What drives the price differential between peak and base electricity prices? An empirical investigation for European market regions

Transcription

1 What drives the price differential between peak and base electricity prices? An empirical investigation for European market regions Enerday 2010, TU Dresden Dr. Hans-Günter Schwarz, RWE AG* Dr. Jochen Blessing, RWE AG* Dr. Christoph Lang, RWE Power AG Volker Liebenau, Universität Paderborn * Corresponding authors, h-guenter.schwarz@rwe.com, jochen.blessing@rwe.com Schwarz et al Enerday page 1

2 Content Motivation Objectives and main results Price volatility indicators Base-peak spread drivers: theory and model parameterisation Model design and data Empirical results and in-sample forecast Schwarz et al Enerday page 2

3 Motivation > Besides the power price level itself, shape of hourly electricity price curve (HPFC) heavily important for assessing utilisation and profitability of power plants, esp. of peaking plants. > Describing an HPFC by the relative price differential between highdemand and low demand hours (base-peak spread) alone might be misleading. Also price volatility within peak and off-peak demand hours matters for the shape. > Shape of curve might change considerably in the next years in Europe. > An empirical assessment of the link of the profile of electricity prices to supply and demand side factors remains surprisingly under-researched so far. Schwarz et al Enerday page 3

4 Objectives > Setup an empirical model that detects main drivers on supply and demand side of the base-peak spread (bps) and price volatilities (volas). > Test empirically whether main drivers relate to the bps and volas in functionally the same way for different countries / market regions. > Assess the quantitative importance of each driver for the overall evolvement of the structure of the HPFC. > Perform an in-sample forecast to check the quality of the model and its predictive power. Schwarz et al Enerday page 4

5 Main results > Supply-side factors matter most in explaining the base-peak spread. > Concentration of the power plant portfolio according to fuel type is key in explaining the differential in prices, both in size and significance. > The gas-coal delta matters only in some market regions. > The bps turns out to be the main driver behind volas in hours of low and high demand. > Demand volatility itself has had no empirical measurable impact on the bps and price volas. Residual demand (demand net of wind production) might have, but series are too short for a proper empirical assessment. Schwarz et al Enerday page 5

6 Electricity price volatility indicators Base peak spread > Most common price volatility indicator because future products exist > Defined as %difference between high demand hours (DE, 8am -20pm on workdays) and the average price > Describes the inter price volatility of price bands (peak and off peak) (Normalised) standard deviation peak and off peak > Defined as standard deviation with prices normalised to a base price of 1 > Describes the intra price volatility of price bands > Standard deviation of base price as result of those both Schwarz et al Enerday page 6

7 Base-peak spread and standard deviations in more detail EUR/MWh Standard deviation peak prices Average daily price (base price) Base-peak spread Standard deviation prices 00:00 08:00 20:00 23:00 Off-peak demand Peak demand Off-peak demand Schwarz et al Enerday page 7

8 Drivers of price volatility: theory p demand d1 peak demand d2 s Very specific conditions in the electricity market > low storage possibilities p2 > binding short-run capacities p1 > price-inelastic short-run demand which is quite volatile during day, week and year (peak definition DE: workday, 8am- 20pm) The flatter the supply (s) slope and the larger the demand (d) volatility (wider gap between d1 and d2) is, the higher is the price (p) differential (gap between p1 and p2), the higher is the price volatility q > necessity for anytime demand-supply balance Therefore: Price volatility a result of the slope of the supply (merit order) curve and demand volatility Schwarz et al Enerday page 8

9 Supply side / cost drivers behind the volatility > Concentration of conventional generation portfolio (s pivots) The higher is the concentration of the portfolio, the more demand fluctuates about the same fuel-technology along the merit order curve leaving short run marginal cost of generation mainly unchanged under high/low demand. Thus the more closely peak and prices are together and thus the lower the bps. > Reserve margin (s becomes increasingly inelastic/elastic) The higher the reserve margin, the less will peak demand fluctuations bring other more expensive - fuel generation technologies online, thus the lower the bps. > CO2-price (s pivots) The higher the CO2-price the more similar become short run marginal costs of coal and gas fired power plants in the merit order, thus the lower the bps when demand changes. > Gas-Coal Spread (s pivots) The higher the price delta between fuel prices gas and coal (expressed in same currency), the more short run marginal costs of neighbouring technologies in the merit order will differ (for a given CO2 price), thus the higher is the price differential between peak and -hours when demand changes. Schwarz et al Enerday page 9

10 Demand side and other drivers > Peak and off-peak demand volatility (d shifts) Demand has highly recurring pattern mainly determined by daily/weekly/seasonal factors. Also vola differs for demand within high-demand and within low-demand time spans. For a given supply stack, vola in high and low demand hours impact the bps and price volatility. > Country-specific add factor (s and d shift) Institutional factors (Market Design, Market Coupling and interconnector capacity, Demand side management). Captured in catch-all factor cross-section fixed effect. Schwarz et al Enerday page 10

11 Drivers: operationalisation and expected impact Driver Measured by Expected effect on bps Expected effect on volas Concentration of conv. gen. Portfolio Hirschman-Herfindahl index based on squared shares of conventional generation capacities (nuclear, coal, gas, oil) in total conventional generation capacity Reserve margin (conventional capacity + flexible hydro capacity peak demand) / peak demand CO 2 price CO 2 price in EUR/MWh Gas-Coal Spread gas price coal price in EUR/MWh (evaluated at lower heating values) Peak and off-peak demand vola standard deviation of hourly electricity demand in MWh/h for peak and off-peak demand time spans Market-specific add factor country-fixed effects Schwarz et al Enerday page 11

12 Model design for bps > Dynamic panel regression model used, which allows to assess common effects across different observation units and through time (see Wooldridge 2003, 2002, Hsiao 2003, Baltagi 2005, Greene 2007) > Yearly data from 2000 to 2009 considered > Seven market regions i pooled: Central West Europe (CWE), United Kingdom (UK), North Pool (NP), Italy (IT), Spain (ES), Greece (GR), Romania (RO > Panel equation for market i with cross-section fixed effect is: bps + δ CO 2, t = α + α i + β hhi + γ reservem peak peak + η demvola + η demvola + λ i ( gas coal ) + ε + K + K (1) t 2000,..., 2009, i = { CWE, UK, NP, IT, ES, GR, RO } Schwarz et al Enerday page 12

13 Model design for price band volas > Own panel model set up to explain price-band volatility. Observation that higher prices come along higher volatility (log-normal nature of prices) translates to expected positive link between bsp and price band volatilities. > Demand volatility directly impacts price volatility. > Regression equations for price volatilities in market region i are σ σ σ peak base = χ = χ peak + κ i, t i, t peak 2 2 ( σ ) + (1 υ) ( σ ) ( ) 2 2 υ peak + κ bps + τ bps peak + τ t 2000,...,2009, υ = 3132 /8760 i, t demvola 0.5 peak i, t demvola + ω peak σ i, t 1 + ω peak + υ peak i, t σ i, t 1 + υ i, t (2) (3) (4) > For the in-sample forecast later on, (1) is used to predict the bps and plugged into (2) - (4) to forecast volatilities for base-/peak/ prices. Schwarz et al Enerday page 13

14 Considered power markets market / country CWE (AT, BE, CH, DE, FR, LU, NL) NP (DK, FI, NO, SE) ES GB GR IT RO source of hourly price EEX NordPool Nordic Power Exchange OMEL APX HTSO GME OPCOM period Schwarz et al Enerday page 14

15 Data input drivers Concentration of conv. gen. portfolio Reserve margin CO2 price Gas-Coal Spread Peak and off-peak demand vola. Country-specific add factor source CERA (European Power Watch Data Center) CERA, EURPROG EEX ARA API#2 for coal, BAFA gas index ENTSO-E, National Grid UK, Nordic Power Exchange constant of regression period Schwarz et al Enerday page 15

16 Regressions results for the bps Dependent variable: BPS_? Coefficient Std. Error t-statistic Prob. C HHI_? RESERVEM_? CO2_PRICE_? STDEV_DEM_PEAK_? STDEV_DEM_OFFPEAK_? CWE--GAS_COAL_SPREAD_CWE UK--GAS_COAL_SPREAD_UK NP--GAS_COAL_SPREAD_NP IT--GAS_COAL_SPREAD_IT ES--GAS_COAL_SPREAD_ES RO--GAS_COAL_SPREAD_RO GR--GAS_COAL_SPREAD_GR Sample: , yearly values. Included observations: 56. Adjusted R 2 =.87. Prob(F- Statistic)=0.00. Durbin-Watson: Schwarz et al Enerday page 16

17 Regression results for price band volatilities Dependent variable: STDEV_PEAK_? Coefficient Std. Error t-statistic Prob. C BPS_? STDEV_DEM_PEAK_? STDEV_PEAK_?(-1) Sample: , yearly values. Included observations: 51. Adjusted R 2 =.55. Prob(F- Statistic)=0.00. Durbin-Watson: Dependent variable: STDEV_OFFPEAK_? Coefficient Std. Error t-statistic Prob. C BPS_? STDEV_DEM_OFFPEAK_? STDEV_OFFPEAK_?(-1) Sample: , yearly values. Included observations: 51. Adjusted R 2 =.08. Prob(F- Statistic)=0.07. Durbin-Watson: Schwarz et al Enerday page 17

18 Historical out-of-sample exercise Central Europe United Kingdom 50% 36% 48% 34% 46% 44% 32% 42% 30% bps 40% bps 28% 38% 26% 36% 34% 24% 32% 22% 30% % CE Nordpool CE_forecast UK Italy UK_forecast 16% 50% 14% 12% 45% 10% 40% bps 8% bps 6% 35% 4% 30% 2% 0% % NP NP_forecast IT IT_forecast Setup: Sample runs until 2007 and out-of-sample forecast is performed for 2008 and Schwarz et al Enerday page 18

19 Disclaimer > The results presented here are in sole responsibility of the authors. They do not necessarily represent the view of RWE AG. Schwarz et al Enerday page 19