Price Convergence and Information Efficiency in German Natural Gas Markets



In 2007, Germany changed network access regulation in the natural gas sector and introduced a so-called entry–exit system. The spot market effects of the reregulation remain to be examined. We use cointegration analysis and a state space model with time-varying coefficients to study the development of natural gas spot prices in the two major trading hubs in Germany and the interlinked spot market in the Netherlands. To analyse information efficiency in more detail, the state space model is extended to an error correction model. Overall, our results suggest a reasonable degree of price convergence between the corresponding hubs. Market efficiency in terms of information processing has increased considerably among Germany and the Netherlands.

1. Introduction

The introduction of the European Gas Directive (98/30/EC) and the EU ‘Acceleration Directive’ (2003/55/EC) brought fundamental changes in the natural gas sector across many European countries. The natural gas industries were transformed from vertically integrated monopolies to more competitive structures coupled with price deregulation and privatisation. While some countries, notably the United Kingdom and the Netherlands, have been relatively progressive in the liberalisation process, others such as Germany have moved more cautiously. Germany did not effectively open its natural gas market until the EU Directive 2003/55/EC was transposed into national law in 2005.

The implementation of the European directive into national law led to a new institutional design called the entry–exit system. The primary aims of introducing the entry–exit system were to facilitate market entry in the energy sector and establish competition in wholesale and retail energy markets, as per the EU policy of creating a common integrated market for natural gas in Europe. As a consequence, regional natural gas wholesale markets in the form of different virtual trading points (hubs) were established in Germany. Two major German hubs emerged, with NetConnect Germany (NCG) roughly covering the South and GASPOOL (GPL) covering the North, essentially creating a liquid market for natural gas in Germany. Whether these two subnational markets are well-functioning remains to be empirically tested.

To evaluate whether the German natural gas wholesale market is functioning yet, we study the price convergence and the development of market efficiency in Germany. Price convergence shows how strongly the prices are related across similar markets, whereas information efficiency depicts the ability of the market to respond to available information and incorporate it into the prices. Therefore, prices should reveal and aggregate all available information in an efficient market. However, analysing the prices across only the two major German hubs may be misleading as it could neglect the possible importance of the directly connected, large Dutch gas market. The Dutch Title Transfer Facility (TTF) hub could be considered to be one of the most liquid wholesale gas trading hubs in continental Europe. The large number of transactions and participants at the TTF create a sound competitive environment. A high level of trading activity by domestic and foreign traders also illustrates the importance of the TTF for the European gas market. However, the balancing regimes between TTF and German markets vary as TTF, at least on a physical level, is an hourly market. Nonetheless, TTF can serve as a (competitive) benchmark for the German natural gas spot markets considering its liquidity and proximity to Germany.

Furthermore, well functioning, connected markets, in theory, should show equal prices for a certain good (law of one price). Such markets share a common long-run equilibrium price and can be said to be economically integrated. Persistent price differences beyond transaction costs should not exist. This is to say that the price differential does not have to vanish for the market to be efficient. In reality, the price of energy is different most of the time, implying that the long-run equilibrium price levels and the resulting variation could be crucial, though not an explicit part of our analysis. Likewise, the market's response to external shocks and corresponding reversion to the competitive equilibrium should be fast, i.e. information should be processed efficiently.

To analyse whether the prices at the different hubs have common long-run equilibrium, we use cointegration analysis (Johansen, 1988, 1991). One implicit assumption of cointegration analysis is that the structural relation among the prices is fixed over the time period considered; however, due to the ongoing changes in the regulatory framework in Europe and Germany and the different mergers of Germany's gas transmission system operators (TSOs), market integration could also be a gradual and ongoing process. In a second step, we therefore examine the convergence path of the natural gas spot market prices and the degree of market integration by estimating a state space model using a Kalman filter. In contrast to cointegration analysis, this estimator allows time-varying coefficients and thus explicitly accounts for possible dynamic structural changes. Finally, the state space model is extended to an error correction model to analyse how efficiently markets respond to new information. The time-varying nature of this approach allows us to draw conclusions as to how market efficiency evolved over time and whether changes in the regulatory framework have led to better market functioning. We find a reasonable degree of price convergence between the corresponding hubs. Also, information efficiency has increased considerably among Germany and the Netherlands.

Natural gas market integration and price convergence as an aftermath of market liberalisation have been the focus of a variety of international studies. However, the methodology used to account for market integration and price convergence differs across studies. Using cointegration analysis, Walls (1994) and De Vany and Walls (1993, 1996) find that the opening of network access in the aftermath of FERC Order 436 in 1985 led to greater market integration, as prices across different locations converged in the North American natural gas markets. Serletis and Rangel-Ruiz (2004) show that the main driver for North American natural gas prices is the price trend at the Henry Hub. Applying a vector error correction (VEC) model, Cuddington and Wang (2006) find different degrees of market integration across regions. While the East and Central regions are highly integrated, the Western market is only loosely connected to common price trends.

In the European context, Asche et al. (2002) apply cointegration techniques to test for the law of one price across the French, German and Belgian markets using monthly natural gas import prices. Their results show an integrated gas market where prices across the regions considered follow a similar pattern over time. Using a Kalman filter, Neumann et al. (2006) study the price relation between the United Kingdom (National Balancing Point) and the Belgian spot market (Zeebrugge). They conclude that prices between these two markets have fully converged. Other empirical studies like Ripple (2001), Siliverstovs et al. (2005) and Neumann (2009) have investigated price relations for natural gas between different continents. All studies find evidence for an increased price convergence across continents.

Our article differs from the existing literature on natural gas price convergence and market integration in three ways. First, we focus our analysis on the market effect of a new network access regime on a subnational gas market level by looking at the dynamic price interactions between two major German gas spot markets. Moreover, these regional market developments are put into a European context by relating them to the Dutch spot market. Second, we apply both cointegration analysis and a time-varying coefficient approach. Extending the time-varying coefficient approach to an error correction model enables us to draw conclusions not only about price convergence but also about how efficiently new information is absorbed by the market and how this information efficiency has evolved over time. Third, we explicitly control for transportation costs, which have been neglected to a certain extent in previous literature.

This article is structured as follows. Section two describes the institutional design of the German natural gas market. The econometric methodology is laid out in section three. Data description follows in section four. Section five includes the estimation results and interpretation. Finally, section six concludes with potential policy recommendations.

2. Institutional Design and Recent Developments in Germany

In 2005, a new Energy Law (Energiewirtschaftsgesetz, EnWG) was enacted in Germany to transpose the European Directive 2003/55/EC into national law. Its major purpose is to develop and establish competition in the German energy sector. A system charging regime based on simple entry and exit charges (entry–exit system) was imposed concerning third-party access to the natural gas networks. The transmission network operators and the regulator concluded an agreement regarding the institutional design of the new regime in October 2006. Since 1 October 2007, the entry–exit system has become mandatory for all TSOs. The agreement initially divided Germany into 19 entry–exit zones (also called market areas or transmission system zones). Meanwhile, the number of zones have decreased to three, one for L-gas and two for H-gas (as of April 2011) due to several poolings mandated by additional governmental intervention in the form of an amendment of the German Energy Law (§20 Sec. 1b EnWG). H-gas is a high caloric natural gas primarily delivered from Norway and Russia to Germany. L-gas is a low caloric natural gas and has a lesser energy content than H-gas. Our focus is solely on H-gas, as low calorific gas plays only a minor role in the German economy. The market zones for H-gas were developed as follows: NCG initiated a major pooling which became operational on October 1st, 2008 and combined the former areas of E.ON and Bayernets. In October 2009, GRTgaz Deutschland, ENI and GVS joined NCG. In April 2011, Thyssengas, the former RWE gas TSO, joined NCG as the latest member. While NCG covers the South and West of Germany, GASPOOL (GPL), is located in the North of Germany and is the second major market zone.1 The core of this area had already been established in 2006 as a result of a cooperative arrangement between BEB, StatoilHydro and DONG Energy. In July 2008, Gasunie, operating the Dutch transmission network, took over the transportation services of BEB. In October 2009, ONTRAS and Wingas Transport joined GPL.

The entry–exit system requires that the natural gas shippers book capacity at the relevant entry and exit points separately; hence, the charges to be paid for the transportation of natural gas (the so-called entry and exit charges) are no longer based on the distance between the entry and exit points (also known as the contractual path) as previously practised in Germany. The abolition of such a ‘path-based’ charging system was meant to promote price transparency, as shippers no longer needed to obtain individual quotations for each separate customer, thereby reducing pricing complexity. Moreover, the trading possibilities at multiple hubs, as a result of an entry–exit system, should deliver a competitive price signal to the German natural gas market as a whole. The entry–exit system should facilitate both domestic as well as cross-border transports for third parties, thus encouraging market entry and eventually competition. Furthermore, the market redesign aimed at increasing the flexibility and comfort in booking procedures as no capacity reservation is required for the individual pipeline sections used to fulfil transport contracts. In sum, consumers and distributors were intended to benefit from increasing gas-to-gas competition as a result of the implementation of the entry–exit system after gas market liberalisation.

However, the potential economic success of the reform is not clear. The German Antitrust Commission complained about the illiquidity of the natural gas wholesale market, congestion and grandfathered capacity rights held by the incumbents (Monopolkommission, 2009). Whether such market barriers could effectively rule out the aim of achieving competition and liquidity in the natural gas sector, even with the introduction of the entry–exit regime, remains an empirical question.

3. Econometric Methodology

In this section, we (1) deduce empirical criteria for measuring German natural gas market competitiveness and its development, and (2) present the corresponding econometric methodology. To test the law of one price, we adjust the regional natural gas spot prices at GASPOOL (GPL), NCG and the Dutch TTF hub accounting for transmission charges. We employ cointegration analysis to test whether prices tend to move towards a common long-run equilibrium price. We use these results as a first indication as to whether markets are integrated or not. In a final step, we estimate a time-varying coefficient model using the Kalman filter to study price convergence over time. Here, we develop a time-varying error correction model to identify the development of information efficiency.

3.1. The law of one price and transmission charges

The starting point for our analysis is the spatial arbitrage condition for efficient markets, which says that prices of identical products traded at regionally distinct locations should differ only in transaction costs (law of one price). In natural gas markets, the most noteworthy transaction costs are the network charges that shippers have to pay for gas transmission. Thus, freedom from arbitrage for gas traded at different locations is assured as long as the price in the exporting region (Pj,t) plus the cost of transmission (TCji,t) equals the price in the importing region (Pi,t). The spatial arbitrage condition can be generalised as follows:

display math(1)

where the equality holds only if trade between the two regions occurs for any day t. If the price differential is strictly less than the cost associated with gas transport, market participants have no incentive to trade. Moreover, efficient markets should not involve any significant price differences that go beyond the transaction costs necessary to realise spatial arbitrage. To detect these price differences, we adjust the gas price series for transportation costs if the price difference between two regions exceeds the corresponding transmission charges. Thus, we assume rational arbitrage behaviour of shippers. In an entry–exit regime, these costs in one direction consist of the exit charge for the exporting region and the entry charge for the importing region. With regard to continental Europe, transmission charges are direction-specific, i.e. TCij,t ≠ TCji,t. Due to this asymmetric characteristic, the adjusted spot price (ex transmission charges) math formula can be calculated as suggested by Zachmann (2008) who analysed convergence of European electricity spot prices controlling for the outcome of capacity auctions:

display math(2)

We include a set of binary variables d, mapping potential flows between the regions i and j. A binary variable dji,t is equal to 1 if the price Pj,t in the low price region plus the costs of shipping the natural gas to the high price region TCji,t are less than the price Pi,t, and zero otherwise. Thus, spot prices are adjusted only if the price differential between two regions exceeds transportation costs, i.e. arbitrage is profitable.

The price adjustment enables us to detect sustainable price differences other than transportation charges. It should be noted that binary variables for two regions can both be zero at the same time, but they can never be unity at the same time. The former situation prevails if spot price differentials are too low to exceed the transmission charges, in which case trading gas between the two regions would be unprofitable.

Using log prices math formula we will analyse if the markets are economically integrated, i.e. if prices at distinct locations show similar price patterns. If prices move apart and price differences other than transportation charges prevail, markets are not economically integrated and not efficient. However, real world complexities in trading imply that several factors may exist leading to deviations from these conditions. For example, the price differential, among others, can contain different factors like balances of demand, day-ahead booking based on scheduled and forced outages, interruption risk variations, maintenance, exogenous gas flows, oil expected variance, winter versus front season demand and market liquidity, which are not possible to explicitly account for in our model. Our model implicitly assumes that any variation resulting from non-captured risks are equal to zero within the period of examination over the long run. Nonetheless, the hypothesis to be tested is that the establishment of an entry–exit system has removed some of the causes of inefficiency, and that prices converged as a result thereafter.

3.2. Cointegration tests

A precondition for cointegration analysis is that the individual time series are integrated of order 1, I(1). Therefore, we test for unit roots using the Augmented Dickey Fuller (ADF, Dickey and Fuller, 1979) and the Kwiatkowski Phillips Schmidt and Shin (KPSS, Kwiatkowski et al., 1992) stationarity tests. Both tests allow us to detect unit roots in individual time series, though the assumption on the test statistic obtained under the null hypothesis varies. ADF obtains the test statistic under the null of the series being non-stationary, whereas KPSS assumes the null hypothesis of stationary to run the test. ADF tests also tend to have low explaining power, especially in small sample dataset when compared with KPSS. Hence, double testing can mitigate the risk of a false conclusion when unit root tests face the problem of poor power properties.

Applying the Johansen cointegration test (Johansen, 1988, 1991), cointegration to natural gas spot price series of two regions, one expects exactly one cointegrating relationship if these regions have a long-run equilibrium. The corresponding two dimensional VEC model with the cointegration rank r (less than and equal to k) is given in the following equation:

display math(3)

where Δ is the first difference operator, math formula is the vector of the two log spot prices, ɛt ∼ n.i.i.d.(0, Σ), Π is a matrix of the form Π = αβ, with β comprising the cointegrating vector and α representing the corresponding loadings. α and β are k * r matrices and Γk is a k * k matrix. While β coefficients show the long-run equilibrium relationship between price levels, α coefficients measure the adjustment speed towards equilibrium. The closer β is to (minus) one, the better economically integrated the markets are. A high absolute value for α, in turn, indicates a high speed of price adjustment and a more efficient market. Furthermore, the value of α can also increase due to the introduction of a new pipeline which engenders a competitive effect and due to over-capacity in both markets.

According to equation (1), we expect β = {1, −1}. In order to control for price differences other than transmission charges, we allow for a constant in the cointegration relationship; however, for efficient markets, this constant should be (close to) zero.

The Johansen approach assumes a constant cointegrating vector over time. As pointed out by King and Cuc (1996), the cointegration relationship does not shed any light on the dynamics of possible price convergence or divergence. Given the background of the dynamic regulatory framework in Europe, as pointed out earlier, the assumption of a fixed relationship between spot prices over time may be problematic. Following the line of argument of Barrett and Li (2002), we use results from cointegration only as a kind of pretest as to determine whether markets are integrated or not.

3.3. Time-varying coefficient model

An approach with time-varying coefficients can overcome the drawbacks of cointegration analysis and can account for the dynamics of parallel price developments from regionally distinct markets. The introduction of a time-varying coefficient into the linear relationship of prices enables us to analyse the path of price convergence or divergence over time.

By recalling equation (1) and introducing a constant cij, we can formulate the following state space model:

display math(4)

where math formula and math formula are white noise processes and βt is the scalar of unobservable coefficients at time t which resembles a random walk process without drift. This implies that βt is neither variance nor covariance stationary, but rather mean stationary. The constant cij is expressed in log levels, whereas the original transmission charge is expressed in price levels.

The coefficient βt represents the strength of price convergence across regions. If βt = 0, it implies that there is no relation between the natural gas spot prices and that markets are completely decoupled. If prices have converged and markets are perfectly integrated and competitive, βt should be equal to 1.

The state space model is estimated using the Kalman filter (Harvey, 1987; Kalman, 1960). This technique processes the data in two consecutive steps. It first estimates βt by using available information for the period t − 1. In a second step, the estimates of βt are updated by incorporating prediction errors from the first step (to time t − 1) to compute values for time t. Applying the Kalman filter provides information for cij and for βt at each point in time, and thus enables us to obtain detailed information on the common development of prices.

In employing the Kalman filter, it is important to determine the initial variances for ɛt and νt, as well as of the expected value of β0. This is because the maximum likelihood function usually has several local maxima. Therefore, inadequately chosen starting points can lead to undesirable results. Exaggerated values of math formula would lead to the inclusion of short-term behaviour, making it difficult to distinguish random shocks from structural relationships. In contrast, setting the variance too low would ignore significant developments in the convergence process over time.


display math

provides suitable noise reduction and signal preservation.

Finally, we use the framework of time-varying coefficients to formulate an error correction model in the following way:

display math(5)

where αt measures the time it takes to bring the system back towards equilibrium after new information appears on the market. The larger the absolute value for αt is, the higher the speed of price adjustment is and the more efficiently information can be converted into price signals. Therefore, the time-varying framework enables us to draw conclusions not only on how efficiently prices are adjusted to new information but also how efficiency evolved over time. Since the entry–exit regime was introduced to ease gas transmission and foster gas-to-gas competition, we expect the absolute value of αt to increase over time.

The error correction model specification needs a new calibration of the starting values for the Kalman filter:

display math

Setting the expected value of α0 to zero assumes inefficient information processing at the beginning of the observation period. This seems especially plausible, since market participants generally need time to adapt to a new regulatory regime (Growitsch and Weber, 2008).

4. Data

The aim of this article was to test for market integration and price convergence between the major two entry–exit zones in Germany, namely GPL and NCG. In addition, we analyse price relations with respect to the Dutch trading hub TTF, which is well connected with Germany and a major European gas trading point. It should be noted that the TTF hub is connected to LNG terminals, implying that the LNG flow may influence the price surrounding the Netherlands, whereas the gas flow from Russia has a greater influence on price in NCG areas. For the German trading hubs, we have used the day-ahead spot market settlement price for natural gas, publicly obtained from the European Energy Exchange (EEX), whereas data for TTF has been obtained from Energate. Although only 10% of the total volume is traded at EEX (90% is traded over-the-counter, OTC), we have chosen EEX data as it is the only publicly available price series covering a longer time period without structural breaks. The use of high frequency data should better capture the markets’ reactions to ongoing regulatory and market reforms, thereby facilitating the investigation of market integration. The observation period is from 1 October 2007, when the mandatory introduction of the entry–exit system came into operation, to March 2011. Price data for both German market areas prior to October 2007 hardly exist and face the problem of low reliability. The time period considered covers three and a half so-called ‘gas years’, each starting with the heating season in October. Thus, we have 915 observations (trading days) for every price series. The prices have been transformed into logarithmic form. Log-transformed prices better capture the underlying distribution of the residuals used in our model as assumed under the financial pricing theory. As spot prices are log-normally distributed, the log returns in the ordinary least squares of cointegration are also assumed to be normally distributed.

It can be seen from Figure 1 that the day-ahead prices for natural gas were volatile throughout the time period considered for all market areas. The prices became more volatile during the last quarter of 2007, whereas the volatility declined from the first quarter of 2008 to the third quarter of 2008. Starting in the first quarter of 2009, prices witnessed a steep decline. The change is likely – at least to some extent – to be explained by the falling prices for crude oil. Gas prices in mainland Europe are often index-linked to those of crude oil, with an expected time lag of between a quarter and a season. While the price for Brent Crude Oil peaked on 3 July 2008, GPL prices reached their maximum on 23 September, 2008. NCG and TTF prices both peaked on 17 September 2008. This also indicates that the development of oil price can be crucial to our model. From the third quarter of 2009 onwards, gas prices started to rise again due to the ongoing recovery of the economy after the crisis in 2008–09. All in all, the price series show a rather similar development pattern over time,2 leading to the expectation of highly integrated markets and thus values close to one for the β coefficient in equations (3) and (4).

Figure 1.

Logarithmic day-ahead spot prices (€/MWh)

Information on transmission charges was obtained from the websites of the relevant TSOs as well as in interviews with the shippers. However, consistent historical data on transmission charges is not publicly available, which can be problematic. This is because the quality of the analysis of the spot gas markets depends on the quality of the data on transmission charges. For each possible relation between the three considered markets, we chose a representative connecting point. To get consistent data, we converted the corresponding capacity-based entry and exit charges expressed in [€/(kWh/h)/a] into [€/MWh]. We assumed that a transport of one MWh of natural gas corresponds to a capacity of one MW. However, the capacity charge could contain a time value for the provided flexibility (real option). Consequently, the capacity utilisation does not have to be aligned with the activities on spot gas markets, which may lead to partial mismatches of maturities. The charges are frequently changed by TSOs. The resulting development of transmission charges over time is summarised in Table 1. While the cost of transporting natural gas within Germany has decreased quite substantially (by about 25%), charges for cross-border transmission have decreased slightly.

Table 1. Transmission charges
Direction of transportConnecting pointTransmission charges [€/MWh]
  1. GPL, GASPOOL; NCG, NetConnect Germany; TTF, Title Transfer Facility.

GPL → NCGBunder Tief0.6190.5820.5650.5340.5340.4830.4830.453
NCG → GPL0.5570.5230.5120.4970.4970.4340.4340.417
GPL → TTFOude Statenzijl0.4200.4200.4200.4100.4000.3660.3610.361
TTF → GPL0.4090.4090.4090.3940.4080.3380.3340.334
NCG → TTFBocholtz0.3870.3990.3860.3860.3980.4110.4060.398
TTF → NCG0.4860.4980.4840.4840.4900.4290.4220.397

All three regions show a significant number of trading days, at which price differentials exceed transportation costs (Table 2). While within Germany price differences are mostly below transportation charges, nearly 50% of the time arbitrage would have been profitable regarding the two neighbouring regions GPL and TTF. Last but not least, price differences above shipping costs are observed more often during the first half of the period of examination.

Table 2. Summary statistics on price differences
  1. GPL, GASPOOL; NCG, NetConnect Germany; TTF, Title Transfer Facility; TC, cost of transmission.

Average [€/MWh]0.250.420.17
Mean absolute [€/MWh]0.520.560.56
No. of observations915915915
Days with price differences above TC (total)281453364
Days with price differences above TC (first half)204238235

5. Results

To answer the question as to whether the introduction of an entry–exit regime has led to more competitive market conditions, we first present the results of the cointegration analysis, and then the results of the time-varying coefficient approaches.

5.1. Cointegration analysis

To check whether the price series fulfil the precondition for cointegration analysis, we test for unit roots. The results from both ADF and KPSS provide a clear picture. All-time series have a unit root and are I(1), as first differences are stationary.

Next, we estimate equation (3), applying the approach developed by Johansen (1988, 1991). Each pairwise price relation has exactly one cointegrating term indicating that long-run equilibria do exist. Price series share a common stochastic trend and hence will not drift apart greatly in the long run.

All β coefficients from the VEC model are very close to one. A likelihood ratio test detects if the restriction of the cointegrating vector is β = {1, −1}, meaning that a 1% price change in region i is accompanied by the same price change in region j. Only in the case of GPL and the Dutch TTF does this indication of very strong market integration have to be rejected at a 10% level. An insignificant constant in the long-run cointegrating equation signals that no other significant price differences exist in addition to the already captured transmission charges, and therefore no persistent price differential occurs. Looking at the error correction coefficient α, we see significant bidirectional price adjustments for GPL-NCG and GPL-TTF. The relationship of natural gas spot prices between GPL and TTF is stronger since the level of significance and adjustment speed are a bit higher. However, for GPL-NCG only around 10% of an external shock is absorbed within one period (i.e. one trading day), GPL prices adjust for 20% of an imbalance with TTF prices. The corresponding half-lives, defined as the time in which a marginal change in the stationary component becomes half of the initial jump and computed as ln(0.5)/ln(1 − α), are 6.4 and 3 days respectively. The stronger interrelation between GPL and TTF may be due to the fact that both networks are now run by the same TSO, Gasunie. The asymmetric price adjustment in the latter case, with GPL prices adjusting faster than TTF prices, indicates that TTF, as the larger and more liquid market, is the leading market for GPL. The same argument holds for the NCG-TTF results, with only the NCG adjusting to deviations from equilibrium. A likelihood ratio test restricting αTTF to zero confirms that TTF is weakly exogenous for NCG. These results support the hypothesis that the Dutch TTF can be considered as a kind of reference or leading market for both German market areas.

To summarise, results from cointegration analysis provide evidence of market integration across the two German market areas considered in this study; however, one has to interpret these results with care. The presence of cointegration does not necessarily imply the stability of the estimated β parameter. Moreover, the long-run β coefficient may not stay constant over time, as several structural changes occurred in the natural gas markets within the period considered.

5.2. Time-varying coefficient

The time-varying coefficient approach is particularly suitable for accounting for these structural changes. Table 3 presents the main results of the analysis of market integration via price convergence (equation (4)), as well as the outcomes of the error correction model (equation (5)) that give insights into the development of information efficiency.

Table 3. Results of the time-varying coefficient models
RegionPrice convergenceInformation efficiency
[Equation (4)][Equation (5)]
β Constant α Constant


  1. For the coefficients the final state is provided. Numbers in brackets report the root mean square error for the coefficients and standard errors for the constant. GPL, GASPOOL; NCG, NetConnect Germany; TTF, Title Transfer Facility. *, **, *** indicate significance at the 10, 5 and 1% levels.


NetConnect Germany and TTF show the highest degree of price convergence (0.983) in the final state of the convergence process and a very high speed of adjustment to changes in available market information. This means these highly integrated markets seem to be almost efficient.

Figure 2 shows the development of the β coefficients over time. All three pairs of market zones started with a rather high degree of price convergence (above 0.96)3 that increased moderately after the introduction of the entry–exit regime in October 2007. The coefficients for GPL-NCG and GPL-TTF peaked around July 2008, which coincides with the takeover of BEB by Gasunie. A plunge in the β coefficient for both price relations can be observed right after the establishment of NCG as a merger of E.ON and Bayernets (October 2008). While prices between NCG and GPL started to converge again, the gap between GPL and TTF prices decreased only slightly. Concerning NCG and TTF, the price gap decreased more or less steadily until the first quarter of 2009, but increased thereafter. By April 2011, the β coefficient of GPL-NCG swung back above the initial level of price convergence, whereas the other two relations approximately reached their starting values. All in all, the Dutch-German gas market appears to be well integrated. However, one might argue that the formations of the β coefficients are mainly explained by the oil price movement as a common exogenous factor driving the gas prices. We believe that such argument seems to be implausible since the decline of price convergence between NCG and TTF started later than in the other two cases. Furthermore, the oil price has already increased since beginning of 2009.

Figure 2.

Price convergence β

However, as pointed out above, market integration in terms of price convergence has to be distinguished from market efficiency. Therefore, the question remains as to whether the observed degree of price convergence is sufficient to allow efficient adjustment to new information. The development of information efficiency is depicted in Figure 3. The coefficient α of the error correction model from equation (5) indicates how fast prices turn back to equilibrium once new information has appeared. The higher the absolute value of α is, the faster prices adjust to new information and the more efficient markets are. The price relation between GPL and NCG shows the lowest efficiency over most of the time period considered. In the first year after the entry–exit system became mandatory, prices adjust pairwise between NCG and GPL to around 10% of new information within one day, which is about half the speed of pairwise price adjustment between GPL and TTF (Figure 3). The information processing between NCG and TTF, being the highest for nearly the entire observation period, may be explained by considerably higher trading volumes at NCG compared with GPL.4 The churn rate of NCG, measuring the ratio between traded and physically delivered volumes, has risen from 1.6 in September 2007 to 3.2 in September 2010, thus becoming closer to the churn rate of the Dutch TTF.5 For GPL-TTF and NCG-TTF, a major increase in the speed of information processing can be observed during the fourth quarter of 2008, shortly after the creation of NCG.

Figure 3.

Information efficiency (α)

The zero value of α between GPL and NCG around January 2009, which indicates that prices did not adjust to new information in any way, can be explained by the gas conflict between Russia and Ukraine. On 1 January 2009, Russia stopped gas delivery to the Ukrainian transmission network completely. Russia accused the Ukraine of consuming gas illegally, importing gas that had been designated for transit to Germany and other European countries. Since the EU imports approximately 25% of its gas from Russia, with the main transit route via the Ukraine, the suspension of delivery resulted in gas shortages in Central and Southeast Europe, which necessitated a rearrangement of gas flows. While cross-border gas transport only changed the direction, the rearrangement within Germany led to temporary inefficient pricing behaviour. Nevertheless, after GRTgaz, ENI and GVS joined NCG in October 2009, the information processing between GPL and NCG dropped to zero again. At the same time, information efficiency between NCG and TTF increased considerably up to approximately 100%. Both developments indicate the relative importance of NCG over GPL.6

To summarise, although price convergence did not substantially increase since the mandatory introduction of the entry–exit system, information efficiency with regard to cross-border trades has increased significantly over the following 3.5 years. As Barrett and Li (2002) point out, market integration has to be distinguished from efficiency issues in spatial price analysis. Regarding the price relations between the two major German market zones for natural gas as well as the connection to the Dutch TTF, markets are sufficiently integrated to provide for an improved processing of new information. Prices between NCG and TTF adjust within one trading day.

6. Conclusions and Policy Implications

The aim of this article was to study the development of market integration and efficiency in Germany after the mandatory introduction of an entry–exit network pricing regime in the natural gas market. We applied Johansen's cointegration analysis and a time-varying coefficient approach (Kalman filter) to test for price convergence. The state space model is extended to an error correction model to analyse how fast prices are able to adjust to new information, i.e. how efficient the markets are.

We explicitly account for transportation costs to test the spatial arbitrage condition using price series for the two major German market areas, NCG and GPL, as well as data from the Dutch TTF hub as a competitive benchmark. Results of the Johansen approach show a level of price convergence close to one between all three locations, with the Dutch TTF – as the more mature market – leading the pricing behaviour at both German hubs. The time-varying coefficient model reveals lower levels of convergence, which are nonetheless sufficient to indicate an improved processing of new information between the larger German hub NCG and TTF. Since the mandatory introduction of the entry–exit system, information efficiency has increased significantly, especially after the merger of the two market zones resulting in the foundation of NCG in October 2008 and the additional enlargement of the NCG consortium in October 2009. Prices adjust to new information within one trading day.

Future research may extend the analysis to entire Europe. Moreover, it could be interesting to analyse different regulatory regimes in terms of their effect on the functioning of the respective wholesale markets. And a special focus should be put on the importance of transport capacity.


We thank Jörg Breitung, Felix Höffler, Reinhard Madlener, Sebastian Nick, Heike Wetzel and Georg Zachmann and two anonymous referees for helpful comments and discussions. Financial support from Bundesnetzagentur is gratefully acknowledged.


  1. 1

    GASPOOL has been renamed several times over the considered period. We use the current name GASPOOL for this market zone throughout this article to avoid any confusion.

  2. 2

    This is confirmed by descriptive statistics, which are available from the authors upon request. For example, the standard deviation, as a proxy for volatility, is around 0.40 €/MWh (in log terms) for all three regions.

  3. 3

    The high value at the beginning is not sensitive to the assumption of E(β0) = 1. Setting E(β0) = 0 reveals similar results.

  4. 4

    Ninety-five% of natural gas trading volume at EEX is related to the market area of NCG, only 5% to GASPOOL.

  5. 5

    The number 1.6 is related to the market zone of E.ON as a precursor of NCG. Compared to the British National Balancing Point (NBP), the most liquid hub in Europe, churn rates in Continental Europe are still significantly lower. NBP has a churn rate of around 10.

  6. 6

    The drop back of α to zero for GPL and NCG during the second half of the observation period can also be interpreted differently. Since the price statistics in Table 2 show few price differences exceeding transportation cost differentials for these two market zones, the development of α might be due to the fact that price arbitraging between the two German market zones is already done intraday. This hypothesis goes against the straight forward interpretation of α as an indicator for information efficiency. We thank an anonymous referee for this comment.