On the integration of the natural gas market in the
European Union
Groningen, October 2010
Msc-Thesis - international economics & business University of Groningen
Faculty of economics and business
By: Rudolph Harmsen
Supervisor: Prof.dr.mr. C. J. Jepma Co-assessor: Dr. P. Rao Sahib
Abstract
Table of contents
1 Introduction ... 4
2 The natural gas market structure in the EU... 7
2.1 The traditional EU natural gas market ... 7
2.2 The reform... 7
2.3 Organized trading markets ... 8
3 Literature survey ... 10
4 Research framework... 13
5 Methodology ... 16
5.1 Descriptive statistics... 16
5.2 Cointegration ... 17
5.3 Vector error correction model ... 20
6 Empirical results... 23
6.1 Results of the Engle & Granger cointegration test... 23
6.2 Results of the VEC model ... 25
7 Conclusion... 28
References ... 32
Appendix A: Information on gas hubs ... 35
1 Introduction
Until after World War ІІ natural gas was routinely found to be a worthless byproduct of oil production. However, in the second half of the 20th century world demand for natural gas rose steadily. It is now used extensively worldwide in power plants, residential heating, and industries. Natural gas consists of a mixture of hydrocarbon gases, primarily methane. It is highly combustible and emits lower amounts of carbon dioxide per unit energy than other fossil fuels (NaturalGas.org, 2010). By far the world’s largest natural gas reserves are found in Russia, Iran, and the Middle East (U.S. Energy information administration, 2010).
For industrialized countries, like those of the European Union (EU), natural gas has become an important aspect of their energy market. In 2008 natural gas accounted for 25% of the total gross energy consumption in the EU. The actual value chain of the EU natural gas industry extends beyond its borders. Consequently, the EU is highly dependent on gas imports, although this dependency varies per member state. In 2008 the EU imported 62% of its gas consumed (Eurostat, 2010). Western Europe imports mostly from Norway, North Africa and South America (in the form of LNG). Southern Europe imports mainly from North Africa. Finally, gas imports to Central and Eastern Europe are dominated by gas from the Russian Federation. The import dependency of the EU is likely to increase in the near future since consumption increases and internal extraction decreases. Moreover, upstream players such as Russia and Algeria no longer rely solely on gas deliveries to Europe. The globalization of the natural gas market through the introduction of LNG trade and upcoming markets in Asia (most notably China and India) makes these players less dependent on Europe’s imports. Excluding Russia, Europe’s largest gas producers are Norway, the Netherlands, and the United Kingdom (U.S. Energy information administration, 2010).
countries. However, the import dependency of the EU as a whole does not decline. The liberalized single market should also lower end consumer prices and increase the overall welfare. Liberalization principles have been introduced to the former segmented markets to pursuit this full integration.
Given the intention of EU politicians to reform the natural gas industry one can ask whether this reform has been successfully so far. It has been stated earlier that the objective of the natural gas industry reform is the completion of the internal market. Therefore this paper addresses the following research question:
To what extent is there an integrated market for natural gas in the European Union?
Integration and liberalization are two strongly correlated processes. Mostly integration follows liberalization. The focus in this paper is however solely on the issue of integration and not on liberalization. The above research question raises several sub-questions. How is the natural gas industry in the EU organized? What defines an integrated market? How can economic integration be measured? The main research question, as well as these sub-questions, are addressed in the proceeding sections. To my best knowledge the data used in this research is unique. Therefore, the conclusion should shed some new light on the issue. Answering the main research question gives in the first place insight in the extent to how far EU natural gas reforms have been successful. In the second place it also contributes to a greater and increasing variety of papers and discussions on the economic integration of markets in general. As Taylor (2001) already observes: “[..] The study of international and regional markets, the choice of open- versus closed-economy models, and many other important issues for theory, empirics, and policy rest on what economists can say about the existence of one market or many”.
2 The natural gas market structure in the EU
2.1 The traditional EU natural gas market
In the pre-reform period the value chain of the West-European natural gas market consisted of three stages: suppliers (upstream), transmission companies (downstream), and consumers (Jepma et al., 2009). The upstream side was dominated by a few gas extracting companies (e.g. Statoil, Shell, ExxonMobil, BP and Gazprom). The transmission companies were large national monopolistic firms. These firms stood at the center of the chain, and typically had the exclusive right for signing bilateral import contracts with suppliers. Examples of such monopolies where the Gasunie (Netherlands), ENI (Italy), Gaz de France (France), and merger E.ON Ruhrgas AG (Germany). Typically these monopolies owned the transmission networks (the pipeline grid). Traditionally gas was traded as follows: gas producers maintained supply contracts with gas transmission firms, gas transmission firms signed supply contracts with distribution companies, and the distribution firms finally sold gas to end consumers. In this pre-liberalization period consumers had no choice between different suppliers of gas (Harris & Jackson, 2005). Throughout the value chain the trade was arranged in long term take-or-pay contracts. These type of contracts were most prevalent because of the large upstream investments in production and transmission systems. Contracts therefore lasted for approximately 15 to 30 years, thereby matching the duration of upstream investments. After the first oil crisis, prices of these gas contracts were linked to alternative forms of energy, usually some crude oil basket, to ensure the competitiveness of natural gas.
2.2 The reform
The reform of the EU natural gas market started back in the 1990’s. This reform is based mostly on three EU legislative acts between 1998 and 20051 (Haase, 2008). As stated in the introduction, integration of former segmented (national) markets into one single EU market is the main objective. The integration should lead to more competitiveness of the EU natural gas market, member state access to various sources of natural gas, and a choice for consumers between different suppliers resulting in lower prices (Cronshaw, Marstrand, Pirovska, Simmons, and Wempe, 2008). Liberalization principles such as, unbundling of activities, full market opening, and third-party-access to the network are key mechanisms in the integration progress. Effective regulation is also required for the integration. Not all member states have
implemented the EU principles to the same extend. Next to that, the initial degree of liberalization differed among member states in the first place. For example, in the UK the downstream industry has been privatized at an earlier stage. The reform has in most countries allowed third-party access, and consumers can nowadays freely choose from who they buy gas (Harris & Jackson, 2005). Efforts have also been made to create organized trading markets for natural gas. The organized trading market is most prevalent at the level where monopolistic transmission firms trade gas with retailers. Organized markets brings together buyers and sellers from a specific geographic region independent of country borders, and thus enhances competition. A (liquid) organized trading market provides equal opportunities for market players. It should ensure a secure supply of gas for retailers, enhancing entry of new retailers, and lowering end-consumer prices. Organized gas markets are a requirement for the market integration process (see for example European Commission, 2009 and Laczkó & Lajtai, 2009). Thus, organized markets create the opportunity for gas to be considered as a commodity that can be traded. The organized markets for natural gas are often referred to as (gas) hubs.
2.3 Organized trading markets
Gas hubs can be either virtual or physical. A virtual gas hub covers a whole area (entry-exit regime). In such a regime gas enters at an entry point, and exits the regime at an exit point. Gas that enters the regime can come from production sites, or other grids (imports). Gas that leaves the regime can flow to a regional transmission grid (consumers), industry, or to other grids (export) (Gas Transport Services, 2010). At a physical hub the market is defined at a single point of interconnecting pipelines. Despite their difference, both types of hubs share the important common characteristic that they facilitate the trading of natural gas. The trading at hubs is facilitated by an independent operator, the hub operator.
Liquidity characterizes the efficiency of a gas hub. Factors that determine liquidity are for example: volumes traded, the churn ratio2, differences between bids and offers, and the number of trading parties. Trading hubs are far more liquid in the United States and the United Kingdom than they are in continental Europe (Jepma et al., 2009). A much larger share of the gas traded in these former two countries is traded on hubs. Thus, despite efforts to create liquid trading hubs in the EU, the bulk of gas trades at the upstream level in the EU are still done via the long term take-or-pay contracts (European Commission, 2010). However, the number of transactions on European gas hubs has gained much importance over time. An example is the Dutch gas hub TTF. The total volume traded at this hub rose from 0.1 billion cubic meters in January 2003, to 9.2 billion cubic meters in March 2010 (Gas Transport Services 2, 2010). Similar patterns arise at other gas hubs in Europe. The volumes traded at European gas hubs rose by 44% in 2006, and by 33% in 2007 (European commission, 2009). The European commission (2010) also reports an increase in liquidity at European gas hubs in 2009. Figure 2 in appendix A shows the traded volume development at the Dutch gas hub TTF. Figure 3 in appendix A shows the traded volume development at the German gas hub NCG. A clear rising trend can be observed in both figures, indicating that indeed traded volumes at European gas hubs are rising. The British gas hub National Balancing Point (NBP) is by far the most liquid hub in the EU. See figure 4 in appendix A to get an impression on the size of different European gas hubs. The gas hubs Title Transfer Facility (TTF) from the Netherlands and Zeebrugge from Belgium are the most liquid continental European hubs. As the proceeding sections will make clear, gas hubs form a mechanism for market integration
2
3 Literature survey
There are multiple studies on the energy integration program of the EU. For example, Meeus & Belmans (2008) show that the electricity market in the EU is still not harmonized. There appear to be large wholesale price differences, and the interconnection capacity between countries is scarce. However, only few studies focus on the integration of the natural gas market. In this section I review the major studies on natural gas market integration. Some of these studies have used similar approaches to that of this paper.
the following hubs: NBP (UK), Zeebrugge (Belgium), and Bunde (Germany). The time period they analyze is between 2000 and 2005. They argue that organized markets have emerged due to the liberalization. According to them the relative policy question should therefore be whether these organized markets are integrated. The relationship between hubs is studied using the Kalman filter. The Kalman filter results in a time-varying coefficient describing the relation between prices at different points in time. Neumann et al. find that the organized markets between the United Kingdom (NBP) and Belgium (Zeebrugge) are well integrated for the time period under consideration. On the other hand they find that the market places between Belgium (Zeebrugge) and Germany (Bunde), show no signs of integration or convergence towards integration for the period 2000-2005. Bunn & Micola (2007) examine the effect of the Interconnector (a gas pipeline connecting Belgium and the UK) on prices of the organized trading markets NBP and Zeebrugge. As part of their analysis they examine market integration between NBP and Zeebrugge by performing a unit root test on their price gap. The findings of Bunn & Micola support market integration between the NBP and Zeebrugge hub.
There have also been studies that assess the degree at which there is an integrated world market for natural gas. L’Hégaret, Siliverstovs, Neumann, and Hirschhausen (2003) examine the extent of international market integration for natural gas between Europe, North America, and Japan. The Johansen cointegration test is performed on price series. They argue that LNG has led to an increasing integration of the natural gas markets. Importantly, the authors assume that the market for natural gas in Europe is an integrated one. Standard international import prices are used. In addition, the spot market price of the American Henry hub is included as well. Their analysis shows co-movements within the European/Japanese and North American gas prices.
4 Research framework
The main objective of the EU reform, as well as the focus of this paper, is the creation of one market for natural gas in the EU. Market integration beyond national borders is a phenomena that has become increasingly important in the globalizing world. For some securities and commodities there even is a wide notion that there exists an integrated world market. Integration of markets is often triggered by competition and trade. The economical model of this paper however begins with an examination of market definitions and characteristics.
There is an important question that should be addressed first: what defines an integrated market? On the one extreme there are perfectly integrated markets, whereas on the other extreme there are perfectly segmented markets. The problem is where and how to draw a line between what is an integrated and what not. There is no formal definition of an ‘integrated market’. Nevertheless, generally speaking an integrated market is a connection of former segmented markets into one market. A perfectly integrated market does in essence not function differently than a ‘normal’ single market. Like in a ‘normal’ single market, there will be comparable rules and terms throughout the integrated market. Therefore the definition of a market should also apply to a perfectly integrated market:
“The role of the market is to facilitate the making of exchanges between buyers and sellers” (Stigler & Sherwin, 1985)
A question that immediately rises from the above market definition is: what characterizes a (perfectly integrated) market? An answer to this can provide the empirical tools to assess the degree of market integration. Building upon work from Cournot and Jevons, neo-classical economist Alfred Marshall states about a market the following:
“[…] the more nearly perfect a market is, the stronger is the tendency for the same price to be paid for the same thing at the same time in all parts of the market […]” (Marshall, 1920).
like these are often said to be perfectly integrated. For these securities prices are kept at the same level in almost all stock exchanges throughout the world. If, for example, for a given security the price is raised in, lets say location A. Then almost immediately does the price for the same security tend to rise in location B and C as well. Any delay in price adjustment gives an opportunity to arbitrage. Arbitrage is the simultaneous buying and selling of the same good at different locations. Consequently, it can generate a riskless profit for an arbitrageur. Following standard demand and supply theory this mechanism forces the prices to move to the same level (equilibrium) everywhere. The theory of one price in one market is referred to as the law of one price (henceforth: LOP). If the LOP holds across markets then these markets are said to be integrated (Chen & Knez, 1995). The LOP only holds if the homogeneity assumption is satisfied. The same price is only paid in two locations if the product in question is exactly identical, or homogeneous. The absolute version of the LOP described above may however not hold in all cases. In his introduction on markets Marshall went on:
“[…] but of course if the market is large, allowance must be made for the expense of delivering the goods to different purchasers; each of whom must be supposed to pay in addition to the market price a special charge on account of delivery.”
The above citation refers to fundamental systematic price differences between markets. These price differences can be caused by for example delivery costs and differential taxes. A good example concerning natural gas are the transportation costs through the pipeline network. Trading with relative remote markets could lead to more transportation costs and thus higher prices in that remote market. A restricted pipeline network may as well increase the costs of transporting natural gas since pipeline access might be scarce. It is said that the relative version of the LOP holds if there are only certain systematic price differences between locations
presence of capacity constrains between two locations, the LOP between these two locations can still hold. A third location, trading with these two locations separately, can cause the LOP condition between all three locations to hold (see Cuddington & Wang, 2006 p. 203 for a mathematical clarification).
The theory above has provided an economic tool to assess the degree of market integration. The problem this study faces is to determine the geographical domain over which the LOP holds for the natural gas market in the EU. The LOP allows one to formulate the market integration measurement problem as a study of price relationships According to the theory above, there can be spoken of an integrated market between two locations if: (1) two locations charge an equal price, or (2) if there is an equilibrium price relation between these two locations. The first statement refers to the absolute version of the LOP. The second statement refers to the relative version of the LOP. For this study the existence of the relative version of the LOP is more important since one expects transportation costs and other systematic price differences in the natural gas industry. The hypothesis that need to be tested to determine whether different geographical locations form an integrated market for natural gas is:
Hypothesis: There is an equilibrium relation between prices for the same natural gas product in different locations.
Allowances are being made for systematic price differences by not putting the restriction that prices are necessarily equal. Testing the above hypothesis for a location pair provides enough information to determine whether the locations from this pair are integrated. Thus, testing the hypothesis for all location pairs assesses the degree of market integration. An econometric model in the next section is specified to test the hypothesis empirically.
5 Methodology
From the literature survey and the from the previous section it can be concluded that it is a very common, if not a universal practice, to examine price relationships when testing for market integration. When speaking of a price relationship it is meant the relationship between two price series of the same homogeneous product in geographically distinct regions. In this paper I will use price analysis to test for market integration as well. As in most of the related literature, I will use a cointegration test. The cointegration technique tests for a long-run equilibrium relationship between prices. Cointegration is the appropriate technique since it is suitable for handling non-stationary time-series data, such as price series used in this paper. In addition to that I estimate a vector-error correction model. Cointegration and the vector-error correction model are explained further on. However, first I examine the data and some of its properties.
5.1 Descriptive statistics
To analyze the integration of the EU natural gas market I will use price data on gas hubs in the EU. This data is available for six gas hubs from five different EU countries: NBP (UK), Zeebrugge (Belgium), TTF (Netherlands), Gaspool (Germany), NCG (Germany), and PEG (France). Thus, in contrast to data used in some other papers (e.g. Asche et al. 2000 & 2002), I will use hub prices in Europe in stead of long-term take-or-pay contract prices. The reasons for this are (1) the amount of gas traded on hubs in Europe is rising (see section 2.3 and figures 2 and 3 from appendix A), and (2) hub prices are believed to reflect the perception of scarcity and hence can be considered as the ‘real’ market prices. Price data is known as time-series data3 where the observations are ordered according to time.
The unique dataset comes from the company ICIS Heren and gives prices for the day-ahead contracts. Day-ahead contracts are the most frequent traded products on the gas hubs. These are contracts for gas to be delivered following the day of the contract. Prices are volume-weighted averages of all day-ahead over-the-counter transactions to be delivered at a specific hub. The prices are published for every British working day and are two decimal accurate. The data covers a period from 23 April 2007 until 7 May 2010. The NBP and Zeebrugge prices are reported in pence per therm, whereas all other prices are reported in euro per
3
megawatt hour. Therefore the prices of NBP and Zeebrugge have been converged to euro per megawatt hour. Conversion values for therm to megawatt hour are based on the ‘units of measurement regulations 1995’ by the British office of public sector information4 (therm to megajoules) and on the ‘Guide for the use of the international system of Units (SI)’ by the National institute of standards and technology5 (megajoules to megawatt hour). Currency exchange rates have been obtained from the ECB. The movements of the different hub prices are shown in figure 5 from appendix A. Visual inspection shows that the hub prices move closely together over time. Since all prices are naturally positive, this paper follows the standard practice to use their natural logarithms instead. It is well know that this transformation reduces the skewness of the data greatly. A summary of the data before transformation to natural logarithms can be found in table one below.
Table 1: Summary of the data ( prices are reported in euro’s per megawatt hour)
Source: ICIS Heren
5.2 Cointegration
An interesting statistical property of time series variables, such as price data, is called cointegration. This concept has been introduced by Engle and Granger (1987). When two time series, say x and y, each are integrated of order d, I(d), then any linear combination of them will generally be also I(d). However, in special cases the linear combination of x and y can be integrated to a lower order I(d-b), b>0. If this is the case then x and y share similar stochastic trends and are said to be cointegrated at order CI(d, b). The typical case is where d=b=1. Importantly, cointegration implies that there is a long-run equilibrium relationship between these variables. Cointegrated variables share similar stochastic trends and never diverge too far from their equilibrium. By equilibrium is meant an econometric equilibrium; it may refer to any long-run relationship. Recall the hypothesis that needs to be tested:
4
http://www.opsi.gov.uk/si/si1995/Uksi_19951804_en_2.htm
5
Hypothesis: There is an equilibrium relation between prices for the same natural gas product in different locations.
An equilibrium price gap can be seen as a long-run equilibrium (Enders, 1995); if cointegrated, in the short run prices may differ (substantially) because of stochastic shocks, but arbitrage will prevent the prices of drifting too far apart, and eventually prices will move to their long-run equilibrium again. The difference between the prices is their systematic difference. Thus cointegration is the appropriate tool for testing market integration, it has been used in most of the related literature on the LOP and purchasing power parity (e.g. Asche et al., and L’Hégaret et al.,). I argue that, if two price series from two separate locations appear to be cointegrated, then the LOP holds and these locations form actually one market.
Two methods are most common to test for cointegration. These are the Engle & Granger method and the Johansen method. In this paper I will use the two-step Engle & Granger method. I prefer the Engle & Granger method because advantages of the Johansen method over the Engle & Granger method are specifically present only when one needs to determine more than one cointegrating vector at the same time including more than two variables (Enders, 1995), which is not the case in this paper. The Engle & Granger method consists of several steps explained below.
Step 1 of the Engle & Granger cointegration method
The first step of the Engle & Granger method is to determine if the five price series are individually non-stationary and each are integrated of the same order. Price series integrated of a different order cannot be cointegrated. Non stationary series do not have the property of mean reversion. I will use the most popular augmented Dickey-Fuller test to determine the order of integration of each of the individual price series. Formally, the augmented Dickey-Fuller test is:
t s it s m s t it p a p v p = + + Σ ∆ + ∆ − = − 1 1
γ
α
(1)of the residuals vt. By adding enough lags of the dependent variable the autocorrelation can be eliminated from the model. The number of lags is determined by examining the significance of the lag coefficients as. Note that I have added the intercept term
α
to the test as well, because from their nature prices are different from zero. The above test is repeated on the next difference of each price series until it becomes a stationary series, thus revealing the order of integration. The hypothesis that are tested:H0: γ =0 a series is non-stationary
H1: γ <0 a series is stationary
If the null-hypothesis is rejected, then it can be concluded that a series is stationary. To test the hypothesis we examine the tau-statistic for the hypothesis that γ =0. The null hypothesis is rejected if the tau-statistic (τ) of γ is lower than the critical value for the Dickey-Fuller test. Note that continuing to the next step now only makes sense if the results show that each price series is non-stationary and integrated of the same order, say d.
Step 2 of the Engle & Granger cointegration method
The next step is to pair every price series with every other prices series (6 price series = 15 pairs) and to estimate their relationship using ordinary least squares. This equation is called the cointegrating equation:
t bt
at p e
p =
β
0 +β
1 + (2)Where pa are prices from location a, and pb from location b, thus forming a pair. Prices of one location serve as the dependent variable and prices of the other location serve as the independent variable. The intercept term
β
0 can represent systematic price differences, such as transportation cost differences.β
1 represents the relationship between the prices. Note that for the absolute version of the LOP to hold,β
0needs to be zero andβ
1 needs to be one, leading to equal prices. However, by not putting these restrictions on the coefficientsβ
0 and1
Step 3 of the Engle & Granger cointegration method
If there is actual cointegration between these two price series then the errors et from equation (2) need to be stationary. If the series are cointegrated then et is stationary, because then any deviation from the long-run equilibrium is only temporary in nature. The test for stationarity of the residuals is based on the estimated residuals eˆt from equation (2) because the true residuals can never be observed. The augmented Dickey-Fuller test is therefore also performed on these estimated residuals:
t s t s m s t t e a e v e = + Σ ∆ + ∆ − = − ˆ ˆ ˆ 1 1
γ
(3)Equation (3) is estimated for all 15 pairs. Here eˆt is the estimated residual of equation (2). This time, however, there is no need to include an intercept term as in equation (1) because the estimated residual eˆt is expected to have a mean of zero. Again, as for equation (1), the null hypothesis can be tested against the alternative:
H0:
γ
=0 the series are not cointegratedH1:
γ
<0 the series are cointegratedThe null hypothesis that two series are not cointegrated can only be rejected if the tau-statistic of
γ
from equation (3) is lower than the critical value. Since the test is based upon estimated residual values from equation (2), the critical values used at this point are special critical values for the cointegration test. If it is not possible to reject the null hypothesis,γ
=0, then it is not possible to reject the hypothesis that the series are not cointegrated. In other words, rejection of the null hypothesis, is evidence that the series are cointegrated. Hence, it implies that there is a long- run equilibrium relationship between two price series, and therefore there can be spoken of one market.5.3 Vector error correction model
obtained from a vector error correction model (VEC) because cointegration only considers the long-run relationship. If two price series are cointegrated, then their time-paths are influenced by the extent of the deviation from their long-run equilibrium. In case of a temporary disequilibrium, at least one of the prices must respond to restore the equilibrium in the short run, this movement is called the error correction. As is written in Enders (1995: 366): “In an error-correction model, the short-term dynamics of the variables in the system are influenced by the deviation from the equilibrium”. If two price series, say pa and pb, are cointegrated, and
there is a temporary disequilibrium between them, then one of these prices or both of them, have to respond to restore their equilibrium. The VEC model gives insight in the extent to which each price will respond to a disequilibrium situation. Thus, the VEC model gives some additional insights, because, even though several locations can be integrated with each other, the efficiency among the locations may vary. In other words, for example, arbitrage between the NBP and Zeebrugge hub might work better and more efficient than between the PEG and NCG hub, indicating that NBP and Zeebrugge are better integrated. Formally the VEC model is as follows: at bt at x w at p p v p = + − − + ∆
α
α
( −1β
0β
1 −1) (4) bt bt at z y bt p p v p = + − − + ∆ α α ( −1 β0 β1 −1) (5)Importantly to note is that in the above equations (4) and (5) pat−1 −β0 −β1pbt−1 is the error part, it is the deviation from the long run equilibrium of two price series from equation (2). Note as well that it is the independent variable. Equation (4) and (5) imply that the change in pa and pb at time t, is determined by the deviation from the long-run equilibrium in time t-1
and some error term vt. The parameters of interest are αx and
α
z, which are the error correction coefficients. They are the speed of adjustment parameters. The VEC model can be estimated using a two-step least squares procedure. First I estimate the cointegrating relationships as in (2) using OLS and generate the lagged residuals since1 0
1
1 ˆ ˆ
ˆt− = pat− − − pbt−
e
α
β
. Next I can estimate the following equations using OLS:The coefficients αx and
α
z show the responses of pa and pb when there is a positivecointegrating error (et=1 >0) in equation (2). In case of a positive cointegrating error one
would expect the coefficient αx to be negative, and
α
z to be positive, thereby ‘correcting’ the error. In the case of a positive cointegrating error pa is larger (or pb is smaller) than onewould expect in the long run equilibrium. Thus one expects pa to decrease and/or pb to
increase, thereby restoring the equilibrium. Therefore the hypothesis that are tested for equation (6) are as follows:
H0: αx =0 The error is not corrected
H1: αx <0 The error is corrected
The hypothesis tested for equation (7) are:
H0:
α
z =0 The error is not corrected H1:α
z >0 The error is corrected6 Empirical results
6.1 Results of the Engle & Granger cointegration test
The analysis begins with the first step of the Engle and Granger test for cointegration. This step is to examine the order of integration for each individual price series. The results are found in table 2 from appendix B. The numbers reported in this table are the tau-statistics (τ) of the coefficients γ from equation (1). The test is performed on the price series of every hub, and on the first difference of every series (denoted by ∆). Both the Dickey-Fuller (DF) test and the Augmented Dickey-Fuller (ADF) test with two lags are performed. I have chosen two lags for the ADF test since the coefficients of most of these lags where found to be significantly different from zero at the 1% level. The critical tau (τc)values of Dickey-Fuller tests with a constant for the following significance levels are 1%: 3.43, 5%: 2.86, and 10%: -2.57. Only ifτ ≤τcthe null of a non-stationary series can be rejected.
The DF-test in the first column of table 2 shows that it fails to reject the null of non-stationary at any common significance level for the price series TTF, NCG, Gaspool, PEG. The null of non-stationarity can only be rejected for the NBP and Zeebrugge series at the 10% level of significance. However, when applying the ADF-test with 2 lagged variables (column 2) the test fails to reject the null of non-stationarity for all price series at any common significance level (since τ ≠≤τc). Therefore I argue that all individual price series are non-stationary. Tests on the first difference of the price series (denoted by ∆) leads to a rejection of the null hypothesis (since τ ≤τc), both according to the DF-test and the ADF-test with 2 lags. I therefore conclude that all price series are integrated of order one, I(1), because to make them stationary they had to be differenced once. The important result from this is that all individual price series are integrated to the same order, I(1), which is a condition for series to be cointegrated.
appendix B. Recall that if these estimated residuals appear to be stationary, then a series is cointegrated. Table 3 shows the tau-statistics from both the DF-test and ADF-test on the estimated residuals from the cointegration equation (2). The first column of the table shows the bilateral market pairs. The most left hubs in the first column acted as independent variables in the cointegration equation (2). The constant terms in the DF and ADF-tests are suppressed (since residuals are expected to have a mean of zero). Moreover, the ADF-test now uses three lags, since in most cases the coefficients of the three lags appear to be significant at the 10% level. Since there is a constant term in equation (2), and given that the cointegration test is based upon estimated values for the residuals, the following critical values are of interest: 1%: -3.96, 5%: -3.37, and 10%: -3.07. Table 3 shows, according to both the DF and the ADF test, that for all market pairs the null of no cointegration series can be rejected at the 1% level of significance. This leads to accept the alternative hypothesis for all cases. In other words, every gas hub appears to be cointegrated with every other gas hub at order CI(1, 1). The market pairs NBP-Zeebrugge, Zeebrugge-TTF, Gaspool, and NCG-PEG have the strongest rejection of the null hypothesis, these pairs have the lowest tau-statistics.
For robustness I analyze if the co- integrating relationships also hold independent of the choice of the independent variables in the co integrating equation (2). Therefore I switch for every market pair the independent and dependent variables in the cointegrating equation (2). The result is found in table 4 from appendix B. It shows that the cointegrating relationship holds for all market pairs, independent of the choice of the independent variable in the cointegrating equation (2).
6.2 Results of the VEC model
Since every pair appears to be cointegrated, the VEC model can be estimated for every bilateral market pair to estimate the speed of adjustment in case of a deviation from the equilibrium. The VEC model is a useful addition to the cointegration technique because it can confirm cointegration and it examines the efficiency of bilateral market pairs. The results from estimating the VEC model (equations 6 and 7) without the lagged changes of each price are found in table 5 from appendix B. The table shows estimates for the coefficients ax and az
from equations (6) and (7). The column ‘First hub’ shows the speed of adjustment (ax from
equation 6) of the price from the first hub of a bilateral market pair listed in the most left column. The column ‘Second hub’ shows the speed of adjustment (az from equation 7) of the
price from the second hub of that same pair. Thus, the ‘First hubs’ are the dependent variables in equation (6), the ‘Second hubs’ are the dependent variables in equation (7). The t-statistics are reported in parentheses. Note that all signs are as expected! In case of a positive cointegrating error (if the estimated residual from equation 2 is larger than zero), prices of the first hubs will fall and prices of the second hubs will rise, thereby restoring the long-run equilibrium between them. Moreover, as expected all coefficients are different from 0 and lower than |1|, indicating that the system is not explosive and that the error indeed leads to a correction. The t-statics reveal that all coefficients are significant at the 1% level, and thus the null that the error is not corrected can be rejected for all cases. This supports the hypothesis that every bilateral market pair is cointegrated. Cointegration namely implies that the long-run equilibrium will be restored after temporary deviations.
equilibrium are completely restored the next day. Analogously, if both coefficients are zero, then deviations from the long-run equilibrium are not restored at all, which would imply as well that the locations are not cointegrated. Note that a coefficient size does not tell anything about the deviation sizes or the frequency with which deviations occur. The coefficients do not tell anything about causality either. The VEC model ‘only’ gives insight in the extent to which prices respond to equilibrium distortions, and not to which prices respond to each other. To determine causality between the prices one should perform special causality tests, such as a test for Granger causality. This is however outside the scope of this paper since it does not contribute to answering the main research question. Thus, a large coefficient only tells that a price series has a rapid price adjustment. Analogously, a small coefficient tells that a price series adjusts relatively slow to equilibrium distortions.
Table 6 in appendix B reports the same result of the VEC model but now with two lagged changes of each price, since most of these lagged changes are significant at the 10% level. Generally, the coefficients in table 6 are lower in absolute value than those in table 5, this is caused by introducing the lagged changes of each price. Relatively the coefficients do not change drastically; a relatively low/high coefficient in table 5 is followed by a relatively low/high coefficient in table 6. However, the t-statistics of the coefficients have dropped and not all coefficients are significantly different from zero at the 10% level anymore. The coefficient of TTF for the pair NBP-TTF is not significant at the 10% level anymore. This suggests that in case of a deviation from the equilibrium between the NBP and TTF prices, the price of NBP tends to do the adjustment. The coefficient of NCG for the NCG-Gaspool pair is only significant at the 10% level. Taking this into consideration, and looking at the absolute value of the coefficient, this might indicate that Gaspool tends to do all the adjustment. Nevertheless, table 5 and 6 present the same pattern. That is, deviations from the equilibrium are always restored, though never immediately. Moreover, there are no locations that are perfectly integrated with each other, and the degree of integration differs per pair.
could indicate that there are barriers to trade between NBP and continental European hubs. Third, one could expect that the more liquid market places, with more buyers and sellers, would respond relatively faster to equilibrium distortions. This pattern can be observed from table 5 as well. The relatively mature markets NBP and Zeebrugge respond faster to equilibrium distortions than their counterparts. This could be explained by the fact that these hubs are home to relatively many buyers and sellers looking for arbitrage opportunities. Interestingly, it is found that TTF, which is considered as one of the most mature trading hubs on continental Europe, does not seem to respond as fast as NBP and Zeebrugge do.
6.3 Some similarities and contrasts with existing literature
7 Conclusion
In this paper I have investigated the extent of market integration for natural gas in the European Union. This integration of the natural gas market is enhanced by EU politicians and is embedded in the creation of a single EU energy market. Overall, market integration results in eliminating existing price differences between countries (with the exception of systematic price differences), this is known as the relative law of one price (LOP). The mechanism ensuring that prices converge to the LOP is called arbitrage. Prices of organized natural gas markets (gas hubs) from several EU countries are used in the analysis. These hub prices are, in contrast to long-term contract prices, believed to reflect the perception of scarcity for natural gas. Prices in this paper come from the following hubs: NBP (the UK), Zeebrugge (Belgium), TTF (The Netherlands), NCG (Germany), Gaspool (Germany), and PEG (France). If there appears to be a long-run equilibrium relation between natural gas prices of these different hubs, then the relative LOP holds, and thus these locations are believed to form an integrated market for natural gas. I used the Engle and Granger cointegration test to test for a long-run equilibrium relationship between prices. Next to this, I estimated a vector-error correction model to examine the short-term market dynamics.
impact. Most importantly it implies that these countries no longer solely rely on there own national gas market. Countries can secure there supply of natural gas better since they have access to a larger, integrated market. Furthermore, in an integrated market there will be more suppliers competing with each other, leading to a more efficient industry and possibly lower prices for end consumers.
In this paper there are some limitations one should be aware of. First, the data has unfortunately restricted this research to Western EU countries only. Measured to population and GDP, these countries make up an important part of the European Union. However, whether other parts of the EU (Scandinavia, Southern- and Eastern Europe) are part of this integrated market as well remains unclear. Thus the conclusion that there exists a single market for natural gas in the entire EU might be too opportunistic. Further research with more data, for example on gas hubs in Austria (CEGH), Italy (PSV), Spain (MS-ATR), and Scandinavia (GTF), could shed more light on this issue. Second, this study analyzes only prices of the day-ahead gas contracts on European gas hubs. Although these are the most frequent traded contracts on hubs, it would be interesting as well to examine other contracts, such as month-ahead and year-ahead contracts. Third, although the results do not point to it, a bias might have been caused by converting prices from the NBP and Zeebrugge hubs from Pence per term to Euro per megawatt hour. Conversion of prices through exchange rates might distort the analysis if the purchasing power parity between the Euro and the British pound sterling does not hold. In other words, at the going exchange rate one should be able to buy the same amount of goods for one Euro in both the UK and the EU. If this is not the case, then there is not parity in purchasing power.
This paper leaves some interesting topics to explore in further research. First, it would be interesting to examine market integration with other parts of the EU. Second, it is believed that the rise of LNG enhances world market integration for natural gas since it makes trade in natural gas less dependent on pipeline networks. Thus, to what extent is there a world market for natural gas? For example one could investigate if there is a relation between natural gas prices of hubs in Europe and hubs the USA. Third, what are the dynamics of the integration of the EU natural gas market? For policy makers it would be interesting to know whether the integration process is increasing, stable, or decreasing. Fourth, this study has concluded that at least the relative version of LOP holds in the long term. It would however be interesting to find out how systematic price differences between locations evolve over time. As the gas market is becoming better integrated, an analysis on the absolute version of the LOP might be interesting.
References
Asche, F., Osmundsen, P., & Tveteras, R. 2000. Market integration for natural gas in Europe.
SNF-Report, no. 45/00.
Asche, F., Osmundsen, P., & Tveteras, R. 2002. European market integration for gas? Volume flexibility and political risk. Energy economics, 24: 249-265.
Bunn, D.W. & Micola, A.R. 2007. Two markets and a weak link. Energy Economics, 29(1): 79-93.
Chen, Z. & Knez, P.J. 1995. Measurement of market integration and arbitrage. The review of
financial studies, 8(2): 287-325.
Cronshaw, I., Marstrand, J., Pirovska, M., Simmons, D. and Wempe, J. 2008. Development of competitive gas trading in continental Europe. How to achieve workable competition in European gas markets? IEA information paper.
Cuddington, J.T. and Wang, Z. 2006. Assessing the degree of spot market integration for U.S. natural gas: evidence from daily price data. Journal of regulatory Economics, 29: 195-210.
Enders, W. 1995. Applied econometric time series. New York: Wiley.
Engle, R.F. & Granger, W.J. 1987. Cointegration and error correction: representation, estimation, and testing. Econometrica, 55(2): 251-276.
European Commission, 2009. Report on progress in creating the internal gas and electricity
market.
European Commission, 2010. Report on progress in creating the internal gas and electricity
Eurostat. Energy Database.
http://epp.eurostat.ec.europa.eu/portal/page/portal/energy/data/database, Accessed June 08, 2010.
Gas Transport Services 2. TTF volume ontwikkeling.
http://www.gastransportservices.nl/nl/transportinformatie/ttf-volume-ontwikkelingen, accessed July 2, 2010.
Gas Transport Services. Entry- en exitcapaciteit.
http://www.gastransportservices.nl/nl/corporate/gastransport/entry-en-exitcapaciteit, Accessed June 28, 2010.
Haase, N. 2008. European gas market liberalization. Competition versus security of supply? Unpublished doctoral dissertation, University of Twente, Enschede.
Harris, N. & Jackson, M. 2005. A picture of the European gas trading market in 2005.
Pipeline and gas journal, 232(8): 54-58.
Jepma et al. 2009. Gas market trading. Groningen: Drukkerij van Denderen B.V.
Keyaerts, N., Meeus, L. & D’haeseleer, W. 2009. First results of the integrated european
gas market: one for all or far from one. Paper presented at the 24th world gas conference, Buenos Aires.
L’Hégaret, G., Siliverstovs, B. Neumann, A. and Hirschhausen von, C. 2003. International market integration for natural gas? A cointegration analysis of gas prices in Europe, North America and Japan. DIW Berlin, discussion paper 393.
Lazckó, M. and Lajtai, R. 2009. Gas hub as a source of regional market development in the
Central and Eastern European region. Paper presented at the 24th world gas conference, Buenos Aires.
Meeus, L. & Belmans, R. 2008. Electricity market integration in Europe. Paper presented at the 16th PSCC, Glasgow.
Naturalgas.org. Background. http://www.naturalgas.org/overview/background.asp, Accessed June 08, 2010.
Neumann, A. Siliverstovs, B, & Hirschhausen von, C. 2006. Convergence of European spot market prices for natural gas? A real-time analysis of market integration using the Kalman-filter. Applied economics letters, 13: 727-732.
Stigler, G.J. & Sherwin, R.A. 1985. The extent of the market. Journal of law and economics, 28(3): 555-585.
Taylor, A.M. 2001. Potential pitfalls for the purchasing-power-parity puzzle? Sampling and specification biases in mean reversion tests of the law of one price. Econometrica, 69(2), 473-498.
U.S. Energy information administration. International Energy Statistics.
http://www.eia.doe.gov/cfapps/ipdbproject/IEDIndex3.cfm?tid=3&pid=26&aid=24, Accessed June 08, 2010.
Appendix A: Information on gas hubs
Figure 1: Major European gas hubs and gas import routes
Source: E.ON Energy trading: http://www.eon-energy-trading.com/cms/en/downloads/gas_trading_routes.pdf, accessed 2 July 2010.
Figure 2: Traded volume development at the Dutch gas hub TTF in cubic meters
Figure 3: Traded volume development at the German gas hub EGT/NCG, (red bars are broker screen data, black bars are TSO data)
Source: E.ON energy trading: http://www.eon-energy-trading.com/cms/en/downloads/hub_volumes.pdf, accessed 20 October 2010
Figure 4: Traded volumes and churn estimates for different European gas hubs
Figure 5: Movement of the hub prices for day ahead products in Euro per megawatt hour
Appendix B: Statistical results
Table 2: Tau-statistics from the DF- and ADF test on the order of integration
Table 4: Tau-statistics from the DF- and ADF test on stationarity of the residuals
