Professional Documents
Culture Documents
net/publication/329098216
CITATIONS READS
5 157
2 authors:
All content following this page was uploaded by Christopher Koch on 15 November 2019.
Version 2018-11-09
d Mercator Research Institute on Global Commons and Climate Change (MCC), Germany
Abstract – Previous studies have noted that, surprisingly, Germany’s dramatic expansion of wind and solar energy coincided
with a reduction of short-term balancing reserves. This paper provides further and updated evidence, supporting this
“German Balancing Paradox”: since 2011 wind and solar energy nearly doubled while reserve requirements and reserve
activation declined by around 50%. We quantitatively explore one reason for reduced balancing needs: increased and
improved short-term wholesale electricity trading. Electricity trading is now commonly done around the clock and based on
quarter hours, rather than hours. The shift to quarter-hourly products alone explains a decrease in balancing energy by 17%.
we also find strong evidence for market parties to respond efficiently to imbalance charges, suggesting that market-based
approaches to balancing work.
Keywords: balancing energy; intraday market; variable renewables; wind energy, solar energy; market design; renewables
system integration
Highlights
• Since 2011 wind and solar energy nearly doubled while balancing energy declined by 50%.
• It is now common to trade electricity around the clock, which helps reduce imbalances.
• Electricity trading is now done on a 15 minute basis, which also reduces imbalances.
• In general, we find widespread support for the notion of efficient short-term electricity markets.
• With good market design, integrating large volumes of renewable energy seems feasible at low cost.
2 Background
2.1 The Balancing System
2.1.1 Regulatory Framework
2.1.2 Self-Dispatch, Balancing Responsibility and Portfolio Management
2.1.3 Balancing Reserves
2.1.4 Imbalance Settlement and Incentives
2.2 Recent Developments in Balancing
3 Quarter-Hourly Products
3.1 Market Development
3.2 Components of System Balance
3.3 Quarter-hourly Trading and System Balance
3.4 Impact of reduced Deterministic Imbalance
4 24/7 Trading
4.1 Market Development
4.2 Peak vs. Off-Peak Balancing Needs
6 Conclusions
7 Reference list
Figure 1: Share of wind and solar energy in annual electricity consumption. Data source: IEA
Monthly Electricity Statistics.
Balancing reserve requirements. Until five years ago, it was conventional wisdom that
expanding renewable energy capacity drives up the demand for standing and spinning short-
term reserves used for balancing. This seems highly plausible: holding everything else fixed,
higher wind and solar capacity translates into larger forecast errors, increasing the reserve
requirement. While there are diverging estimates on the size, the mode-based literature is
unambiguous regarding the sign of the effect (Batalla-Bejerano and Trujillo-Baute, 2016;
Brouwer et al., 2014; dena, 2010; Deutsches Zentrum für Luft- und Raumfahrt et al., 2012;
Holttinen et al., 2011; Morbee et al., 2013; NREL, 2013).
Contribution of this paper. Hirth & Ziegenhagen (2015) remained quite speculative about the
reasons for reduced balancing requirements. Among the most plausible hypotheses are
advanced forecasting techniques, closer cooperation among system operators and expanded
intraday trading. Ocker and Ehrhart (2017) discuss cooperation among system operators. It is
the last channel that we assesses in detail. Intraday electricity markets have become much
more liquid, especially during nighttime and weekends. Also, 15 min trading was introduced
and has been increasingly used, complementing hourly products. Both developments mean
that generators can balance short-term deviations from schedules through bilateral trading,
reducing the pressure on balancing systems. Hence the contribution of this paper is to show
that the “German Balancing Paradox” is no so paradoxical in the end.
Findings. We find further evidence of reduced pressure on German balancing systems, despite
a continued expansion of renewable energy: during the years 2011-17, Germany’s wind and
solar generation more than doubles. At the same time, the use of balancing reserves declined
by 55% and we estimate that reserves could have been reduced by 50% (in fact, reserves were
reduced by just 20%). 15 minute trading helped a lot to reduce imbalances, in particular
predictable imbalances stemming from diurnal pattern in solar generation and electricity
consumption. Such “deterministic imbalances” were reduced by 80%. To understand the
relevance consider the following thought experiment: if those improvements had not been
realized between 2012 and 2017, balancing reserve would have been 7% larger and activation
volumes 17% larger. We also find empirical evidence that 24/7 trading helped reducing
imbalances: in the past, imbalances in non-office hours were both more frequent and more
persistent – not any longer.
Related literature. This research relates to the empirical literature of intraday markets for
electricity. A handful papers discuss intraday markets in the Nordic region (Scharff and Amelin,
2016), Italy (Gianfreda et al., 2016), the Iberian peninsula (Frade et al., 2018) and Germany
(Märkle-Huß et al., 2018). This paper also relates to papers that assess the impact of factors
such as renewable energy generation and forecast errors on intraday market prices (Frade et
al., 2018; Gianfreda et al., 2016; Kiesel and Paraschiv, 2017; Pape et al., 2016; Pape, 2017;
Selasinsky, 2016; Ziel, 2017). The paper that is most closely related to ours is probably Remppis
et al. (2015), who assess the impact of 15 min trading on balancing energy needs. The present
paper is broader in scope (discussing also the impact of trading around the clock), uses a much
expanded dataset, and applies more rigorous methods (multiple regressions as opposed to
simple regressions).
The European target model. The European electricity market “target model”, while not being
formally defined, can be understood as comprising the following elements:
1https://docstore.entsoe.eu/publications/system-operations-reports/operation-handbook/Pages/default.aspx
2 http://data.europa.eu/eli/reg/2017/2195/oj
3 http://data.europa.eu/eli/reg/2017/1485/oj
Balancing responsibility. Each electricity market actor is a “balancing responsible party” (BRP)
or “program responsible party”. Each physical connection point of the grid is associated with
one BRP; in Germany, there exist a few thousand BRPs. They have the responsibility of
balancing a portfolio of generators and/or loads through dispatch of physical assets or through
trade. Non-metered consumers are assigned to their connecting distribution grid. BRPs
provide schedules to the associated system operator and are financially accountable for
deviations from these schedules. Physical consumption/generation quantities are called
“(final) positions” and deviations between positions and schedules “imbalances”. The time
steps of schedules are called “imbalance settlement period” (ISP); they are currently 15, 30 or
60 min in different European countries – by the year 2020 all countries are supposed to use
15 min ISPs.
Portfolio management. As time unfolds, BRPs actively reduce deviations between submitted
schedules and expected physical generation/consumption by either trade or dispatching own
assets. Collectively, these activities are called “portfolio management”. Portfolio management
may include the deliberate choice not to close imbalances.
Spot markets. Spot markets as we use the term comprise the day-ahead auction, usually taking
place at noon for the individual hours of the next day, and intraday markets. Depending on
the country, intraday markets are either organized as a series of auctions, as continuous
trading, or as a mix of both. On some intraday markets quarter-hourly and half-hourly
products are traded in addition to hourly products.
Germany’s intraday market. On Germany’s leading spot power exchange, EPEX Spot, there is
an opening auction for quarter-hourly intraday products each day 3 p.m. for delivery the next
day, followed by continuous trading. Continuous trading closes 30 minutes before real time;
this time gap is called “gate closure”. Thereafter there is an “extended trading” until five
Balancing reserves. To this end, TSOs keep balancing reserves (or balancing capacity) that can
respond quickly to deviations. Due to unbundling regulation, system operators do not own
generation assets, but instead procure balancing reserves from generators or loads. In real
time, system operators activate reserves upward and downward, thereby obtaining balancing
energy.
Types of reserves. European system operators hold a variety of reserves, mainly differentiated
by the time they take to be activated. Historically, they used to be called primary control,
secondary control and tertiary control (or minute reserve), each of which may raise or lower
the power balance (upward or downward reserve). The electricity guidelines introduced
between 2016 and 2017 use a new terminology:
• Frequency Containment Reserve (FCR)
• Automatic Frequency Restoration Reserve (aFRR)
• Manual Frequency Restoration Reserve (mFRR)
Sizing. The FCR is sized in a deterministic approach to match the largest credible contingency
(N-1 criterion). It is set to be 3000 MW for the interconnected system of continental Europe
(“UCTE region”) and distributed to system operators pro rata relative to annual electricity
consumption. The sizing of aFRR and mFRR is done by system operators through a range of
different methodologies, resulting in quite different levels of reserves among European
countries.
Procurement. Balancing reserves are procured through legal supply obligations, bilateral
negotiations, or more formalized balancing markets for periods ranging from days to years
(ENTSO-E, 2018b).
Germany’s balancing market. German balancing reserves are procured through the online
platform Regelleistung.net, which is now used by more and more neighboring TSOs as well.
The design of balancing market auctions is prescribed by Germany’s regulator. During the past
years, the duration of contracts was shortened from months to weeks and days and minimum
bid sizes were reduced, leading to a strong increase in participation and competition. Both
capacity and energy is remunerated on a pay-as-bid bases and the award criterion has been
changed in 2018 to include the energy price also. Market clearing pricing and free bids are to
be introduced in late 2019.
Incentives. Close to real time, market parties regularly face the choice to close open positions
through intraday trading or through imbalance settlement. Some countries, including
Germany, impose a legal obligation to close all open positions on intraday markets, but the
inherent uncertainty in wind and solar generation means this is rarely enforced in practice.
The economic incentives for BRPs to reduce imbalances is the difference between the
imbalance price and the intraday price they would have paid (in a sense the opportunity cost),
a spread that we dub the “imbalance price spread”. To be precise, we define the imbalance
price spread as imbalance price minus ID3 price. We calculate the imbalance spread ex-post;
it should be noted that at gate closure the imbalance price is not yet know by market
participants.
Balancing reserve procurement. In fact, however, German TSOs procure less balancing capacity
than expected. During 2011-14, the total frequency restoration reserve (FRR) procured – aFRR
4
http://www.erneuerbare-
energien.de/EE/Navigation/DE/Service/Erneuerbare_Energien_in_Zahlen/Zeitreihen/zeitreihen.html
Figure 2: Frequency restoration reserves procured by TSOs (balancing capacity) and generated
energy from wind and solar plants.
Balancing energy activation. The amount of balancing energy – i.e. the activation of reserves –
has declined even more: it dropped from well above 7 TWh (resp. 840 MWh/h) in 2011 to 2.5
TWh (resp. 280 MWh/h) in 2017. However, this ignores the fact that quite a lot of balancing
is now done (or avoided) through international cooperation among TSOs, such as imbalance
netting. If those measures are accounted for, 4 TWh of balancing energy (resp. 470 MWh/h)
was used in 2017, a reduction of 45% as compared to 2011 (Figure 3). So a 110% increase in
wind and solar energy coincided with a 20% decline of balancing capacity and a 45% decline
of balancing energy.
Table 1: Number of quarter hours with an activation of at least 80% of the balancing reserve.
Year Activation
2011* 99
2012 168
2013 33
2014 1
2015 1
2016 0
2017 0
*The numbers for 2011 contain only data of the
second half-year.
Ex-post sizing. We use the same quarter-hourly reserve activation data to estimate the
necessary reserve ex-post. Obviously such descriptive statistics of realizations is different from
estimating a reserve requirement for the uncertain future, but we will see that our findings
are interesting nevertheless. For the calculations we use quarter-hourly data and,
conservatively, ignore cross border netting because it is not available reliably. We assume a
reliability target of 99.975% (1 – 0.0225%), which is the same as the German TSOs use
(Consentec GmbH, 2018). This translates to less than one quarter hour per quarter of the year,
which was the period of sizing (except the week around Christmas, when additional reserves
are procured). We hence report the minimum and maximum quarter hour for each quarter of
the year. Figure 4 compares the reserves procured by TSOs with our ex-post measure of
reserve needs. It turns out that in the years 2011 through 2013 both measures match well on
average while there was actually a number of quarters where TSOs did not meet the reliability
target with the amount of reserves procured. In sharp contrast, the reserve requirement as
calculated by us drops strongly since 2014. In 2017, we estimate that TSOs could reduce
reserves by another 35% without compromising their reliability target. In other words: the
10
The German Balancing Paradox. Hirth and Ziegenhagen (2015) dubbed the decline of
balancing reserves despite an increase in renewable energy the “German Balancing Paradox”.
This section has not only confirmed the paradox, it seems now starker than ever: during a time
when renewable energy grew by 110%, balancing reserves could have probably been reduced
by 50% and the use of these reserves by 55% (Figure 5).
Figure 5: Energy generated from renewable energy sources, use of balancing energy and ex-
post sizing for balancing reserve compared to the base year 2011. Note: unlike in Figure 1,
here we report only data from 2011 onwards due to availability of balancing energy data.
11
This section provides a market overview. We then define systematic imbalances as those that
can be addressed best with quarter-hourly trading and show how they have been reduced
over time, along with an increase in quarter-hourly trading. Observing that quarter-hourly
trading volumes have increased from 1 TWh in 2012 to 10 TWh in 2017, our regression analysis
indicates that trading on the two markets reduce systematic imbalances and this positive
effect increases with higher quarter-hourly deviations of load and wind. Accordingly we report
that the strong decline in systematic imbalances has reduced the total absolute system
balance by 17% and the procurement of FRR potentially by 7%.
5
EPEX SPOT introduced an auction for quarter hourly products in Germany in December 2014. They
call it intraday auction even though it takes place at 3 pm the day before.
6
The trading data are provided by Agder Energi Solutions GmbH and consider all single transactions
with at least sell or buy area in Germany. The information includes inter alia the price, the traded
volume and the trading time.
12
13
System balance over time. These systematic patterns clearly reduced in 2014. There were still
some regular patterns left in the evening probably caused by load portfolios, because there is
no solar production during the night and the load drops steeply during these hours. The
introduction of the quarter-hourly intraday auction in December 2014 helped to mitigate the
systematic quarter-hourly imbalances. It ensured an entire management of the systematic
quarter-hourly deviations as they are accurately predictable the day before and the effort of
a single auction is much lower for the market participants. Therefore, a considerable share of
the trading volume shifted from the continuous market to the auction (Neuhoff et al., 2016),
but the total quantity of both markets increased in the following years (Figure 6). Finally, the
average system balance for 2017 demonstrates no systematic quarter-hourly deviations. 7
There is a similar development of the utility frequency as Weißbach et al. (2018) show. So, the
demanded quarter-hourly portfolio management is implemented by the BRPs.
7
Section Fehler! Verweisquelle konnte nicht gefunden werden. discusses the shift to positive system
balances in 2017 in detail.
14
ℎ is the hour of the year (1,2,…,8760) and 𝑞ℎ are the quarter hours of each hour (1,2,3,4).
Average hourly system balance. We divide 𝐻𝑆𝐵 into three components: Stochastic imbalances
𝑆𝑡𝑜𝐼 like power plant outages and forecast errors of solar and wind production or load,
strategic imbalances 𝑆𝑡𝑟𝐼 to benefit from a bias of the imbalance price spread (cf. chapter 5),
and deterministic imbalances 𝐷𝑒𝑡𝐼 that stem from coarse portfolio management, e.g. if
market actors trade only peak/off-peak blocks rather than hourly products.
Within-hour deviation. Similar, one can divide 𝐷𝑒𝑣𝑞ℎ,ℎ into the same three constituents:
stochastic imbalances due to inaccurately predicted gradients, strategic imbalances portfolio
managers take consciously based on a forecasted imbalance price spread, and deterministic
imbalances that stem from parties trading only hourly (but no quarter-hourly) products.
𝐷
1 (5)
̂ 𝑞𝑑,𝑑
𝐷𝑒𝑡𝐼 = ∑ 𝐷𝑒𝑣𝑞𝑑,𝑑
𝐷
𝑑=1
with 𝑞𝑑 = 1,2,…,96 denoting the quarter hours of each day and 𝑑 = 1,2,3,…,365 the days of
the year. To account for differences in diurnal solar and load patterns between winter and
summer as well as working days and non-working days, we calculate equation (5) separately
for the quarters of the year and working / non-working days.
15
Figure 8: Development of average absolute 𝐷𝑒𝑣, deterministic imbalance 𝐷𝑒𝑡𝐼 and average
quarter-hourly intraday trading volume per week from 2012 to 2017. The seasonal pattern in
trading volumes can be explained by the pattern of solar generation.
Assessing the impact. We analyze the described link with a multiple linear regression model
estimating 𝐷𝑒𝑣 as the dependent variable for every quarter hour in the sample 𝑞 . The
regressors are the trading volume of the quarter-hourly intraday markets and the absolute
quarter-hourly deviations to the hourly mean for solar (𝐷𝑒𝑣𝑆𝑜𝑙𝑎𝑟 ), wind (𝐷𝑒𝑣𝑊𝑖𝑛𝑑 ) and
electricity load (𝐷𝑒𝑣𝐿𝑎𝑜𝑑 ). Figure 9 illustrates the example for solar generation. We consider
absolute numbers for all deviations, as the trading volume is always positive as well and
include dummies for the quarter hours of a day (qd) to cover other systematic influences.
16
DevSolar
Long positions
DevSolar
Short positions
Models with and without interdependencies. We test two models: the first one includes only
the aforementioned variables and provides intuitively interpretable coefficients. The second
one covers also interdependencies between trading volume and the three within-hour
deviations and between 𝐷𝑒𝑣𝑆𝑜𝑙𝑎𝑟 and 𝐷𝑒𝑣𝐿𝑜𝑎𝑑 , as there is probably a relation between these
parameters.
Significant parameters. Table 2 summarizes the results of both regression models. The
interdependency of trading volume and 𝐷𝑒𝑣𝑊𝑖𝑛𝑑 as a regressor is significant on a 5% level.
All other parameters are significant on a 0.1% level. The coefficient estimates show, that an
increase of absolute quarter-hourly deviations for photovoltaic, wind and electricity load lead
ceteris paribus to higher quarter-hourly deviations of the system balance. This indicates that
a fraction of the deviations is not balanced at the intraday market. The coefficients for load
and wind are higher than for solar. A possible explanation for wind is that the gradients are
more difficult to predict, because they are not systematic as the other two. The high
coefficient for load suggests that quarter-hourly portfolio management was partly neglected
by retailers during the considered period. Solar portfolios are predominantly managed by
TSOs, which have a special interest in accurate portfolio management for reasons of system
stability.
Positive impact of quarter-hourly trading. The coefficient for trading volume is -0.102 for the
model without interdependencies, which means that the quarter-hourly deviation decreases
by 0.102 MW with every megawatt of trading volume. So, the results of the linear regression
show the positive impact of quarter-hourly intraday trading on the on the system balance by
balancing within hourly deviations. The average trading volume is around 1200 MW today.
Based on our estimation, this reduces the within-hour deviation for each quarter hour by 120
MW.
17
* and *** denote that a test statistic is statistically significant on a 5% respectively 0.1%
level of significance.
18
We want to quantify both effects by adding the difference between the deterministic
imbalances of 2012 and 2017 to the system balance of 2017:
2012
𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑡𝑖𝑐𝑎𝑙 𝑆𝐵𝑞2017 = 𝑆𝐵𝑞2017 + (𝐷𝑒𝑡𝐼𝑞𝑑 2017
− 𝐷𝑒𝑡𝐼𝑞𝑑 ) (7)
The deterministic imbalances 𝐷𝑒𝑡𝐼 are calculated with the actual numbers for both years as
described in equation (5).
Implications. Table 3 summarizes the results of the calculations. The total absolute system
balance is 3131 GWh with the actual numbers of 2017 and 3652 GWh by adding the difference
between the deterministic imbalances of 2012 and 2017. This corresponds to a growth of
16.6%. There is also a significant influence on the largest absolute system balances as the
results of the ex-post sizing show. The cumulated capacity of positive and negative FRR is on
average 6.77% higher for the synthetic system balance (5021 MW compared to 4703 MW).
The results confirm the two hypotheses mentioned above. The total activation volume and
the numbers for FRR ex-post sizing are higher, if we include the deterministic imbalances of
2012. Active trading on the quarter-hourly intraday markets led to a reduction of the regular
patterns and improved thereby the system stability.
4 24/7 Trading
The Federal Network Agency criticised that especially aggregators of non-dispatchable
renewable generation units made too little effort on weekends, holidays and during the night
in terms of active portfolio management until short time before delivery. We try to illustrate
this behaviour by empirical analyses and want to figure out whether this has been changed
19
This section starts with an analysis of trading volumes and median execution time in 2012 and
2017 indicating a development towards more 24/7 trading. Afterwards, we examine the effect
on the system balance detecting a reduction of high absolute system balances. One reason is
that active portfolio management enables a quick reaction on new information leading to a
fast compensation of high imbalances.
Table 4: Average hourly trading volume divided into peak and off-peak products. The trading
volume grew significantly faster for off-peak products.
Execution time. The execution time of trades is another valuable indicator to analyse the
trading activity on a continuous market. We look at the time difference between the execution
time of the single transactions and the gate closure time for the associated trading product to
achieve a comparability between the different trading products. A lack of 24/7 portfolio
management means the majority of trading for night and weekend hours is done with a large
time difference to delivery. This is critical for wind and solar portfolios as forecast accuracy
improves with less time to delivery (Weber, 2010; Zhang et al., 2015).
Course in 2012. Figure 10 shows the median execution time before gate closure for the hourly
intraday products of 2012 and 2017. The image confirms that in 2012, a lot of market
8
Off-peak time is between 8 pm and 8 am on working days and during the whole non-working days.
20
Course in 2017. The course of the median execution time for hourly products in 2017 shows a
behavioural change of market participants. The overall pattern is still the same, but there is a
significant reduction of the median for non-office hours so that the numbers in the morning
hours are similar to the afternoon hours. The products between 9 and 11 am are the only ones
with a higher median in 2012 than in 2017. This is explainable by a more coninuous trading
activity meaning that not all open positions must be closed at the beginning of the morning
shift. The time lag to gate closure is still the highest for the late evening hours, but the slope
is smaller than in 2012.
Remaining pattern. A reason, why the median execution time is still higher in the evening, is
the update of the important global weather model of the European Centre for Medium-Range
Weather Forecasts (ECMWF) between 6 and 7 am and pm UTC (ECMWF, 2018). These data
are the basis for renewable production forecasts (Sperati et al., 2016) and updates provide
significantly higher accuracy. Consequently, it is appropriate to make a first adaption of
9
The indicated time refers to the beginning of the hour. Thus, the trading product for delivery from 7
to 8 am is represented by 7 am.
10
There is a jump discontinuity between 11 pm and midnight, because midnight is the beginning of the
day and 11 pm its end. Trading for midnight is possible between 3 pm and 11:30 pm the day before (8.5
hours) and for 11 pm between 3 pm the day before and 10:30 pm same day (31.5 hours).
21
24/7 trading on track. There are probably still some BRPs who have not implemented 24/7
trading for their portfolio yet (either be done by themselve or by service providers), but the
analysis above indicates a behavioural change of many market particpants.
Peak vs. off-peak activations. The emperical data show that the absolute system balances are
higher for peak than for off-peak hours (Figure 11). Median, lower and upper quartile are all
lower for the off-peak quarter hours in 2012 and 2017 proving that there is mostly less
demand for capacity reserve compared to peak quarter hours. From 2012 to 2017, the relative
deviation between the system balances for peak and off-peak hours is nearly constant for the
lower quartile and the median. Little deviations occur even by using the latest available
forecasts and active trading has no effect on these situations. But the relative deviation
increased for the upper quartile from 3% in 2012 to 12% in 2017. It seems that active portfolio
management has the largest impact on high imbalances.
22
Distribution of high absolute system balance. This hypothesis is confirmed by Table 5 that
shows the distribution of quarter hours with the 5% highest absolute system balances for 2012
and 2017. In 2012, 67% of these situations occurred in off-peak hours. This share dropped to
53% in 2017. A further analysis illustrates the relation to the 24/7 trading. Active trading
means a quick reaction on new porfolio information and public market data. Therefore,
portfolio managers should be able to balance high imbalances faster. This can be aproved by
an examination of the average duration of consecutively high system balances. The limit is the
95th percentile seperately for the years 2012 and 2017 (Table 5).11 It turns out that the average
duration dropped stronger for the off-peak hours. In 2012, market participants reacted slower
on high imbalances during off-peak than during peak hours, whereas they were faster in 2017.
The results confirm the assumption that active portfolio management has the largest impact
on high imbalances, because market participants can react on large forecast adaptions and
asset outages.
2012 2017
Number of observations
11
We dropped intervals of one or two quarter-hours for this calculation. In these cases, exceeding and
falling below the limit cannot be caused by active intraday trading as its lead time was at least 30
minutes.
23
In this section, we discuss the apparent paradox of a systematically unbalanced system with
arbitrage-free markets. We start by discussing the status quo and then explore developments
over time.
24
Asymmetric imbalance price spreads. These two observations seem to be inconsistent. They
can be explained with the fact that absolute imbalance spreads tend to be larger if the system
is long (Figure 13). In other words, from a BRP’s perspective, being short when the system is
short involves a small penalty, but being long when the system is long involves a much larger
penalty. For example, during the past two years the imbalance price spread was around EUR
20 when the system was somewhat short of energy (up to 1000 MW). When the system was
oversupplied with the same amount of energy, the imbalance price spread averaged EUR -37.
This ratio hold also for higher absolute imbalances: surplus led to about twice the imbalance
price spread as shortage. It is therefore economically rational for market parties to rather
error on the right sight (be short).
25
26
Figure 15: The mean system balance went from surplus to shortage of energy. An average bias
of 200 MW (as in 2016) corresponds to 1.8 TWh annually.
Trading. The introduction and adaptation of quarter-hourly trading products as well as more
widespread 24/7 trading helps explaining the decline in absolute system balances, as we have
argued in sections 3 and 4. We cannot think of any reason these changes by themselves should
have an effect on bias in balancing. However, as we will argue in the following, those
advancements in trading give parties additional possibilities to exploit economic incentives.
Outages. In contrast to forecast errors of load and renewable energy, unplanned outages are
a biased source if imbalances in the sense that they can only lead to shortage of energy, not a
surplus. However, a systematic change in the distribution of outages of power plants, loads or
interconnectors seems unlikely.
27
Figure 16: TSO forecast errors of wind and solar energy (day-ahead forecast minus
extrapolation of actual generation).
Incentives. Chaves-Ávila et al. (2014) and van der Veen et al. (2012) already emphasize that
imbalance prices are an important motivation for active portfolio management. Given the
collective evidence, we find incentives the most plausible drivers behind the development.
Significant arbitrage opportunities in 2011 lead market actors to reduce systematic long
positions. BRP behavior is quite efficient since 2015 with respect to systematic deviations: they
cannot make a systematic profit anymore. But this implies a systematic shortage of energy of
100 MW to 200 MW because of the asymmetric imbalance price spreads.
Implications. If system operators and regulators would like to see a systematically unbalanced
system (zero average system balance), they should consider the incentives that market parties
face. With symmetric spreads, we expect a quick return to zero systematic system imbalance.
28
15’ trading helped. During that time, portfolio management at the 15 minute scale (rather
than full hours) became increasingly common, as evidenced by the sharp increase in quarter-
hourly trading. This helped a lot to reduce imbalances, in particular predictable imbalances
stemming from diurnal pattern in solar generation and electricity consumption. Such
“deterministic imbalances” were reduced by 80%. To understand the relevance consider the
following thought experiment: if those improvements had not been realized between 2012
and 2017, balancing reserve would have been 7% larger and activation volumes 17% larger.
This is further supported by econometric evidence – we find a highly significant reducing
impact of higher trading volumes on within-hour deviations of the system balance.
24/7 trading also helped. Electricity trading around the clock has become more common:
trading volumes increased particularly in night time and during weekends and trading now
takes place close to real time also during these off peak times. In the past, imbalances in non-
office hours were both more frequent and more persistent – not any longer.
Parties respond to imbalance changes. We see strong evidence for efficient markets when in
comes to imbalance charges. The system balance is biased (100 MW average shortage in
2017), which can be fully explained by optimizing market parties: the penalty paid when being
long (and the system also being long) is twice as high than the penalty for being short (when
the system is also short), so portfolio managers try to err on the right side. In past years, parties
could make a profit from being consistently short. This arbitrage opportunity is gone – an
indication of efficient and mature markets.
Incentives work. Taken to a higher level, the evidence presented in this paper let us conclude
that “incentives work”. If the incentives are right and proper market design is available,
markets efficiently integrate quite highly volumes of wind and solar energy. This should give
us confidence to rely on market-based approaches to short-term electricity system operation.
The evidence that arbitrage opportunities will be exploited and hence markets will be biased
if incentives are biased, however, should serve us as a reminder to design such market-based
approaches carefully.
29
30
31
32
33