FORWARD MARKETS AND THEIR INTERACTIONS WITH SPOT MARKETS AN EMPIRICAL ANALYSIS OF THE LIBERALISED EUROPEAN ELECTRICITY MARKET

Transcription

1 FORWARD MARKETS AND THEIR INTERACTIONS WITH SPOT MARKETS AN EMPIRICAL ANALYSIS OF THE LIBERALISED EUROPEAN ELECTRICITY MARKET Christian Redl 1, Vienna University of Technology Reinhard Haas, Vienna University of Technology Claus Huber, EGL Austria GmbH and Vienna University of Technology Abstract Since the liberalisation of the European electricity sector, forward and futures contracts have gained significant interest of market participants due to risk management reasons. For pricing of these contracts an important fact concerns the non-storability of electricity. The classical cost of carry approach therefore cannot be used to determine a relationship between spot and futures prices. Instead, a second theory on futures pricing relates forward prices to expected spot prices which are built on fundamental market expectations. In the following article the crucial impact parameters of forward electricity prices and the relationship between futures and spot prices will be assessed by an empirical analysis of two of the biggest European electricity exchanges: The European Energy Exchange and the Nord Pool Power Exchange. Introduction The goal of various, worldwide liberalisation efforts of the electricity supply industry was the introduction of competition as precondition for an efficient energy supply. In Europe, the liberalisation process dates back to the 1990s when the first electricity directive concerning common rules for the internal market in electricity was adopted in Besides introduction of competition, the creation of a common European electricity market was a major objective of this directive (EC, 1997). Along with the introduction of competition, a transition from a cost based price regulation towards a market orientated price formation took place and market platforms for short- and long-term trading of electricity were established. In these power markets, the wholesale price is determined by the generation costs of the marginal technology (i.e. the SRMC of the most expensive plant needed to meet demand merit order principle). In this new, competitive environment risks emerged for market participants unknown in the previous regulated area. Longterm contracts like futures or forwards, traded both at power exchanges and bilaterally over-the-counter, allow for management of the price risk by effectively locking in a fixed price and therefore avoiding uncertain future spot prices. Hence, the determination of the relevant influence factors on the price formation on long-term electricity markets is of great interest. Both econometric as well as fundamental marginal cost models are used in this paper to conduct quantitative and qualitative analyses of futures and forward prices at the European Energy Exchange (EEX) and the Nord Pool Power Exchange. Interesting properties arise from the non-storability of electricity which will be considered by having a closer look at the relationship between spot and forward prices. It proceeds as follows: The next section focuses on year-ahead futures and an econometric analysis of these prices. Section 3 studies the relationship between spot and forward prices. Finally, section 4 concludes. Econometric analysis of year-ahead electricity prices Figure 1 depicts the price evolution of spot and year-ahead baseload electricity prices at the EEX and Nord Pool power exchanges from January 2003 to June Spot and forward prices were rising continuously from early 2003 to mid 2006 at the EEX. The EEX is the leading exchange in the western/central European sub market and since this market is dominated by fossil fuelled power plants, increasing power prices reflected rising primary energy prices (especially for natural gas). The highest increases could be 1 Energy Economics Group, Vienna University of Technology; Gusshausstrasse 25/E373-2, A-1040 Vienna, Austria; Tel: ; redl@eeg.tuwien.ac.at; Web:

2 observed during 2005 due to the commencement of the European emission trading scheme (Prices for CO2 emission allowances started trading from 8 EUR/t CO2 and rose dramatically during 2005 peaking several times at 30 EUR/t CO2). Spot prices at the EEX were falling, with a short exception, from March 2006 on mainly due to a massive drop-off in emission allowance prices. Still, year-ahead prices have maintained their high level from early 2006 because of high gas and later on high (2008-) forward emission allowance prices [EUR/MWh] Spot_EEX Year-ahead_EEX Spot_Nord Pool Year-ahead_Nord Pool Jan 03 Apr 03 Jul 03 Okt 03 Jan 04 Apr 04 Jul 04 Okt 04 Jan 05 Apr 05 Jul 05 Okt 05 Jan 06 Apr 06 Jul 06 Okt 06 Jan 07 Apr 07 Figure 1. Evolution of spot and forward prices for baseload electricity at EEX and Nord Pool. Source: EEX, Nord Pool Nord Pool prices follow a somewhat different pattern. The Scandinavian power market is mainly characterised by hydro and nuclear generation with 74% of total generation in 2006 stemming from these two generation sources whereas hydro and nuclear generation corresponds to about 60% in the Central European sub market. 2 Clearly, also fossil fuelled plants stay at the margin in Scandinavia though in times of high hydro generation only highly efficient plants are needed to satisfy demand resulting in lower pool prices. Still, low hydro availability as e.g. in early 2003, implying congested transmission grids, and increased generation in inefficient thermal power plants cause prices soaring above EEX levels. Compared to spot prices, year-ahead forward prices follow a less volatile regime for both markets whereas Nordic prices are generally lower than their EEX counterparts due to different power plant park structures. At first sight, spot and forward prices show a higher correlation in the Scandinavian market most likely due to the high amount of hydro storage capacity especially in Norway (see chapter 3 for a more thorough analysis). Figure 2 shows crucial influence parameters for continental European forward electricity prices, namely futures prices for hard coal (North-western Europe port prices ARA ports), natural gas (Zeebrugge hub), CO2 allowances and year-ahead baseload futures traded at the EEX. Additionally, German import border prices for natural gas and hard coal published by the German Federal Office of Economics and Export Control (BAFA) are also shown in this figure (BAFA prices are average prices of current deliveries hence, no year-ahead deliveries are considered in the import prices). A positive relationship between CO2, natural gas and electricity futures as well as no correlation between stable coal and rising electricity futures quotations can be observed. Therefore the influence of the gas price on the electricity price prevails also in baseload. 2 See Nordel (2007) and UCTE (2007) on detailed statistics about the considered markets. Redl, C., Haas, R., Huber, C. 2

3 [EUR/MWh], [EUR/tCO2] Dez 04 Feb 05 Apr 05 Jun 05 Aug 05 Okt 05 Dez 05 Feb 06 Apr 06 Jun 06 Aug 06 Okt 06 Dez 06 Feb 07 Apr 07 Jun [EUR/MWh] Year-ahead coal Coal_BAFA Year-ahead gas Gas_BAFA Year-ahead EUA Year-ahead Baseload EEX Figure 2. Monthly averages of year-ahead coal, gas and CO2 emission allowance prices and German border prices (left scale) vs. EEX year-ahead baseload futures (right scale). Source: EEX, BAFA In competitive markets, marginal generation costs are relevant for price formation. Due to the dominance of fossil fuelled power plants in the Central European power market, generation costs of these technologies crucially determine electricity prices. Hence, (short run) year-ahead marginal costs of conventional thermal power plants are calculated by formula (1) using input data shown in Figure 2: pprim pco2 f CO2 SRMC = + (1) η η where SRMC are short run marginal costs [EUR/MWh], p PRIM are primary energy prices [EUR/MWh], p CO2 are CO2 emission allowance prices [EUR/t CO2], f CO2 is the CO2-emission factor of the fuel [t CO2/MWhprimary], and η is the efficiency of the plant. Year-ahead electricity futures traded at the EEX show a higher correlation with generation costs of gas-fired (CCGT) plants compared to coal-fired power stations. The comparison of year-ahead marginal costs therefore proves the higher influence of the CCGT technology on the electricity price (see Table 1). However, EEX year-ahead prices also show a high correlation with the moving average of last 12 months spot prices indicating pronounced adaptive expectation formation behaviour of market participants on the futures market where the spot price average serves as an estimator of year-ahead electricity prices. A slightly weaker link between forward prices and generation costs of fossil fuelled plants prevails in the Nordic market though correlation coefficients between prices and CCGT generation costs are still high (see Table 1). Again, a strong adaptive price formation component can also be observed in this market with a strong link, opposed to the EEX market, between forward and current spot prices. As mentioned earlier, the high amount of reservoirs in the Nord Pool area can serve as an explanation. Redl, C., Haas, R., Huber, C. 3

4 Table 1. Correlation coefficients between EEX and Nord Pool year-ahead baseload prices and explanatory variables from December 2004 to June Correlation coefficient EEX Correlation coefficient Nord Pool Year-ahead Base / SRMC CCGT_Zeebrugge Year-ahead Base / SRMC HC_ARA Year-ahead Base / SRMC CCGT_BAFA Year-ahead Base / SRMC HC_BAFA Year-ahead Base / Spot Year-ahead Base / Spot 12 month average To quantitatively assess the influence of the relevant parameters an econometric analysis is performed by testing several regression models: LnYearAhead n Base( t, T ) = b1 + bi X i t, T ) i= 2 ( (2) where YearAhead Base(t,T) is the monthly front year average for baseload on the EEX and Nord Pool futures market respectively and X i(t,t) are explanatory variables. SRMC, determined from input prices shown in Figure 2, represent fundamental explanatory variables to the model. 3 Both the actual spot price as well as the historic moving 12 month spot market average constitute independent variables to model adaptive behaviour of market participants (see also correlation coefficients in Table 1): i= t 1 1 i= t 12 Spotaverag e t = Spot i (3) 12 Comparison of all independent variables with futures prices let us to anticipate a strong influence of year-ahead CCGT generation costs and moving 12 month average spot market prices on EEX futures prices. A simple linear regression model (4) is applied to test this hypothesis: LnYearAhea d = b + b LnSRMC + b LnSpotaverage (4) Base( t, T ) 1 2 CCGT ( t, T ) 3 ( t) Table 2 shows the results of the econometric model for monthly baseload futures prices at the EEX for the period December 2004 to June Marginal year-ahead generation costs of CCGT plants as well as the historical yearly spot market average provide a good explanation of the year-ahead electricity price whereas the spot market exerts a more dominant influence on the futures market compared to the CCGT generation costs. This result indicates a pronounced adaptive expectation formation on the futures market. Still, certain statistics (DW and ADF) reveal methodical problems. Table 2. Results of regression analysis (4) and (5) for Ln Year-ahead baseload futures at the EEX (t-statistics in brackets). (4) n=31 (5) n=30 b 1 (constant term) 1.00 (4.31) 0.19 (0.96) b 2 (Year-ahead SRMC CCGT ) 0.07 (1.16) b 2 (SRMC CCGT_BAFA ) 0.72 (6.05) b 3 (Spot average) 0.69 (10.41) 0.26 (3.03) R 2 (R 2 corr) 0.87 (0.86) 0.95 (0.95) DW ADF (95% critical value) (-4.05) (-4.05) 3 Compared to primary energy prices, SRMC are relevant for price formation on wholesale markets. Besides, the problem of multicollinearity in econometric analyses can be avoided as gas and CO2 prices show a linear relationship amongst them. Redl, C., Haas, R., Huber, C. 4

5 Especially since the introduction of the EU-ETS, CCGT plants are commonly used also in baseload generation. However, assuming CCGT as the marginal technology, if CCGT operators offer baseload products on the futures market and use Zeebrugge gas prices for their calculations, futures prices are up to 20 EUR/MWh lower than generation costs. Still, besides opportunity reasons, Zeebrugge gas futures prices not necessarily need to be the relevant primary price benchmark for CCGT plants. To test this hypothesis, another simple linear regression model (5) is considered using CCGT generation costs calculated from current BAFA gas prices and (yearly) average spot market prices as explanatory variables: LnYearAhea d = b + b LnSRMC + b LnSpotaverage Base( t, T ) 1 2 CCGT _ BAFA( t) 3 ( t) (5) In Table 2 the results of this econometric model for monthly baseload futures prices at the EEX for the period January 2005 to December 2006 are shown. Marginal generation costs of CCGT plants calculated with border prices as well as the moving 12 month spot market average provide a very good explanation of the year-ahead electricity price. The quality of the regression is increased compared to model (4). Moreover, the CCGT generation costs exert a more pronounced influence on the electricity futures compared to the moving spot market average. Nevertheless, regression coefficient b 3 is significantly different from zero. The stronger influence of current gas border prices on the electricity futures price compared to year-ahead gas prices is consistent with a lacking liberalisation of the European gas sector. Gas supply is still characterised by long-term contracts ( Take or pay contracts) (Maisonnier, 2006). Therefore, as possibilities to resell gas from these long-term contracts at the hub are limited, Zeebrugge prices do not reflect procurement costs of natural gas for CCGT plant operators correctly. Nevertheless, the border prices do not contain year-ahead deliveries. Similarly, Nord Pool s year-ahead forward prices also show a strong dependence on CCGT generation costs and moving spot market averages (see Table 3). Table 3. Results of regression analysis (4) and (5) for Ln Year-ahead baseload futures at Nord Pool (t-statistics in brackets). (4) n=31 (5) n=30 b 1 (constant term) (-4.15) (-1.88) b 2 (Year-ahead SRMC CCGT ) 0.60 (12.15) b 2 (SRMC CCGT_BAFA ) 0.94 (8.45) b 3 (Spot average) 0.71 (12.77) 0.23 (2.66) R 2 (R 2 corr) 0.91 (0.90) 0.84 (0.82) DW ADF (95% critical value) (-4.05) (-4.06) Relationship between spot and forward prices Along with the liberalisation of the electricity supply industry and the introduction of competition the necessity has arisen to establish market places for power. Consequently, electricity became a tradable commodity. However, certain characteristics of power differ significantly from other, more traditional goods, namely the necessity of an exact match of supply and demand which is amplified by electricity being virtually a non-storable commodity. Especially since electricity cannot be stored economically the cost of carry approach, normally used in the area of finance and traditional commodity markets, cannot be used to determine a no-arbitrage condition between spot and futures prices: F ( t, T ) ( r+ s)( T t) = Ste (6) where F (t,t) is the current futures price for delivery in T, S t is the current spot price, r is the interest rate and s are storage costs. 4 4 See e.g.skantze and Ilic (2000) on arbitrage pricing. Redl, C., Haas, R., Huber, C. 5

6 Instead, a second theory of futures pricing relates prices of long-term contracts to expected spot prices. The latter are built on fundamental expectations of market participants. Differences between these prices indicate risk or forward premiums which are a compensation for bearing the spot price risk (Longstaff and Wang, 2002). Finally, contracts are discounted to account for opportunity costs. Ex-post determined differences between forward prices in the trading period and spot prices in the delivery period can be used to assess whether the futures price represents a good predictor of the future spot price: F( t, T ) ST Forecaster ror = (7) T S 1 1 T 1 t = 1 T These ex-post determined errors reveal considerable misjudgement of future conditions by market participants. The reason behind this is a deviation between actual generation conditions and fundamental generation conditions due to shocks between futures and spot trades like unexpected cold or warm weather, high or low CO2 prices, high or low hydro availability, etc. in the delivery period (see Table 4). If one looks at each contract separately, the forecast error for peak load shows for every contract the highest value, whereas for off-peak errors shows the lowest value for every contract. Due to a higher slope of the supply curve when approaching system capacity, clearly, forecast errors are more strongly affected in peak load compared to off-peak. Monthly futures, where trading at Nord Pool and EEX starts six months ahead of the delivery period, show significant forecast errors during the analysed sample which comprised contracts with monthly delivery periods from January 2005 to July 2007 (see Figure 3 and Table 4). 140% 120% 100% 80% rel. forecast error 60% 40% 20% EEX Nord Pool 0% -20% Jan 05 Mrz 05 Mai 05 Jul 05 Sep 05 Nov 05 Jan 06 Mrz 06 Mai 06 Jul 06 Sep 06 Nov 06 Jan 07 Mrz 07 Mai 07 Jul 07-40% Figure 3. Forecast errors of monthly futures traded at the EEX and Nord Pool with respect to the actual spot price during the delivery period. Source: EEX, Nord Pool, own calculations Redl, C., Haas, R., Huber, C. 6

7 Table 4. Forecast errors of various futures contracts traded at EEX and Nord Pool. Delivery period Forecast error baseload Forecast error peak load Forecast error off-peak EEX Nord Pool EEX Nord Pool EEX Nord Pool Cal-04-10% -21% 14% -44% Cal-05-37% -11% -23% -59% Cal-06-31% -41% -20% -51% Q % Q % Q % Q % -30% -48% Q % -24% -27% -50% Q2-06 0% -31% 16% -25% Q % -48% -18% -31% Q % -15% 22% -9% Q % 71% 119% 35% Q % Jan 05 22% 50% 49% -15% Feb 05-7% 27% 9% -30% Mrz 05-25% -4% -15% -38% Apr 05-19% -15% -4% -39% Mai 05-16% -18% -2% -35% Jun 05-25% -1% -20% -35% Jul 05-19% -15% -6% -37% Aug 05 5% -6% 20% -17% Sep 05-9% 13% 2% -26% Okt 05-5% 8% 9% -24% Nov 05-31% 24% -30% -33% Dez 05-21% 15% -16% -30% Jan 06-19% 3% -11% -31% Feb 06-18% -5% -6% -38% Mrz 06-12% -25% 2% -32% Apr 06 17% -22% 38% -12% Mai 06 40% 11% 54% 19% Jun 06 28% -3% 48% -4% Jul 06-30% -12% -30% -30% Aug 06 15% -26% 38% -18% Sep 06 19% -15% 38% -10% Okt 06 26% 5% 46% -7% Nov 06 26% 28% 38% 3% Dez 06 60% 89% 87% 13% Jan % 127% 142% 57% Feb % 100% 153% 31% Mrz % 93% 170% 21% Apr 07 31% 62% 55% -5% Mai 07 5% 36% 17% -16% Jun 07 13% 9% 27% -16% Jul 07 54% 34% 100% -26% Redl, C., Haas, R., Huber, C. 7

8 Table 5 summarises some additional statistics of the forecast errors of monthly futures. On average, baseload contracts were traded 14% and 18% above actual spot prices in the delivery periods of the futures and forwards at EEX and Nord Pool respectively. On the EEX market, forecast errors for peak load even amounted to 32%. Calculating the off-peak component out of baseload and peak load contracts yielded and error of -15%. Forecast errors are significantly different from zero for a double-sided test. Errors for baseload and peak load are significantly larger than zero at the two exchanges whereas the virtual off-peak EEX contract is characterised by a significant negative error. Table 5. Summary statistics of forecast errors for monthly futures with delivery from January 2005 to July 2007 traded at EEX and Nord Pool. EEX Baseload EEX Peak load EEX Off-peak Nord Pool Baseload Mean 14% 32% -15% 18% Standard dev. 40% 52% 23% 40% Minimum -31% -30% -39% -26% Maximum 112% 170% 57% 127% t-statistic 1,94* 3,47** -3,60** 2,58** */** denotes significance at the 0.1 and 0.05 level for the double-sided test Considering perfect foresight, the identified differences in spot and forward prices would indicate the existence of pronounced (mainly positive) risk premiums in the forward markets. 5 Clearly, perfect foresight of market participants is a very strong assumption in many cases not correctly reflecting real expectation formation. However, the considered sample of futures contracts may still imply that dominating motives of certain market participants contribute at least partly to a certain sign and, hence, bias of the forecast error. Bunn (2006) identifies positive risk premiums for peak hours when comparing the day ahead and prompt market and the week ahead and day ahead market. On the other hand risk premiums for off-peak hours are negative when comparing the day ahead and prompt market and the week ahead and day ahead market. He argues, that during peak hours the demand side has a higher willingness to pay day ahead in order to avoid high volatility in the intra-day market whereas during offpeak hours the supply side behaves more competitively in the day ahead market in order to avoid shutdown and start-up costs. Similarly, e.g. Hadsell and Shawky (2006) and Douglas and Popova (2006) find positive risk premiums in long-term electricity markets. It is worth to mention that in their nomenclature long-term markets are day-ahead markets and short term markets are real time markets. Yet, many papers compare day ahead with intra-day markets and attribute price differences to risk avers behaviour of market participants. However, as Bunn and Karakatsani (2003) put it, the arguments for absence of price convergence between day-ahead and real-time markets still underestimates the fact that different timing of the two markets implies different information uncertainties and plant flexibility requirements, which are converted to costs. Price differences may, therefore, be attributed mainly to shocks between futures and spot trades (e.g. unexpected cold or warm weather, high or low CO2 prices, etc. in the delivery period, etc.) though persistent differences between off-peak and peak premiums can be explained, in part, by the special form of the supply curve in electricity markets and competitiveness reasons. However, the caveats of a limited sample size still apply calling for further analyses. Conclusions Electricity forward prices at the European Energy Exchange and the Nord Pool power exchange have been analysed within this paper. Econometric analyses show high influences of generation costs of CCGT plants and moving 12 month spot market price averages on year-ahead electricity futures which, on the one hand, indicates both adaptive expectation behaviour, but also shows a stronger influence of the CCGT technology on the electricity futures price compared to hard coal power plants. 5 Compare also Gjolberg and Johnsen (2001) and Botterud et al. (2002) on the existence of positive risk premiums in the Nordic market. Redl, C., Haas, R., Huber, C. 8

9 Due to lacking storage possibilities, no exact relationship between current spot and forward prices can be formulated. Instead, when assessing the pricing of futures and forward contracts the starting point is the expected spot price in the delivery period. Hence, forward prices are built on fundamental expectations of market participants, updated by applying risk or forward premiums and, finally, discounted to account for opportunity costs. Ex-post determined differences between forward prices in the trading period and spot prices in the delivery period reveal considerable misjudgement of future conditions by market participants due to a deviation between actual generation conditions and fundamental generation conditions. Biased forecasts result from shocks between futures and spot trades (e.g. unexpected cold or warm weather, high or low CO2 prices, high or low hydro availability, etc.). Nevertheless, significant positive signs of the forecast errors for baseload and peak load contracts may also arise from dominating motives of certain market participants. Further research is indicated to clarify this issue. References Botterud, A., Bhattacharyya, A., Ilic, M. (2002), "Futures and spot prices an analysis of the Scandinavian electricity market". Bunn, D.W. (2006), Risk and electricity price dynamics, Platts.com News Feature. Bunn, D.W. and Karakatsani, N. (2003), "Forecasting Electricity Prices," London Business School, London. Douglas, S., and Popova, J. (2006), "Storage and the Electricity Forward Premium" 26th USAEE/IAEE North American Conference, Ann Arbor, Michigan, September 24-27, 2006, Proceedings. EC (1997), Directive 96/92/EC of the European Parliament and of the Council concerning common rules for the internal market in electricity, L027, Brussels. Gjolberg, O., and Johnsen, T. (2001), "Electricity Futures: Inventories and Price Relationships at Nord Pool" Discussion Paper #D-16/2001. Hadsell, L., and Shawky, H.A. (2006), "Electricity Price Volatility and the Marginal Cost of Congestion: An Empirical Study of Peak Hours on the NYISO Market, " The Energy Journal, 27-2, Longstaff, F.A., and Wang, A. (2002), "Electricity Forward Prices: A High-Frequency Empirical Analysis" Working paper, University of California Los Angeles. Maisonnier, G. (2006), "The Ties between Natural Gas and Oil Prices" IAEE Newsletter, 15, Third Quarter, Nordel (2007), Annual statistics 2006, Helsinki. Skantze, P., and Ilic, M. (2000), "The Joint Dynamics of Electricity Spot and Forward Markets: Implications on Formulating Dynamic Hedging Strategies" Energy Laboratory, Massachusetts Institute of Technology, MIT EL , Cambridge. UCTE (2007), Online data available at Acknowledgments The presented analyses have been financed by the Vienna Science and Technology Fund (WWTF). Of course views expressed here are under the sole responsibility of the authors. The authors are grateful for valuable discussions to Bernhard Böhm. However, the authors are solely responsible for the expressed views and any remaining errors. We also acknowledge Nord Pool for providing price data. Redl, C., Haas, R., Huber, C. 9