A first revision onreal systems has revealed that the problem

can betackled in verydifferentways, dependingonthe marketsystem characteristic (nodal orsingle-bus), the dispatching method
(sequential vs.joint) orthe ancillaryservices considered (typically
frequency and loadregulation divided into primary,secondary and
tertiary reserves). In additionto these factors, newpossible market
designs might arise, as,for instance, the existence of a uniquemarket in which the whole group of services is cleared bymeansof a
uniqueoffer.
It has beenpointed out that reserve modelingrequires the consideration of short-term constraints such asthe rampinglimits of
the generationunits orthe responding times of reserves. In addition,reserves are related to uncertain contingencies, reinforcing the
stochastic/probabilisticdimensionof the problem.It also requires
the formulation of interaction factors betweencommoditiesas,for
instance, lost opportunity costs betweenservices,orthe dependencyof reserve requirements onenergydemand.
Centralized dispatcheshave been thoroughly studied and a
widevariety of modelshavebeenelaboratedleading to interesting
results. A challenging via for future workwill bethe study of the
couplingbetweenmedium-term decisions (hydrothermalcoordination,fuel and maintenanceplanning, etc.)withthese short-term
centralized energyand reserve marketmodels.
Onthe contrary, oligopolistic models, havenot yet reached the
same level of maturityfor the combined energy and reserve dispatch than for the onlyenergy dispatch. Asshown in this paper,
the degreeof mathematicalcomplexity turns out to beconsiderablyhigher.Furthermore, the inclusion of additional variablesand
constraints criticallyincreases the size of the problem,and consequently, the expected computational times. These drawbackswill
surely besome of the challengesthat future works will haveto cope
with.
The effect of uncertainty is limited to the growth rate of
demand, to which a Markov Chain Monte Carlo approach is
applied. One hundred sequences of thirty years of demand
growth rates were generated. The performance of each of
the different market designs was tested against each of these
demand growth rate scenarios. With such a large number of
scenarios, statistical analysis of the results yields statistically
significant results. The differences between the average outage
rates are an indicator for the relative reliability of the different
market designs. We should not expect large differences in
long-run average prices, as investment is programmed to
take place up to the point that it is just expected to be profitable. Therefore, in every market design, long-run average
prices should approximate long-run average cost. However,
the differences between the standard deviations in de the different models are an indicator for price volatility. In an earlier
version of the model, historic data and a fixed growth rate were
used ( De Vries and Heijnen, 2006 ). However, we could not
generalize from the historic data while the fixed growth rate
provided little indication of the stability of the different market
designs in the presence of random variations in the growth rate
of demand. The Monte Carlo approach makes it possible to
evaluate the impact of demand growth rate uncertainty upon
various market designs in a statistically significant way.
The basic structure of the model is as follows. For each
year, the model calculates the supply function and the loade
duration curve for an imaginary electricity market (loosely

based upon the Netherlands). The supply function is

constructed by combining existing generation capacity with
the new generators that are built in the course of the model
run. From this the price e duration curve is calculated, which
provides the generating companies revenues. Based on the
merit order, fuel type and efficiency, the generating companies variable costs can be calculated. After subtracting fixed
costs, the net profit remains.
All the versions of the model start with the same generation
portfolio, based upon the Dutch stock of large generating
plants. Capacity is somewhat short in the beginning of the
model run and the same amount of new capacity is assumed
to be in the pipeline in each version of the model, hence
the similar volumes of generation capacity in the first years
of each version.
4. Model results
As a first step, one hundred time series of the demand
growth rate were generated randomly, using the function that
was described in Section3.3.
Fig. 5 shows the distribution of the demand growth rates in
the model runs. These 100 Markov chains formed the basis for
a Monte Carlo analysis of the different market designs.
In the presentation of the results, means and standard
deviations are calculated over only the last 20 years of the
model runs, in order to filter out the effect of any transition
effects that might occur during the first five years of the run.
As mentioned before, the same 100 stochastic runs were
used for each different market design. Differences in the
results are therefore entirely caused by differences in the
market designs, not by random differences between the input
data. The standard deviation was 0.9% points. This is a fairly
modest demand growth rate with also a limited variation in
growth rates. Thus our model provides a conservative testing
environment, in the sense that in many markets, uncertainty
about demand growth is higher than here.
A statistical analysis of the model results is presented. The
main performance indicators of the different market designs
are the average price of electricity (which includes payments
for capacity), the standard deviation of the price and the
average number of hours per year with insufficient generation
capacity. In addition, for each market design the results of one
typical run are plotted.
As a base case, the model is run for an energy-only market,
that is, a market in which there are no provisions for stimulating investment in generation capacity other than the price
mechanism. In addition, several capacity mechanisms are
tested: capacity payments, operating reserves pricing and
capacity obligations. These are all run under the assumption
of effective price competition. Finally, a run is presented in
which market power is simulated.
Where possible, the parameters of the different capacity
mechanisms were chosen to resemble the real-life implementations of these models. The level of capacity payments was
modeled after the payments in Argentina and Chile and the
mandated reserve margin in case of capacity obligations was
taken from PJM. Operating reserves pricing has not been
implemented, but the combination of the size of the operating
reserve and the system operator s willingness to pay for
reserves were calculated to be optimal in theory (De Vries,

2004) and, at 10% of installed capacity, appear to be reasonably large. Of course, choosing different parameters will
lead to different performances. However, because investment
is modeled to always tend towards an equilibrium where price
equals long-run marginal cost, the models will provide
sufficient indication whether an alternative constitutes an
improvement or not. Therefore this comparison provides the
necessary insight into the dynamic behavior of the different
capacity mechanisms, but the results should be interpreted in
a qualitative manner.
outages and schedule maintenance. The outages in generation
units are modeled using a two-state probabilistic generation
model. A forced outage rate (FOR) is assumed for each technology
j . Interruptible contracts are seen as a generation technology
which has a variable cost and no fixed cost.
The demand in each local market is characterized by a load
duration curve and modeled on an hourly basis. The growth rate
of demand is represented by a discrete random variable following
a triangular distribution function similar toDe Vries and Heijnen
(2008).
3
In order to consider the influence of electricity trading
between local-markets on spot prices, demand is modified
according to the power flow between systems.
To calculate the electricity price, we assume a perfect competition framework and it is generally settled by the marginal cost of
generation which in turn corresponds to the variable cost of the
marginal generation technology. However, if demand exceeds the
available generation capacity, the electricity price is equal to the
cost of interruptible load contracts.
3.3. Modeling interaction between local markets in relation to their
interconnection capacity
Within regional electricity markets, local markets continuously interact. These interactions not only allow the most efficient
generation resources to be used, but also it increases the security
of supply. In order to consider these interactions, we rely on a
simplified version of an optimal power flow (OPF) algorithm,
which dispatches generation assets according to the regional
merit order (least-cost) subject to the physical constraints of the
interconnector. Here, the power flow through the transmission
line is the result of the free interaction of generation and demand
in both markets until the constraint of transmission capacity are
reached. It is assumed that system operators will not interrupt
exports to adjacent markets in case of a domestic emergency of
supply. This seems to be the purpose stated in Article 24 of
Directive 2003/54/EC: In taking the measures to be adopted in
emergency situations, Member States shall not discriminate
between cross-border contracts and national contracts .
We consider that the interconnector capacity is constant and
perfectly reliable throughout the simulation period. Although the
relationship between the dynamics of generation and transmission investments is strongly linked, we isolate the dynamic nature
of generation investment in order to focus on the impact of
market design on generation investment.
3.4. Investor s behavior
3.4.1. Investor s anticipation in an imperfect information context
The key economic agents in the model are the power producing companies. Their decisions cover investment in and decom-

missioning of power plants. In the model, investment decisions in

new generation capacity mainly depend upon the expected prices
that result from the expected market condition which is a
function of expected demand and expected available generation.
We model a forward merit-order dispatch
in order to calculate
the future electricity prices. In a mature and liquid electricity
market, spot prices forecasting should be driven by forward
markets. However, in most electricity markets, forward markets
are insufficiently developed to provide a firm indication of future
demand for generation capacity. Alternatively, investors may
make their own forecasts. In an ideal situation without risk
aversion and perfect information, these price forecasts should
converge with the prices agreed in any long-term contracts.
Similar to De Vries and Heijnen (2008), we consider that firms
have information on their competitor s decisions around new
generation capacity and decommissioning. Thus, a strong price
signal may not necessarily result in an investment, if there is
considerable new capacity already under construction. Due to
imperfect information, lumpy investments and long lead-times,
investment decisions cannot be optimal. These investments are
prone to cycles, which are characterized by lumpy adjustment
and frequent periods of inactivity.
At this stage of the model description, it is relevant to discuss
the way in which producers make use of the information available
to calculate their demand and available generation forecasting.
We adopt an approach based on the fact that firms do not know
3
Triangular distribution reflects the tendency towards cycles in the general
economy which creates related cycles in the demand for electricity.
According to Consentec (2012), the question of whether security of supply is to
be considered from a
national or from a European perspective has to be answered by politic
s. There are different
consequences depending on the answer given to this question. Consentec (2012) st
ates that in case
of a European perspective, there would be no imperative need for acti
on whereas a national
perspective would lead to a need for action in the medium to long term. Then, an
efficient solution
could be provided only by comprehensive capacity markets. BET (2011) an
d EWI (2012) look at
national security of supply and see a need for capacity mechanisms. DICE (2011)
sees this issue
primarily at European level, particularly if the introduction of comprehensive c
apacity markets is at
stake. Both DICE (2011) and r2b (2012) consider however that there is a possibil
ity to reach the
target also by means of a strategic reserve when looking on security
of supply from a national
perspective.
The definition of security of supply as a public good in combination with a nati
onal perspective make
the need for action appear more urgent than currently reflected by market prices
because imports
from the internal EU market lead to a reduction in prices. Should security of su
pply be defined as a
public good, the consequence would be that behaviour based on price elasticity w
as excluded from
the market because supply would be guaranteed and all consumers would have to pa

y for that. A
national perspective on security of supply leads to higher costs due to national
excess capacities.
These capacities would be seldomly used, since imports usually substitute their
production. These
definitions should be politically answered, though it would be advisable to take
the consequences of
these definitions thoroughly into consideration.

Assumptions on the supply side play a crucial role for the short to medium-term
perspective on the
need for action. Many electricity market models need to be based on
assumptions regarding the
technical lifetime of the different types of power stations in order to draw con
clusions on the future
need for investment. The BET (2011) and EWI (2012) models have been b
ased on these fixed
lifetimes and have thus determined the future need for new power stat
ions. Consentec (2012)
considers the assumption of typical
lifetimes generally only suited to a limited ext
or a more
detailed assessment of the development of the power plant fleet . The argum
ents of r2b (2012)
likewise refer to potential incentives for extensions or reductions of the lifet
ime of conventional power
plants.
The assumption of fixed technical lifetimes of power stations helps to assess th
e order of magnitude
of future investment needs. This assumption however represents only an aid. In r
eality, it is mainly
the electricity price that plays a decisive role when retrofitting measures and
consequently extended