Negative Prices in Electricity Market
Negative Prices in Electricity Market
Negative Prices in Electricity Market
Electricity Markets
In this paper we describe how liberalisation has lead to the segmentation of trading
opportunities for electricity with different periods to delivery. We clarify the price
characteristics in each segment, including the extreme volatility in short-term prices
and the phenomenon that electricity prices can become negative close to the time of
delivery. With the Dutch market as an example, we show the implications for risk
management and the valuation of derivatives. We argue that a distinct price model is
required for risk management and derivative valuation in each market segment.
Derivative valuation goes beyond the financial contract itself and can be very useful for
taking strategic decisions on flexible generation assets.
By MICHAEL SEWALT & CYRIEL DE JONG 1
LIBERALISATION AND DEREGULATION in electricity premiums on short-term prices are much higher than can be
markets have resulted in active trading between generators, expected from standard pricing models as Black-Scholes
suppliers, distributors, large end users and several intermedi- (1973) or Black (1976) 2 . We therefore analyse the require-
aries for hedging and speculation purposes. In sharp contrast ments that a pricing model should fulfil for the shortest-term
to conventional markets trading has been clearly segmented, (quarter-hourly imbalance) prices. An appropriate price model
both geographically as well as in terms of delivery period. would also improve strategic decisions on flexible real assets.
Geographical segmentation is the result of limited cross bor-
der transport opportunities and different regulations per The Forward Market
country. Most noteworthy however is the segmentation that is The ongoing liberalisation of electricity markets has
due to the non-storability of the commodity: separate trading resulted in a relatively liquid trade of longer-term contracts
mechanisms and markets exist for electricity with different between several market participants. Popular forward con-
periods to delivery, ranging from long-term forward markets tracts are week-ahead, coming months, quarters and years.
to (very short-term) imbalance markets. Each market segment The major part of the trading is settled OTC where the par-
is characterised by distinct price characteristics that provide ties come together, sometimes facilitated by brokers. Even
a challenge for risk management, derivative valuation and though some exchanges are quite successful (e.g. Nord Pool,
EEX), in those markets OTC trades still form a major portion
> Each market segment is characterised by of the total trading in forwards. OTC trading is facilitated
with the adoption of master agreements, which increasingly
distinct price characteristics < follow the standards of the European Federation of Energy
Traders (EFET). In addition, information providers such as
Platt’s and Heren provide some transparency by publishing
asset optimisation. In this paper we clarify the properties of forward prices.
the market segments and focus on the extraordinary charac- Market participants mainly organise electricity trading on
teristics of very short-term prices. We highlight the implica- a country-by-country basis in the form of country desks,
tions for option valuation, which provides the means for valu- because national grids still have their own procedures and
ing flexible generation assets. limited exchange capacities between them. Prices in the for-
The interest in option valuation stems from the limited liq- ward market are quite well comparable to those in other
uidity and large pricing differences between electricity commodity markets. Volatility is limited and forward returns
options. Therefore, market prices do not provide the desired conform to the normality assumptions pretty well, as shown
benchmark on which to base strategic decisions. For example, in Table 1: skewness and kurtosis do not deviate significant-
if a generation plant can be treated as an option on spot elec- ly from 0, so prices exhibit few outliers. Because of this price
tricity prices, then we would ideally value this plant based on behaviour and because hedging with forwards is possible to
tradable options. The illiquid market and the lack of valuation some extent, the standard Black (1976) formula may be
benchmarks is partially due to the inapplicability of standard applied to value European-style call and put options on for-
pricing models to price electricity options with short periods wards .3 Those instruments are traded on the Nord Pool
to delivery, especially when the short-term electricity prices exchange and OTC, and provide a means to manage longer-
become negative. This motivates the growing attention for term risks. Apart from limited excess kurtosis and skewness,
option pricing models for electricity in general and our focus Table 1 also highlights a first indication of term-structure
in this paper on the phenomenon of negative prices. Option effects in forward prices: shorter-term forwards experience
higher volatility than longer-term forwards. This effect is switches in which spikes are modelled as a separate price
much stronger in electricity markets than most other mar- process. The advantage of a separate spike regime is that it
kets and is due to the non-storability of the commodity, better reflects the temporary nature of spikes. An additional
which prevents arbitrage between periods. advantage of this approach is that it allows (under certain
conditions) for the derivation of closed-form formulas for
The Spot Market
Forward trading is mostly organised without the need to
trade on an exchange. On the other hand, spot trading for > Prices in the forward market are quite well
day-ahead delivery is largely conducted on organised spot
markets such as (in Europe) those from Nord Pool, EEX, APX,
comparable to those in other commodity markets <
UKPX, COMEL and Powernext. The advantage of centralised
markets is not only an increase of price transparency, but European-style options on spot prices. Those formulas or
also a reduction in credit and counterparty risk. In the simulation-based methods may be applied to value flexible
Netherlands for example, the APX takes full responsibility for end-user contracts with caps or floors, tradable daily exercis-
able options and generation assets with the flexibility to man-
Table 1. Return Properties of Electricity Contracts
age output on a day-to-day level. Outcomes of those
Volatility Skewness Excess
approaches will generally be quite distinctive from standard
Kurtosis
option-formulas and yield much higher values especially for
M1 32.9% -0.36 -2.14
out-of-the-money call options. This may lead to price caps
M2 20.3% -0.83 1.00
being sold too cheaply and flexibility in energy generation
M3 14.3% -0.01 -0.12
Q1 15.7% 0.04 -0.02 plants being valued too low.
Q2 8.6% -0.18 2.64 Simply modelling baseload or peakload prices is not suffi-
Q3 8.7% -0.25 1.05 cient for some options and the valuation of flexible genera-
Y1 7.4% 0.13 2.11 tion: hourly prices should instead be modelled. Since spikes
Forward statistics are based on weekly (5-daily) returns of German baseload normally last for several hours in a row and revert back to
contracts (volatility is annualised). Period: Jan 2002 – Mar 2003.
normal levels more gradually, it is not so convenient to trans-
Source: Platts/Moneyline.
fer spikes on an hourly level to a spike regime. Approaches
counterparty risk, like a general clearing institute, and facili- that have been applied instead are state-of-the-art time-
tates the exchange of power. On a daily basis potential buy- series models (Guthrie and Videbeck (2002), Cuaresma et al
ers and sellers can hand in bids and offers for power on a (2002)). The main challenge here is to accurately capture the
specific hour for the day ahead. Based on the resulting sup- interdependencies between prices on the same day and
ply and demand curves a market clearing price and a market between similar hours on different days. In Table 2 we
clearing volume are determined for every single hour the observe for example that the correlation between hours that
next day. A transaction will be settled by the APX when a bid are 7 hours apart is lower (0.21) than between a single hour
or ask (buy or sell) is hit. The most important function of the on consecutive days (0.61). A similar complex interdependen-
day-ahead market is giving market participants the opportu- cy exists in volatilities of prices (final column). Finally, some
nity to balance their own delivery or procurement on a short- sort of jump behaviour (positive for peak-hours, negative for
term basis. Both before and after settlement on the
exchange spot trading also takes place on OTC markets.
However, the advantage of the exchange is that it looks after > The Black (1976) model is inappropriate for
the financial settlement and guarantees the physical delivery. valuing daily exercisable options <
Therefore, the counterparty risk is fully reduced in contrast
with bilateral agreements.
Being much closer to delivery than forward contracts, day- off-peak hours) may be required to capture the outliers.
ahead spot price dynamics are inherently different from for- Monte Carlo simulations of hourly price models are used in
ward price dynamics. Since spot price changes are not nor- risk management applications and form the basis for the val-
mally distributed, the standard Black (1976) model is inap-
propriate for valuing (daily exercisable) options, caps, floors Table 2. Correlations Between Hourly Spot Prices
or collars. Distinguishing features of the prices are a strong Time Lag Correlation
level of mean-reversion, seasonality (across seasons and Price Squared Price
weekdays), extreme and possibly time-varying volatility 1 0.78 0.63
(reaching daily levels of 1,000%), and occasional spikes. 2 0.63 0.48
These characteristics have extensively been analysed by aca- 3 0.51 0.37
demics and practitioners alike and different modelling 4 0.44 0.34
5 0.34 0.23
approaches have been proposed. A common approach is a
6 0.27 0.20
mean-reverting model with stochastic jumps to account for
7 0.21 0.16
occasional spikes. Since spikes are often very short-lived, the
24 0.61 0.60
stochastic jump process (which assumes a long-lasting 48 0.36 0.31
impact of spikes) does not work well for some electricity spot Correlations between hourly spot prices and squared spot prices on the
markets. A recent development is the application of regime- Amsterdam Power Exchange.
imbalance markets is still largely unexplored < Combined-cycle installations are basically installed for the
generation of heat (steam) whereby electricity is a co-prod-
uct. Reducing the must-run output is hardly possible from a
at all times. In the Netherlands for example 34 market par- technical perspective or it involves high shutdown costs.
ticipants have the so-called ‘programme responsibility’, Negative prices are acceptable to power suppliers because
which includes the requirement to supply a daily schedule of the opportunity costs of a shutdown period are sometimes
expected supply and demand on a quarter-hourly basis. Just much higher. Generally, prices will be negative in only a
before the electricity is generated and consumed, the net- short period of time and mainly during the night. However,
work operator TenneT settles any discrepancies between Figure 1 shows that negative prices can sometimes last for
forecasted and actual supply and demand (see Figure 1 for long periods of time and can attain extreme levels. The
the discrepancies on 9th March 2003). Apart from its own graph contains the imbalance market results in the
emergency capacity that it may use, the network operator Netherlands on March 9th 2003; it shows that the market is
organises an imbalance market on a daily basis to 'smooth' very volatile and prices can jump from -190 ¤/MWh up to
+120 ¤/MWh within two hours. When negative prices last for
Figure 2. Imbalance Bid Curves
a longer period (corresponding to a positive imbalance),
Quarter-hourly bid curves on 9 March 2003 for the Dutch power market. shutting down generation capacity will pay off and imbal-
Each line (-300, -100, 100, 300 MW) represents the price for which power ance prices will automatically increase.
generators are willing to supply power up to the specific volume.
Negative prices cause sizeable operational problems, for
200
example in energy risk management systems. Not all sys-
tems can handle a negative deal in their VaR-calculations,
150 cash-flow projections or invoice procedures. An even larger
challenge is the appropriate modelling of negative prices for
EUR/MWh
fuel) is positive. Many energy risk management systems use trading of options could be stimulated. Before trading in
the so-called delta hedging strategy 4 in order to forecast short-term options really takes off, option valuation tech-
the optional production capacity in advance. This strategy is niques are already required: to manage and value flexible
based on the same assumptions as other standard option generation assets. In order to determine whether an invest-
formulas. These models assume that electricity and fuel ment is worthwhile or not a power generator can be consid-
prices evolve according to a gradual process (‘Brownian ered as an option on power production. This method of real
Motion’) with no extreme changes, mean reversion or nega- option valuation becomes more and more familiar. However
tive prices: it assumes prices are lognormally distributed. for a proper valuation of the most flexible generation assets,
To understand how negative prices can be dealt with, it is
important to understand why negative prices lead to model-
ling problems. The problems all stem from the fact that stan- > Imbalance prices exhibit similar complex
dard price models are based on price returns (or in fact
logreturns). A return becomes in fact infinite when prices
interrelations as those in hourly spot prices <
approach zero and is not defined at all for negative prices. 5
Incorporating negative prices can basically be achieved with it is important that standard option pricing models will be
two approaches: an indirect (structural) approach and a adjusted to price spikes as well as negative prices.
direct approach. The structural approach does not model The study of negative commodity prices has made clear
prices directly, but models them as the outcome of a price that negative prices have a special impact on option pricing
formation process. This process may include for example models. Standard models as Black-Scholes (1973) and Black
the imbalance (Figure 1) and the imbalance bid curves (1976) are not applicable to options with negative underlying
(Figure 2), from which imbalance prices result. A structural value. New models can be very helpful for financial options
approach offers valuable insights in the formation of prices as well as for strategic decisions (real options) like the man-
and is appealing to industry professionals, who ‘recognise’ in agement of flexible generation assets ■
it the functioning of the market. However, for risk manage-
Footnotes
ment systems they easily become too complex, because
1. The authors thank Kasper Walet (Maycroft Consultancy) and
they need to contain several stochastic variables (such as
Gerard van Baar (Deloitte & Touche) for helpful comments.
imbalance and imbalance bid curves), which provide chal- 2. Black’s (1976) model is similar to the famous Black-Scholes
lenging modelling and implementation tasks by themselves. (1973) model, but applicable to options on forwards and
A direct price modelling approach is not straightforward futures.
either, but at least reduces the problem to one variable: the 3. If returns are skewed or exhibit clear kurtosis the extended
Black formula with separate terms for skewness and kurtosis
price. We propose to allow for negative prices by setting a (Jarrow and Rudd, 1982) yields more reliable results.
lower bound on the actual price and re-scaling prices with 4. Without the possession of the option in reality the increase
respect to this lower bound. An important advantage of this or decrease of an option value can be optimally simulated by
approach is that the lower bound can be based on econom- holding an amount of the underlying asset equal to the option
delta.
ic rationales and market experience. It also allows for the
5. For example, what is the return if prices increase from -10 to +10?
extreme positive outliers, while limiting the negative outliers
in prices. Moreover, this approach permits the usage of rela-
tively standard time-series models on the re-scaled price MICHAEL SEWALT is consultant at Deloitte & Touche’s
returns. Imbalance prices exhibit sudden jumps and similar Energy & Commodity Risk Management Centre of
complex interrelations as those in hourly spot prices: within Excellence. [email protected]
days and across days in both price levels and price volatility. CYRIEL DE JONG is professor at Erasmus University in
For a realistic model it is necessary to include both types of Rotterdam and consultant in Energy Risk Management
interrelations. We suggest that a combination of the period- and Derivative Valuation.
ic autoregressive model in Guthrie and Videbeck (2002) and [email protected]
specifications that model each time period separately
(Cuaresma et al., 2002) can achieve this goal. Literature
De Jong, C. and R. Huisman, 2003, "Option formulas for
Conclusions power prices with spikes", Energy & Power Risk
Options are helpful products for managing unexpected Management, February, Germany special issue
price and volume fluctuations. Given the high volatility of Jarrow, R. and A. Rudd, 1982, “Approximate option valua-
short-term electricity prices it could be expected that elec- tion for arbitrary stochastic processes”, Journal of
tricity options are very popular. However, currently bilateral Financial Economics, 10, 347-369.
options are traded in the OTC market only on a small scale, Guthrie, G. and S. Videbeck, 2002, “High-frequency elec-
and the exchange trade of options is even lower. The few tricity spot price dynamics: an intraday markets approach”,
options that are traded are mainly on forward contracts, research paper, New Zealand Institute for the study of com-
such as for example an option on the forward 2004. But also petition and regulation.
options on day-ahead spot prices (daily exercisable options) Cuaresma, J.C., J. Hlouskova, S. Kossmeier and M.
are traded occasionally. A major explanation for the low Obersteiner, 2002, “Forecasting electricity spot prices
trading volume is the difficulty to value those contracts. using linear univariate time series models”, working paper,
When current methods of option valuation are improved, the University of Vienna.