Node

Probabilistic
Transshipment
Model

DC
Model Deterministic

AC Interperiod
Model Links
Single
Node

Transshipment
Model

DC
Model
Probabilistic

AC
Model
Deterministic
Intraperiod Interperiod
Constraints Constraints

Transmission
Network

Fig. 4. Characterization of some electricity market models according to the modeling of uncertainty, transmission network and interperiod links.

The majority of models surveyed in this paper do not models such as those proposed in Fleten et al. (2002),
consider the transmission network; nevertheless, there Unger (2002) and Kelman et al. (2001) consider not only
are good examples of transmission models. In terms of the hydroenergy constraints implicit in the management
network modeling, some authors consider a transship- of water reserves but also the hydraulic inow un-
ment network that omits Kirchhoffs voltage law (Wei certainty. On the other hand, short-term models such as
and Smeers, 1999) although their model allows for inter- Garc!a et al. (1999) and Ba!llo (2002) consider in detail
temporal constraints regarding investment decisions inter-temporal constraints, such as ramp-rate limits, and
(Fig. 4). Other authors consider both of Kirchhoffs incorporate binary variables to deal with decisions such
laws (Berry et al., 1999; Hobbs et al., 2000; Hobbs, 2001) as the start-up and shut-down of thermal units.
by means of a linearized DC network whereas Ferrero In the case of equilibrium models, two of the revised
et al. (1997) use a nonlinear AC network model. From a approachesthe Otero-Novas model (2000), which
computational point of view, only two of these combines a simulation algorithm with optimization
approaches (Hobbs, 2001; Wei and Smeers, 1999) techniques, and the Rivier model (2001), which is solved
permit solving realistically sized problems. by complementarity methodsreach a degree of realism
similar to that of optimization models. Both models are
6.6. Generating system modeling able to manage realistically sized problems considering
every generation unit as independent with its particular
A high degree of realism regarding the physical constraints. Scott and Read (1996) and Bushnell (1998)
modeling of generating systems involves the representa- are considered to have an intermediate level in terms of
tion of technical limits affecting generators as well as the generation system modeling since they take into account
consideration of accurate production cost functions of independent units but they are not capable of solving
thermal units. large problems. Finally, it is very rare that SFE-based
As shown in Fig. 5, optimization-based models for models include a detailed representation of the genera-
individual rms achieve a high level of accuracy in tion system due to their numerical tractability limita-
system modeling due to the powerful LP and MILP tions.
techniques available to solve them. These models
consider in detail the relevant technical constraints 6.7. Market modeling
affecting generation units. In addition, these models
consider every individual generation unit of interest in a The last attribute considered in this taxonomy is
non-aggregated manner. For instance, medium-term related to the market model under consideration. Pure
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 909

Single-firm Imperfect
Uncertainty Exogenous Residual Market
Price Demand Equilibrium
Low
Probabilistic
Medium

High Deterministic
Market
Modeling
Low
Probabilistic

Medium

Deterministic
High

Exogenous Single-firm Imperfect

Price Residual Market
Demand Equilibrium
Generating
System Modeling
Fig. 5. Characterization of some electricity market models according to the treatment of uncertainty, generation system modeling and market
modeling.

competition-based modelsFleten et al. (2002), Unger the competitors offers and demand-side bids. Usually,
(2002), Pereira (1999)are the simplest in terms of risk management models are also based on optimization
market modeling since they consider the price clearing due to their complexity and size.
process as exogenous to the optimization problem. In contrast, when long-term planning studies are
Models based on the leader-in-price conceptGarc!a conducted, equilibrium models are more suitable be-
et al. (1999) and Ba!llo (2002)are considered to have cause the longer the time scope of the study, the lower
an intermediate level of complexity since they take into the requirement for detailed modeling capability, and
account the inuence of the rms production on prices the more signicant the response of all competitors.
by means of its residual demand function. Finally, the Therefore, the majority of models devoted to yearly
most complex market models are those based on economic planning and hydrothermal coordination are
imperfect market equilibrium as they take into account Cournot-based approaches, which provide more realism
the interaction of all participants. in the representation of physical constraints than SFE-
based approaches, that have numerical tractability
limitations.
7. Major uses As in the case of long-term studies, in market power
analysis and market design, it is also necessary to
As mentioned in Section 2, differences in mathema- consider the market outcome resulting from competition
tical structure, market modeling and computational among all participants. Consequently, equilibrium
tractability provide useful information in order to models and simulation models are the best alternative
identify the major use of each modeling trend. This to traditional anti-trust tools based on indices such as
section summarizes the experience and conclusions HirschmanHerndahl Index (HHI) and Lerner Index.
drawn from the publications referred to in Sections 3 Finally, regarding the analysis of congestion manage-
5 regarding the major uses of single-rm optimization ment in transmission networks, Cournot and SFE
models, imperfect market equilibrium models and equilibrium models are able to simultaneously consider
simulation models (see Table 2). power ow constraints and the competition of several
One-rm optimization models are able to deal with rms at each node.
difcult and detailed problems because of their better In conclusion, Table 3 summarizes the main char-
computational tractability. Good examples of such acteristics of the most signicant models referred to in
models are those related to short-term hydrothermal previous sections. The models are classied into eight
coordination and unit commitment, which require categories depending on their market model.13 Within
binary variables, and optimal offer curve construction
under uncertainty, which not only needs binary vari- 13
CSF: Conjectured Supply Function approach and CV: Conjectur-
ables but also involves a probabilistic representation of al Variations approach.
 ARTICLE IN PRESS
910 M. Ventosa et al. / Energy Policy 33 (2005) 897913

Table 2
Major uses of electricity market models

Major use One-rm optimization Simulation models Imperfect market equilibrium models
models

Risk management Fleten et al. (2002), Unger

(2002) and Pereira (1999)

Unit commitment Garc!a et al. (1999) and

Rajamaran et al. (2001)

Short-term hydrothermal Ba!llo et al. (2001)

coordination

Strategic bidding Anderson and Philpott

(2002) and Ba!llo (2002)

Market power analysis Day and Bunn (2001) Green (1996), Bolle (1992) Rudkevich et al.
(1998), Borenstein et al. (1995), Borenstein and
Bushnell (1999), Baldick et al. (2000) and
Baldick and Hogan (2001)

Market design Bower and Bunn (2000) Green (1996), Baldick et al. (2000) and Baldick
and Hogan (2001)

Yearly economic planning Otero-Novas et al. (2000) Ramos et al. (1998)

Long-term hydrothermal Scott and Read (1996), Bushnell (1998), Rivier

coordination et al. (2001), Kelman et al. (2001) and Barqu!n
et al. (2003)

Capacity expansion planning Murphy and Smeers (2002) and Ventosa et al.
(2002)

Congestion management Hogan (1997), Oren (1997), Hobbs et al. (2000),

Hobbs (2001), Wei and Smeers (1999) and Berry
et al. (1999)

each category, models are listed by year of publication. differ as to their mathematical structure, market
Other columns are related to major use, main features of representation, computational tractability and major
the model, numerical solution method,14 problem size15 uses.
of the case study and the regional market considered. In the case of single-rm optimization models,
researchers have been developing models that address
problems such as the optimization of generation
scheduling or the construction of offer curves under
8. Conclusion and future developmental trends perfect and imperfect competition conditions. At pre-
sent, they are working on two different challenges. On
This paper presents a survey of the literature on the one hand, they are tackling the cutting edge problem
electricity market models showing that there are three of converting the offer curve of a generating rm into a
main lines of development: optimization models, equili- robust risk hedging mechanism against the short-term
brium models and simulation models. These models uncertainties due to changes in demand and competitors
14
behavior. On the other hand, they are developing risk
Benders: Benders Decomposition, DP: Dynamic Programming, management models that help rms to decide their
Enumeration: Exhaustive Enumeration, EPEC: Equilibrium Program
with Equilibrium Constraints, Heuristic: Ad hoc Heuristic Algorithm,
optimal position in spot, future and over-the-counter
LCP: Linear Complementarity Problem, LP: Linear Programming, markets with an acceptable level of risk.
MCP: Mixed Complementarity Problem, MIP: Mixed Integer Models that evaluate the interaction of agents in
Programming, MPEC: Mathematical Programming with Equilibrium wholesale electricity markets have persistently stemmed
Constraints, NI: Numerical Integration, NLP: Non-Linear Program-
from the game-theory concept of equilibrium. Some of
ming, Simulation: Simulation Scenario Analysis, and VI: Variational
Inequality. these models represent the equilibrium in terms of
15
Small: less than 100 variables, Medium: between 100 and 10,000 variational inequalities or, alternatively, in the form of a
variables, and Large: more than 10,000 variables. complementarity problem, providing a framework to
Table 3
Major uses and main features of the reviewed market models

Market model Authors Year Major use Main feature Solution method Size Intended market

Perfect Gross and Finlay 1996 Generation scheduling Deterministic prices LP Large E&W
Competition and Fleten et al. 1997 Hydro and risk management Stochastic prices and inows LP Large Nord Pool
Exogenous price Pereira et al. 1999 Hydro and risk management Solution method Benders Large
Rajamaran et al. 2001 Unit commitment Price uncertainty DP Large
Unger 2002 Hydro and risk management Risk modeling LP Large Nord Pool

Leader-in-price and residual demand function Garc!a et al. 1999 Unit commitment Thermal modeling MIP Large Spain
Ba!llo et al. 2001 Short-term hydrothermal coordination Non-convex prot MIP Large Spain
Anderson and Philpott 2002 Offer curve construction Stochastic demand function NLP Small New Zealand
Ba!llo et al. 2002 Offer curve construction Practical approach MIP Large Spain

Supply function equilibrium Green and Newberry 1992 Market power analysis Symmetric rms NI Small E&W

M. Ventosa et al. / Energy Policy 33 (2005) 897913

Bolle 1992 Market power analysis Symmetric rms NI Small E&W
Rudkevich et al. 1998 Market power analysis Closed-form solution Analytic Small Pennsylvania

ARTICLE IN PRESS
Linear supply function equilibrium Green 1996 Market design Closed-form solution Analytic Small E&W
Ferrero et al. 1997 Congestion management AC Network Model Enumeration Small
Berry et al. 1999 Congestion management DC Network Model Heuristic Small
Hobbs et al. 2000 Congestion management DC Network Model MPEC Medium
Baldick et al. 2000 Market power analysis Piecewise linear SFE Heuristic Small E&W
Baldick et al. 2001 Market design Non-decreasing SFE Heuristic Medium E&W
Day and Bunn 2001 Market power analysis Asymmetric rms Enumeration Medium E&W

Cournot equilibrium Scott and Read 1996 Hydrothermal coordination Hydro-interperiod links DP Medium New Zealand
Bushnell 1998 Hydrothermal coordination Analytic modeling DP Medium California
Borenstein and Bushnell 1999 Market power analysis Radial congestion Heuristic Medium California
Batlle et al. 2000 Risk analysis Stochastic prices and inows Simulation Large Spain
Otero-Novas et al. 2000 Yearly economic planning Agents behavior Heuristic Large Spain
Kelman et al. 2000 Long-term hydrothermal coordination Stochastic inows DP Large Brazil
Rivier et al. 2001 Hydrothermal coordination Hydrothermal modeling MCP Large Spain
Barqu!n et al. 2003 Hydrothermal coordination Stochastic inows NLP Large

Stackelberg Ventosa et al. 2002 Capacity expansion planning Investment decisions MPEC Medium
Murphy and Smeers 2002 Capacity expansion planning Investment decisions EPEC Medium

Spatial Cournot Hogan 1997 Congestion management Network constraints NLP Small
Oren 1997 Congestion management Network constraints Analytic Small
Wei and Smeers 1999 Congestion management Transshipment model VI Large Europe
Hobbs 2001 Congestion management DC power ow LCP Large

CV Garc!a-Alcalde et al. 2002 Price forecasting Fitting procedure LCP Large Spain
CSF Day et al. 2002 Congestion management DC power ow LCP Large E&W
Agent-based Bower and Bunn 2000 Market design Learning procedure Heuristic Medium E&W

911
 ARTICLE IN PRESS
912 M. Ventosa et al. / Energy Policy 33 (2005) 897913

analyze realistic cases that include a detailed representa- non-decreasing constraints, and function space iterations. Program
tion of the generation system and the transmission on Workable Energy Regulation (POWER) PWP-089. University
network. This line of research has also provided of California Energy Institute, Berkeley, CA.
Baldick, R., Grant, R., Kahn, E., 2000. Linear supply function
theoretical results relative to the design of electricity equilibrium: generalizations, application and limitations, Program
markets or to the medium-term operation of hydro- on Workable Energy Regulation (POWER) PWP-078. University
thermal systems in the new regulatory framework. As in of California Energy Institute, Berkeley, CA.
the case of optimization models, the research commu- Barqu!n, J.,Centeno, E., Reneses, J., 2003. Medium-term generation
nity is now trying to develop a new generation of programming in competitive environments: a new optimization
approach for market equilibrium computing. IEE Proceedings,
equilibrium models capable of taking risk management
Generation, Transmission and Distribution (in press).
decisions under imperfect competition. Batlle, C., Otero-Novas, I., Alba, J.J., Meseguer, C., Barqu!n, J., 2000.
On the subject of market representation, there are A model based in numerical simulation techniques as a tool for
recent publications devoted to the improvement of decision-making and risk management in a wholesale electricity
existing Cournot-based models. They propose the CV market. Part I: general structure and scenario generators.
approach to overcome the high sensitivity of the price- Proceedings PMAPS00 Conference, Madeira.
Berry, C.A., Hobbs, B.F., Meroney, W.A., ONeill, R.P., Stewart Jr.,
clearing process with respect to demand representation W.R., 1999. Understanding how market power can arise in network
typical of such models. Obviously, there are still competition: a game theoretic approach. Utilities Policy 8, 139158.
questions to be resolved. For instance, even when the Bolle, F., 1992. Supply function equilibria and the danger of tacit
simple Cournot conjecture is assumed, pure strategy collusion. Energy Economics 14 (2), 94102.
solutions may not exist if there are transmission Borenstein, S., Bushnell, J., 1999. An empirical analysis of the
potential for market power in Californias electricity industry.
constraints. Another example is that non-decreasing
Journal of Industrial Economics 47 (3), 285323.
supply functions may be unstable when generating Borenstein, S., Bushnell, J., Kahn, E., Stoft, S., 1995. Market power in
capacity constraints are considered. California electricity markets. Utilities Policy 5 (3/4), 219236.
The contribution of simulation models has been Bower, J., Bunn, D., 2000. Model-based comparisons of pool and
signicant as well, on account of their exibility to bilateral markets for electricity. Energy Journal 21 (3), 129.
Bushnell, J., 1998. Water and power: hydroelectric resources in the era
incorporate more complex assumptions than those
of competition in the Western US. Program on Workable Energy
allowed by formal equilibrium models. Simulation Regulation (POWER) PWP-056. University of California Energy
models can explore the inuence that the repetitive Institute, Berkeley, CA.
interaction of participants exerts on the evolution of Bushnell, J., Day, C., Duckworth, M., Green, R., Halseth, A., Grant
wholesale electricity markets. In these models, agents Read, E., Scott Rogers, J., Rudkevich, A., Scott, T., Smeers, Y.,
learn from past experience, improve their decision- Huntington, H., 1999. An international comparison of models for
measuring market power in electricity. Energy Modeling Forum,
making and adapt to changes in the environment. This
Stanford University. Available from World Wide Web: (http://
suggests that adaptive agent-based simulation techni- www.stanford.edu/group/EMF/research/doc/EMF17.1.PDF).
ques can shed light on certain features of electricity Daughety, A.F., 1988. Cournot Oligopoly. Cambridge University
markets ignored by static models and therefore these Press, Cambridge.
techniques will be helpful in the analysis of new Day, C., Bunn, D., 2001. Divestiture of generation assets in the
electricity pool of England and wales: a computational approach to
regulatory measures and market rules.
analyzing market power. Journal of Regulatory Economics 19 (2),
As a concluding remark, it should be pointed out that 123141.
the impressive advances registered in this research eld Day, C.J., Hobbs, B.F., Pang, J.-S., 2002. Oligopolistic competition in
underscore how much interest this matter has drawn power networks: a conjectured supply function approach. IEEE
during the last decade. Transactions on Power Systems 17 (3), 597607.
Ferrero, R., Shahidehpour, S., Ramesh, V., 1997. Transaction analysis
in deregulated power systems using game theory. IEEE Transac-
tions on Power Systems 12, 13401347.
References Fleten, S.E., Wallace, S.W., Ziemba, W.T., 1997. Portfolio manage-
ment in a deregulated hydropower based electricity market.
Anderson, E.J., Philpott, A.B., 2002. Optimal offer curve construction Proceedings Hydropower97, Norway, July.
in electricity markets. Mathematics of Operations Research 27 (1), Fleten, S.E., Ziemba, W.T., Wallace, S.W., 2002. Hedging electricity
82100. portfolios via stochastic programming. In: Greengard, C., Ruszc-
Ba!llo, A., 2002. A methodology to develop optimal schedules and zynski, A. (Eds.), Decision Making Under Uncertainty: Energy and
offering strategies for a generation company operating in a short- Power 128 of IMA Volumes on Mathematics and Its Applications.
term electricity market. Ph.D. Dissertation, ICAI, Universidad Springer, New York, pp. 7193.
Ponticia Comillas, Madrid. Garc!a, J., Rom!an, J., Barqu!n, J., Gonz!alez, A., 1999. Strategic
Ba!llo, A., Ventosa, M., Ramos, A., Rivier, M., Canseco, A., 2001. bidding in deregulated power systems. Proceedings 13th PSCC
Strategic unit commitment for generation companies in deregulated Conference, Norway, July.
electricity markets. In: Hobbs, B., Rothkopf, M., ONeil, R., Chao, Garc!a-Alcalde, A., Ventosa, M., Rivier, M., Ramos, A., Relano, G.,
H. (Eds.), The Next Generation of Unit Commitment Models. 2002. Fitting electricity market models. A conjectural variations
Kluwer Academic Publishers, Boston, pp. 227248. approach. Proceedings of the 14th PSCC Conference, Seville. July.
Baldick, R., Hogan, W.W., 2001. Capacity Constrained Supply Green, R.J., 1996. Increasing competition in the British electricity spot
Function Equilibrium Models of electricity markets: stability, market. Journal of Industrial Economics (44), 205216.
 ARTICLE IN PRESS
M. Ventosa et al. / Energy Policy 33 (2005) 897913 913

Green, R.J., Newbery, D.M., 1992. Competition in the British Pereira, M.V., 1999. Methods and tools for contracts in a competitive
electricity spot market. Journal of Political Economy 100 (5), framework. Working Paper CICRE ! Task Force 30-05-09.
929953. Rajamaran, R., Kirsch, L., Alvarado, F., Clark, C., 2001. Optimal self-
Gross, G., Finlay, D.J., 1996. Optimal bidding strategies in commitment under uncertain energy and reserve prices. In: Hobbs,
competitive electricity markets. Proceedings of the 12th PSCC B.F., Rothkopf, M.H., ONeill, R.P., Chao, H.-P. (Eds.), The Next
Conference, Germany, July. Generation of Electric Power Unit Commitment Models, Interna-
Hobbs, B.F., 2001. Linear complementarity models of NashCournot tional Series in Operations Research & Management Science.
competition in bilateral and POOLCO power markets. IEEE Kluwer Academic Publishers, Boston, pp. 93116.
Transactions on Power Systems 16 (2), 194202. Ramos, A., Ventosa, M., Rivier, M., 1998. Modeling competition in
Hobbs, B.F., Metzler, C., Pang, J.S., 2000. Strategic gaming analysis electric energy markets by equilibrium constraints. Utilities Policy
for electric power networks: an MPEC Approach. IEEE Transac- 7 (4), 233242.
tions on Power Systems 15 (2), 638645. Rivier, M., Ventosa, M., Ramos, A., 2001. A generation opera-
Hobbs, B., Rothkopf, M., ONeil, R., Chao, H., 2001. The Next tion planning model in deregulated electricity markets based
Generation of Unit Commitment Models. Kluwer Academic on the complementarity problem. In: Ferris, M.C., Mangasarian,
Publishers, Boston. O.L., Pang, J.-S. (Eds.), Applications and Algorithms of
Hogan, W.W., 1997. A market power model with strategic interaction Complementarity. Kluwer Academic Publishers, Boston,
in electricity networks. Energy Journal 18 (4), 107141. pp. 273298.
Kahn, E., 1998. Numerical techniques for analyzing market power in Rudkevich, A., Duckworth, M., Rosen, R., 1998. Modeling electricity
electricity. Electricity Journal 11 (6), 3443. pricing in a deregulated generation industry: the potential for
Kelman, R., Barroso, L., Pereira, M., 2001. Market power assessment oligopoly pricing in a Poolco. Energy Journal 19 (3), 1948.
and mitigation in hydrothermal systems. IEEE Transactions on Scott, T.J., Read, E.G., 1996. Modelling hydro reservoir operation in a
Power Systems 16 (3), 354359. deregulated electricity market. International Transactions in
Klemperer, P.D., Meyer, M.A., 1989. Supply function equilibria in Operational Research 3, 243253.
oligopoly under uncertainty. Econometrica 57 (6), 12431277. Smeers, Y., 1997. Computable equilibrium models and the restructur-
Murphy, F., Smeers, Y., 2002. Capacity expansion in imperfectly ing of the European electricity and gas markets. Energy Journal 18
competitive restructured electricity market. Operations Research (4), 131.
(submitted for publication). Unger, G., 2002. Hedging strategy and electricity contract engineering.
Neuhoff, K., 2003 Integrating transmission and energy markets Ph.D. Thesis, Swiss Federal Institute of Technology, Zurich.
mitigates market power. CMI Working paper 17, Massachusetts Varian, H.R., 1992. Microeconomic Analysis. W.W. Norton &
Institute of Technology, Center for Energy and Environmental Company, New York.
Policy Research. Ventosa, M., Denis, R., Redondo, C., 2002. Expansion planning in
Oren, S., 1997. Economic inefciency of passive transmission rights in electricity markets. Two different approaches. Proceedings of the
congested electrical system with competitive generation. Energy 14th PSCC Conference, Seville, July.
Journal 18 (1), 6383. Vives, X., 1999. Oligopoly Pricing. MIT Press, Cambridge, MA.
Otero-Novas, I., Meseguer, C., Batlle, C., Alba, J., 2000. A simulation Wei, J.-Y., Smeers, Y., 1999. Spatial oligopolistic electricity models
model for a competitive generation market. IEEE Transactions on with Cournot generators and regulated transmission prices.
Power Systems 15 (1), 250257. Operations Research 47 (1), 102112.