International Market Integration for
Natural Gas
Stan Dorhout Mees
University of Groningen
Abstract: In this paper we investigate whether the global market for natural gas has evolved into a well integrated market with a uniform price or if it is still a regionally segregated market. We use monthly gas prices of Europe, Japan and the US from 2008 to 2018. Using the Johansen procedure we find that there is a split between the US market and the European/Japanese market. Intercontinental trade is not yet substantial enough to integrate all markets. Gas prices within Europe and between Japan and Europe are well integrated. In addition to that, we find that oil still is an important factor in determining gas prices.
Keywords: natural gas prices, Law of One Price, market integration, cointegration JEL: C32, F14, L95, Q41
1. Introduction
understand this market thoroughly. The current debate about gas as a long-term energy source and geopolitical issues like Europe’s dependency on Russian gas make the topic very actual.
We will motivate our research on the basis of five observations about the gas market. The first observation is that gas markets on a continent used to be separated. Gas was only traded via pipelines of neighboring areas or countries which led to a constraint on trade. In other words, the constraint on trade was mainly due to a lack of transport capacity. Due to the limited trade, every area had its own gas price which was characterized by the specifics of that particular market. Think of trade patterns, seasonality, transport costs, supplier costs etc. Currently, a lot has been done to integrate gas markets that were formerly separated.
The second observation is that the basis of the gas market is changing a lot. The separation of gas markets was due to physical capacity constraints, legal constraints and due to the fact that gas markets used to be highly vertically integrated. Over the past four decades, governments have tried to address this segregation by introducing reforms. In the US for example, reforms started by addressing the vertically integrated structure of the market around 1985. In Europe, three gas Directives were adopted to open up the gas market. After all these reforms, the vertically integrated structure disappeared, gas-to-gas trade increased and the gas markets became fairly integrated on the continents. This has been confirmed by many studies, such as Asche et al. (2001;2002) for Europe and King & Cuc (1996) for the US.
The third observation is the changing relationship between oil and gas. Traditionally, gas was priced via pricing formulas in long-term contracts that used the oil price as a benchmark for the gas price. Due to the nature of these contracts, changes in gas prices were predominantly the result of changes in oil prices. Historically seen, this relationship was established because oil and gas are substitutes in terms of energy source. The literature about the relationship between oil and gas is abundant (see, e.g., Panagiotidis & Rutledge (2007), Brown & Yücel (2008) and Caporin & Fontini (2017)). In light of our research question it would be interesting to see whether the increased gas-to-gas trade of the past decade has any influence on this relationship.
The fourth observation is that the means of transport for gas across continents have improved dramatically. In the past, trade via Liquefied Natural Gas (LNG1) was not economically viable. Improved techniques have now made trade via LNG economically profitable. The only constraint left is capacity related. A lot of initial investments are required to set up facilities for the liquefication (making gas liquid) and regassification (from liquid to
gas) of natural gas. Global liquefication capacity increased with 103% from 254 billion cubic meters (bcm) to 516 bcm from 2008 to 2018. Current projects are to increase the total capacity with another 123 bcm. Literature on the influence of LNG on trade is limited, see e.g. Neumann (2009).
Now the question arises why trade via LNG is growing so fast. The need can be explained fairly simple. As said before, the markets around the world used to be separated which entailed that there were large price differences. A new means of transport could therefore enable huge arbitrage opportunities. Countries with low gas prices would suddenly become able to sell their gas at much higher prices on other continents. This would mean that continents are no longer segregated and that a global gas market will emerge. There is little literature available about this matter and some literature is contrasting (see, e.g., Siliverstovs et al. (2005) and Aruga; (2016)). Moreover, the existing studies that consider global market integration of the natural gas market examine older time samples than we will make use of.
The fifth observation is the actuality of former literature. The existing literature on gas market integration and the relationship between oil and gas does not make use of contemporary data. The most recent studies have datasets up until 2012, but most studies are using data until the mid-00’s. Our research will therefore provide an update on the existing literature.
The main research question is whether the global natural gas market has evolved into a well integrated market with a uniform price or if it is still a regionally segregated market. We will approach our research question in the following way. We use monthly data of gas and oil prices. We have data of France, Germany, the LNG import price of Spain, the Henry Hub (US), the export price of LNG from the US, the Japanese import price of LNG, the Brent crude oil and the West Texas Intermediate (WTI) crude oil. In order to test whether the markets have become more integrated we will make use of the Johansen cointegration procedure (see, Johansen (1995)).
Cointegration looks at whether two price series move similar in the long-run. If two price series move similar in the long run, it must be that prices are determined jointly in one market or by the same factor. Such joint price determination may for example be the result of arbitrage. By comparing the gas prices at the different continents, we will be able to check if the markets have become integrated.
that the split between the US market and the rest of the world is starting to disappear. In addition to that, we find that the oil price is still a dominant factor in determining the gas price.
The paper will be organized as follows. The second section will give a broad view of the gas market as it is today. The third section will discuss the relevant literature. The fourth section will argue wat our hypotheses are. The fifth section will explain the methodology we will use. The sixth section will describe our dataset. The seventh section will discuss our results. Finally, the eighth section will conclude our findings.
2. Gas markets
We will start with a brief description of how prices in the gas market are established. The basic price of gas is not only determined by supply and demand. Natural gas is a raw material which has to be extracted from the earth, but more important, gas is a fossil fuel. For fossil fuels, the opportunity costs are also important in determining the basic price. Opportunity costs are the costs of producing one unit of gas now and not being able to supply that unit in the future, which leads to missed revenue in the future. An optimal path of depletion should be derived to determine how much gas should be extracted in a certain period.
Let us now turn to how prices are established in the gas market. Gas used to be traded via physical hubs with long-term forward contracts. Those long-term forward contracts contained pricing formulas that linked gas prices to substitutes for power generation, in most cases a crude oil. These long-term forward contracts made the gas market very inflexible. Nowadays, the system for gas prices has become much more flexible. So-called entry and exit zones have been established which allow for the existence of virtual hubs instead of physical hubs. The virtual hubs increase competition because trade becomes easier, the traders are not responsible for the physical gas flows. Virtual hubs have helped establishing integrated gas markets on continents. Please bear in mind that trade is still limited to physical capacity constraints, e.g. pipeline capacity.
We will now look at the supply and demand of gas. Global demand for natural gas has grown with 3%2 in 2017. About 40% of global demand for natural gas is related to power generation. Despite power plants getting more efficient, it is expected that the share of gas used
2 All data in this chapter is provided by the International Energy Association (IEA), which is partly funded by
for power generation will continue to grow. The main driver for this growth is China, which has a growing industry and is increasingly switching from coal to gas, because it reduces (air)pollution. Global gas production in 2017 was 3740 bcm and is expected to have grown with 10% by 2023. Especially the United Stated contribute to this supply growth via shale gas, which is a relatively new technique which was not profitable in the past.
In 2017 total demand for natural gas in Europe was 547 bcm of which 31.8% is used for power generation. Europe differs from the rest of the world because its expected demand for natural gas is declining with 0.4% annually over the period 2017 to 2023. The main reason for this decline is the increase in renewables with respect to the power supply.
European production was only 263 bcm in 2017 and is expected to decrease even further. The decrease in production has two main causes. First, the gas fields in the North Sea are slowly depleted and will produce slightly less every year. Second, the Dutch government has decided to stop producing from the Groningen gas field by 2030. This production stop in Groningen will further decrease production with more than 30 bcm. As a result of this, Europe will become more dependent on gas imports which may lead to price increases.
Due to the increasing supply gap in Europe, LNG will become more popular, especially for arbitrage. In 2017 Europe imported 64 bcm of LNG which is expected to grow to 79 bcm in 2023. Most growth comes from the southern European countries. There is another geopolitical reason for a move towards LNG. European countries do not want to be completely dependent from Russia. Russian gas currently accounts for about 30-32% of all imports of natural gas in Europe.
In 2017 total demand for natural gas in the US was 764 bcm of which about 35% was used for power generation. It is expected that demand for natural gas will continue to grow with 1.6% per annum for at least five years. The introduction of shale gas in 2007 had two main influences. Firstly, it led to a big increase in gas consumption and secondly, it led to enormous price drops. In the period 2007-2017 gas consumption increased with 150 bcm, mostly used for power generation. In 2017, natural gas production in the US was 760 bcm. North-America is expected to account for almost half of the increase in natural gas production for the next five years, mainly due to further exploitation of shale gas.
Forecasts predict that Qatar, the US and Australia will account for 60% of total LNG supply by 2023.
In 2017 total demand for natural gas in Japan was 120 bcm of which about 65% was used for power generation. The share used for power generation is relatively high compared to other countries. It is expected that the demand growth for gas will stagnate around 2023. In the past, the natural gas market in Japan was dominated by natural monopolies. From 1995 onwards, the Japanese government decided to liberalize the market.
Though Japan’s LNG imports are expected to decrease it will keep its position as largest importer of LNG in the world by 2023. Japan currently has a regassification capacity of 277 bcm/y. Since 2017, China has surpassed Korea and is now the second largest LNG importer of the world. Japan is switching from their so-called ‘legacy contracts’ to more flexible contracts which is comparable to the developments in LNG contracts in the US. Legacy contracts are long-term contracts that are indexed to inflation, to a variety of other fuel commodities like the Japanese Customs-cleared Crude (JCC) and have a fixed destination.
It has become clear that global demand for natural gas is growing. The increase in demand results in more suppliers. Moreover, there are new techniques that make it economically profitable to extract gas from sources that were previously left untouched, e.g. shale gas. The supply side of natural gas is often geographically distant from the demand side. Due to the geographical dispersion of gas, a higher demand evidently implies more trade. The trade in gas is however somewhat more complicated due to the nature of the product.
In the past, gas was mainly transported via pipelines. This implied that continents where isolated from each other with respect to gas trade. This worldview has changed over the past 15 years due to the rise in trade via LNG. The means of transporting have improved considerably and countries are investing a lot in new capacity of either exporting or importing LNG. The increased trade possibilities via LNG will create new arbitrage opportunities, practically meaning that gas from other continents can now compete with pipeline gas. This should result in a large increase in gas-to-gas trade. Investing in LNG only makes sense if LNG is inframarginal to other suppliers in the merit order. The merit order is a stack of all suppliers from the cheapest to the most expensive. If a supplier is inframarginal, it means that the supplier produces at lower costs than the current price, hence the supplier should be able to sell its product.
As said before, gas prices traditionally are pegged to the oil prices through long-term contracts. An increase in gas-to-gas trade should lead to prices moving more independently from the oil prices. Gas prices should become more uniform around the world. Uniform prices may establish as follows. The increase in supply via LNG, assuming that it is inframarginal, will lead to more competition and especially more competition between continents that did not trade natural gas in the past. Suppliers will send their shipments to the highest bidders, and therefore also compete with pipeline prices on the continent. If LNG is the marginal source of gas and trade capacity is substantial enough, it should, on the margin, drive prices closer together up to the transportation and transaction costs.
3. Literature review
Stigler & Sherwin (1985) define the area of a market as follows: “A market for a good is the area within which the price of a good tends to uniformity, allowance being made for transportation costs.” Let’s now add the composite commodity theorem developed by Hicks (1936). Consider a group of goods with different prices. If the prices of those goods change in the same proportion, the composite commodity theorem states that they can be considered as one uniform good. These two definitions can best describe the foundation of the Law of One Price, which will be the underlying principle in this paper. The Law of One Price basically states that two exact same products should have the exact same price, anywhere on the world, with the exception of transportation costs and quality differences. It is important to stress the difference between the absolute LOP and the relative LOP. If the absolute LOP holds, it means that price differences are so small that they are negligible. If the relative LOP holds, it means that there are price differences, but those price differences (e.g. transportation costs) are constant over time. In this research, we will only consider the relative LOP. We will now continue with literature on cointegration in gas markets, which is abundant. Most literature, however, is dated and does only cover specific areas.
These studies use the Kalman filter, which is a time-varying parameter model. The motivation for using the Kalman filter resides in the fact that the integration of the natural gas market in Europe is still an ongoing process (Renou-Maissant, 2012). Examples are King and Cuc (1996), Neumann et al. (2006), Neumann (2009), Renou-Maissant (2012) and Li et al. (2014).
We will go through the most important papers and look at three main subjects. First, papers concerning market arbitrage and the influence of LNG. Second, papers concerning market integration within and between Europe, the US and Asia. Third, papers which investigate the link between oil and gas prices.
3.1 Market arbitrage
Brown & Yücel (2009) investigate whether the increase in LNG capacity around the world indeed leads to more gas-to-gas arbitrage. They use weekly data from 1997 to 2008, including gas prices of the NBP, the HH, WTI and the Brent. The authors apply the Johansen test and find that gas prices are cointegrated with the oil prices which is an expected result. A conflicting result with papers like Siliverstovs et al. (2005) and Li et al. (2014) is the fact that they find the gas prices of the NBP and the HH to be integrated as well. They suggest that there are coordinated prices across the Atlantic, pointing towards LNG trade. This explanation is contradictory, since there was hardly any capacity to export LNG in the observed period according to Brown & Yücel (2009). In other words, arbitrage at a large scale was impossible at that time.
3.2 Market integration: Europe, US & Asia
Siliverstovs et al. (2005) compare import prices of natural gas for Europe, the US and Japan. They use monthly data from 1993 to 2004. In order to make a good distinction they use both LNG prices and pipeline prices. In addition to that the Brent crude oil price is added. The oil price is added to reveal any gas-to-oil indexation(s) which may be present in the long-term contracts. The authors apply a principal component analysis and the Johansen framework. Natural gas prices within the US show a great level of market integration. In Western-Europe there is also a high degree of market integration. Moreover, they also find a strong relationship between the Brent oil price and the gas prices in Europe, which according to Siliverstovs et al. (2005) reveals the presence of oil indexation in the pricing formulae and the dominance of long-term contracts. Third, they find that the gas prices between Japan and Europe are integrated, though the Law of One Price does not hold. The authors state that a good candidate for the price difference could be the transport costs which are assumed to be higher for Japan due to its geographical location. The last result, which the authors state as ‘most important result’, is that both Europe and Japan show no cointegration with the US. They attribute this expected result to the limited possibilities for arbitrage and the higher level of competition within the US at that time.
Li, Joyeux and Ripple (2014) investigate convergence between the natural gas markets of the US, Europe and Asia. The authors use data from 1997 to 2011 and apply the Kalman filter. The most interesting result they find is that the US market is separated from the European and Asian market, which is in line with the findings of Siliverstovs et al. (2005). Following Li et al. (2014), this separation is due to the increase in shale gas production, but the lack of export facilities which drives the prices down. Li et al. (2014) argue that the market integration between the Asian and the European market is due to their linkages to the oil price and not due to arbitrage opportunities.
3.3 Market integration: US
Around 1985 the US started with the deregulation of the national gas market. One of the oldest papers on gas market integration in the US is the paper of De Vany & Walls (1993). Their dataset is from just after the start of the reforms. The first regulatory change was an open access policy for pipelines entailing an increase competition. De Vany & Walls (1993) use daily data from 1987 to 1991 and apply the Engle and Granger procedure. They find that from the beginning of their sample period to the end, market integration grew from 46% to 66%. The authors argue that this increase can be contributed to the increased possibilities for arbitrage which were enabled by the reforms. With the removal of the intermediary dealers, buyers and sellers could trade directly and gas could be drawn from more fields, e.g. the cheapest production field.
King and Cuc (1996), examine the degree of market integration in the US with data from 1986 to 1995. They apply the Kalman filter and find that the price deregulation has increased the market integration, which is in line with the findings of De Vany & Walls (1993). Furthermore, King and Cuc (1996) state that the market still suffers from a lack of transport capacity, resulting in a split between East and West. Due to this lack of transport capacity, the Law of One Price does not hold for the entire North-American market during the observed period. Nevertheless, the deregulation has increased competition which is beneficial for both consumers and suppliers. Consumers can buy their gas at a lower price and suppliers can improve the price they receive irrespective of their location.
In a more recent paper, Park et al. (2008) compare eight North-American gas spot prices including one in Canada, over the period 1998-2007. They find that the gas spot prices are well integrated using the Johansen test. With this result the authors prove that the high deregulation of the market, from 1985 onwards, was successful. They attribute small differences due to seasonality, since the differences between seasons are rather big in the US. Studies like King and Cuc (1996) and Cuddington and Wang (2006) that examine data from just after the start of the deregulation still find a split in the North-American market between east and west.
3.4 Market integration: Europe
The first Gas Directive (98/30/EC), which was adopted in 1998, aimed at introducing competition to the markets by opening them up. At that time, the gas market was dominated by long-term bilateral contracts.
The second Gas Directive which was adopted in 2003 (2003/55/EC), focussed on changing the traditional structure of energy markets. The gas market used to be vertically integrated, meaning that gas producers were also the suppliers to the end-users for both industrial and domestic consumers. The goal of the second Gas Directive was to make sure that consumers could freely choose their gas supplier. The EU tried to unbundle the vertically integrated structure of the market by separating production and supply from transport networks. In practice, there were still a lot of barriers remaining that hindered free competition. Therefore, a third Gas Directive was adopted in 2009 (2009/73/EC). The third directive focused on further unbundling of the system and improve legislation to remove any barriers remaining.
Asche et al. (2001,2002) investigate the European market integration of the gas market before the reforms by analyzing France and Germany respectively. The authors apply the Johansen test to monthly import prices of gas in France (Germany), imported from the Netherlands, Norway and Russia. The data used is from 1990 to 1997. Asche et al. (2001) find that the French import prices are integrated and find no evidence to believe that the Dutch and Norwegians receive a price premium relative to the Russians. Furthermore, they find that there is a close relationship between the French market and the Belgium and German market. The authors state that there is a high correlation between prices of different suppliers in different commodity markets, e.g. the oil market, suggesting the delivery contracts must be rather similar. Asche et al. (2002) find similar results for Germany, the import prices are well integrated. Russian gas, however, is systematically sold at a lower price. The authors argue that Russian gas is systematically sold at a lower price due to the inflexibility of supply. Russian gas has to be transported over enormous distances and it would require extra capacity to be able to provide flexibility of supply (swing). The constant lower price of the Russian gas compared to the relatively high prices that the Netherlands and Norway receive for providing swing is the reason for the LOP not to hold for Russia.
Neumann et al. (2006) have a similar research question as Robinson (2007). The authors use daily day-ahead bid prices of natural gas from the NBP, Bunde and Zeebrugge. The observed period is from 2000 to 2005 and the authors make use of the Kalman filter. Special attention is given to the ‘Interconnector’, which is a pipeline that opened between the UK and Belgium. The authors find strong convergence between NBP and Zeebrugge, moreover they are able to see strong diversions of this convergence on dates when the Interconnector was closed due to technical issues or maintenance. These diversions stress the importance of arbitrage and the physical capacity constraints under which arbitrage is subject. Interesting is that they do not find evidence for market integration between the continental prices, suggesting that the European market at that time was not yet functioning properly as a deregulated market. Renou-Maissant (2012) investigates the cointegration and convergence of industrial gas prices for Italy, France, Belgium, Germany, Spain and the Netherlands. Bi-annual industrial gas prices from 1991 to 2009 are used and both the Johansen procedure and the Kalman filter are applied. The authors find strong convergence in the observed period. Moreover, the authors find the markets to be integrated, but only starting in the mid 00’s. Belgium is an exception as the market is not integrated with the rest. The Law of One Price seems to hold the best between France and Italy, where industrial consumers pay nearly the exact same price for gas.
The most recent paper on market integration for natural gas in Europe is from Asche et al. (2013). The authors look at the oil price (Brent) and the spot price of natural gas at the trading hubs in the Netherlands, UK and Belgium. They apply the Johansen test to monthly spot prices from 1999 to 2010. Asche et al. (2013) find the prices to be highly integrated. More specifically, they find the Brent oil price to determine all the spot prices. From this finding, the authors conclude that the dominating pricing mechanism in Europe is still determined by long-term gas contracts. However, they do not find evidence for the Law of One Price to hold, which may reside in the fact that there might be separate price determination processes.
3.5 Link between oil and gas
Still today, there are long-term contracts in the gas market. In those contracts, gas is often priced via a pricing formula. The main component in those contracts has always been the oil price and it still is today, though it is getting less influential. Traditionally, this relationship between gas and oil has been established because the two are substitutes for power generation. European gas prices are often pegged to the price of Brent crude oil. Japanese prices are often pegged to the Japanese Customs-cleared Crude (JCC), which is an average of the import prices paid on oil. The JCC is also known as the Japanese Crude Cocktail. American gas prices are often pegged to the West Texas Intermediate (WTI), which is another type of crude oil.
Panagiotidis & Rutledge (2007) use the Johansen test to show that during their entire sample period 1996-2003, there was a high cointegration between UK gas prices and the Brent crude oil price. Their result was even robust during the opening of a new pipeline between the UK and continental Europe, the ‘Interconnector’, which enabled new arbitrage opportunities. Oil is still an important factor in pricing natural gas, but it cannot be taken as a rule of thumb to price gas, which was the case in the 90’s. Taking into account the weather, seasonality, crude oil prices, production disruptions and storage seems to fairly explain the largest part of natural gas prices (Brown & Yücel, 2008). Moreover, there is evidence that suggests that the natural gas price adjusts to changes in the crude oil price and not vice versa.
The link between oil and gas prices seems to be weakening a little bit due to increased gas-to-gas arbitrage opportunities. Caporin & Fontini (2017), investigate this matter by applying the Johansen test to prices and quantities of oil and gas between 1997 and 2013. One of the main suggested reasons for the weakening relationship between oil and gas is the production of shale gas, which was first traded at the Henry Hub in 2007. Production has increased dramatically, in 2018 it accounted for 62% of total gas production in the US according to the Energy Information Association (EIA). Caporin & Fontini (2017) introduce a break date, in line with the Perron (1997) procedure, at the moment that shale gas first entered the market. Surprisingly, they find that the impact of oil prices on gas prices doubled after the introduction of shale gas. The authors have no fundamental answer to this unexpected result, but argue that it might be due to the similar technology that is needed for both shale gas and oil production. The influence of technical developments on the link between oil and gas has been researched by Hartley & Medlock (2014).
generation technology and external factors like the weather. In addition, they also find a small exchange rate effect which stresses the influence of fiscal and monetary policy on relative energy prices.
To conclude this section we will summarize the most relevant findings. Increasing arbitrage opportunities seem to be the most important factor for markets to get integrated. In the US, Europe and Asia, a lack of arbitrage due to both legal and capacity constraints seem to have slacked market integration. The enormous growth of shale gas has not only led to lower gas prices in the US, but has also started opening up markets that were previously secluded due to their geographical location. Still today, gas prices seem to be dominated by long-term contracts with pricing formula’s that are related to oil prices. Technology innovations and the exchange rate are of little effect to the gas prices.
4. Hypotheses
We will now combine the knowledge of chapters two and three with our research question. Has the global market for natural gas evolved into a well integrated market with a uniform price or is it still a regionally segregated market? We will work towards an answer by answering nine hypotheses about our countries of interest. For Europe, previous research has shown that prices within the continent are already fairly integrated and the Law of One Price (LOP) holds. Furthermore, previous research has also shown that the oil price still is an important factor in determining the price. We therefore hypothesize:
Hypothesis 1: Prices within Europe are fully integrated.
Hypothesis 2: The European gas prices are integrated with the Brent crude.
For the US, previous research has shown that the prices of the HH were integrated with the WTI crude. After the introduction of shale gas in 2007, gas-to-gas competition has become fiercer. This may have led to a decoupling from the WTI. We therefore hypothesize:
Hypothesis 3: The prices of the HH were integrated with the WTI crude until 2007.
If we look at previous research, we see that it is suggested that there is a split between the US gas market and the rest of the world. Since we have argued that prices should become more uniform after an increase in trade, we think that the split may already have disappeared. We will test if prices of the HH have become more integrated from 2016 onwards. We choose for 2016, because the LNG export capacity in the US starts to increase from 2016 onwards. In 2016 LNG export capacity was 0.5 bcm, in 2017 it was 14.3 and in 2018 it was 26.5. We therefore hypothesize:
Hypothesis 5: Prices of the HH are not integrated with the European and Japanese gas prices until 2016.
Hypothesis 6: Prices of the HH are integrated with the European and Japanese prices from 2016 onwards.
Since we have argued that LNG is traded globally, it should mean that prices are fairly similar. We therefore hypothesize:
Hypothesis 7: The prices of Spain LNG, US LNG and Japan LNG are integrated.
In respect of Japan, we have two additional hypotheses. Since the Japanese gas prices have always been pegged to the JCC instead of the WTI and Brent, we argue that these prices are not cointegrated. In addition to that, we have another important argument concerning the LOP. Both Europe and the US, mainly consume gas which is delivered via pipelines. Transporting costs for pipelines are very stable. Japan on the other hand, is almost completely dependent on LNG imports. Since LNG transporting costs can fluctuate a lot, depending on the distance and efficiency of the ship, we argue that the LOP cannot hold between Japan and Europe/US. Although we argued that LNG could be a marginal source for Europe, trade still seems not substantial enough to make a difference. We therefore hypothesize:
Hypothesis 8: The relative LOP does not hold between Japan and Europe/US.
5. Methodology
The main statistical relationship we want to prove in this paper is cointegration. In order to test for cointegration we first need to analyse the data, which will be time series price data. In order to test for cointegration, the time series have to be nonstationary. We will make use of the Johansen test to test for cointegration. We prefer this procedure over the Engle and Granger method because the Engle and Granger test is somewhat outdated and has some disadvantages which are discussed later in this chapter. The Kalman filter is a technique that focusses on convergence over time rather than on cointegration and is therefore not the best candidate for answering our hypotheses.
5.1 Stationarity
Since our variables need to be nonstationary, we will first elaborate on stationarity. In econometrics, variables of a time series will follow some stochastic process. In case of a (first-order) autoregressive process this will look like equation (1),
𝑌𝑡= 𝛿 + 𝜃𝑌𝑡−1+ 𝜀𝑡 (1)
where 𝛿 is a constant, 𝜃 is some factor multiplied by the previous value of 𝑌𝑡 and 𝜀𝑡 is the value of the unpredictable component. Formula (1) basically states that the current value 𝑌𝑡 is composed of a constant, a factor multiplied by the previous value and plus some value for the unpredictable component. There are three possible cases for 𝜃. The stationary case: 𝜃 < 1, the random walk case: 𝜃 = 1 and the “explosive” case: 𝜃 > 1. For this research we will assume 𝜃 < 1. We make this assumption because it is not realistic that shocks to the system will remain or increase in perpetuity for 𝜃 = 1 and 𝜃 < 1, respectively.
We will make use of three tests for stationarity. The first test is the augmented Dickey-Fuller test (ADF) which was first developed by Dickey and Dickey-Fuller in 1979. It is a test for unit roots. The null hypothesis of this test assumes there is a unit root (nonstationarity), 𝜃 = 1. The alternative hypothesis assumes there is no unit root (stationarity), 𝜃 < 1. A typical feature of a first-order autoregressive model is that 𝛿 = (1 − 𝜃)𝜇, where 𝜇 is the mean of the series. In case of a unit root, it must therefore be that the constant is zero. Dickey and Fuller (1979), prove that the t-distribution is non-standard in case of a unit root, 𝜃 = 1. Therefore we have to use the special critical values developed for the Dickey and Fuller test. Since the augmented Dickey-Fuller test makes use of multiple lags, we need to determine the optimal lag length. The optimal lag length can be found by using for example the Aikake Information Criterion (AIC). Since we are going to use monthly data, we will test for autocorrelation up to the twelfth lag.
The second test we will use is the Phillips-Perron test (PP). This test was developed in 1988 by Phillips & Perron. It is a variation on the ADF test. The difference resides in the fact that the test does not depend on adding lags in the regression, but instead it changes the DF-statistics to adjust for possible autocorrelation in the errors. The hypotheses of the PP test are the same as the hypotheses of the ADF test. The null hypothesis assumes there is a unit root, whereas the alternative hypothesis assumes there is no unit root.
The third test is the Kwiatkowski–Phillips–Schmidt–Shin test (KPSS). It was developed by Kwiatkowski et al. (1992). In contrast to the ADF test and the PP test this is not a unit root test. The main reasoning behind this test is that a time series is built from three components: a random walk, a deterministic trend and a stationary error term. The null hypothesis of this test states that the variance of the random walk is zero, implying stationarity. The alternative hypothesis states that the variance of the random walk is not zero, implying nonstationarity. Please note that these hypotheses conclude the opposite of the hypotheses from the ADF and PP tests. The KPSS test has its own critical values.
5.2 Cointegration
arbitrage. By comparing the gas prices at the different continents, we can see if they behave similar in the long-run or independently. If they move in a similar way (they are cointegrated), it suggests that there is one uniform market where prices are determined instead of having numerous separated markets. In case the two price series behave independently (not cointegrated), the markets are separated.
We will now take a look at the basic principle of the error-correction model (also: Engle
and Granger approach), which can be used to find cointegrating relationships between time
series. Two time series are said to be integrated if they follow a common trend. A common trend implies that the two variables grow in the same proportions, consider the example in Appendix A2. We can use the common trend principle to find a long-run equilibrium, which works as follows.
Assume you have two time series, 𝑋𝑡 and 𝑌𝑡, which are both nonstationary, or in other words, integrated of order one I(1). Now assume that 𝑍𝑡 exists, see equation (2), which is a linear combination of 𝑋𝑡 and 𝑌𝑡.
𝑍𝑡 = 𝑌𝑡− 𝛽𝑋𝑡 (2)
Normally, when two time series are added or subtracted that are both I(1), the result should also be I(1). But now assume that 𝑍𝑡 is integrated of order zero I(0), or in other words, stationary. If 𝑍𝑡 turns out to be I(0), it must be that the long-wave components in 𝑋𝑡 and 𝑌𝑡 have cancelled out producing a series which is integrated of order zero. The cancelling of the long-wave components implies that the two series are cointegrated. To sum up; If equation (2) holds and 𝑍𝑡 is I(0) while 𝑋𝑡 and 𝑌𝑡 are I(1), than 𝑋𝑡 and 𝑌𝑡 are said to be cointegrated and share a common trend. In such a case the cointegrating vector is (1, −𝛽)′, where 𝛽 is the cointegrating parameter. We will now shortly look at the Engle and Granger test, because it is the basis of the Johansen procedure which we are going to use. Granger (1983) proves that cointegrated series can be approached via an error correction model, in the Granger Representation Theorem. This error correction model is further improved in Engle and Granger (1987). It works as follows, consider again that 𝑋𝑡 and 𝑌𝑡 are I(1) and that equation (2) holds, there exists a cointegrating vector (1, −𝛽)′. From this, 𝑍𝑡 = 𝑌𝑡− 𝛽𝑋𝑡 can be written in an error-correction form such as equation (3)
consistent with the short-run. As said before, it is important for the variables to be non-stationary, since stationary variables may lead to spurious regressions.
There are certain problems with the error correction model of Engle and Granger (Verbeek, 2017). First, the unit root test on the residuals does not exploit all information contained in the dynamic interactions of the variables, it therefore lacks power. Second, if 𝑦1𝑡 is not present in the cointegrating vector, this may lead to inconsistent regression estimations. Third, in case there are more than one cointegrating relationships, OLS will estimate a linear combination of those relationships and not reveal that there are multiple relationships. Fourth, the Engle and Granger test is sensitive to which variable is set as dependent variable, while the causal relationship might be unknown. Olsson et al. (2011) present an intuitive solution, by suggesting to run the OLS regression two times and switching the dependent variable. If the two regression estimates are unambiguous there is not a problem.
5.3 Johansen test
Johansen (1988) proposed a procedure which does not suffer from the problems of the Engle and Granger method. Moreover, the Johansen procedure is able to identify multiple cointegrating relationships if present. The procedure is an extension to Engle and Granger, and built upon a vector autoregressive model (VAR). A VAR model uses the common history of two variables to describe their dynamic evolution (Verbeek, 2017). In the Johannsen framework, equation (3) is generalized to obtain equation (4). This is also known as a vector error-correction model (VECM).
∆𝑌𝑡= 𝛿 + Γ1Δ𝑌𝑡−1+ ⋯ + Γ𝑝−1Δ𝑌𝑡−𝑝+1+ Π𝑌𝑡−1+ 𝜀𝑡 (4) In equation (4), ∆ is the lag operator, 𝑌𝑡 is k-dimensional, Γ is a term that represents the VAR in first-differences and 𝜀𝑡 represents the white noises which are normally and independently distributed with mean zero and covariance matrix ∑. The term 𝑌𝑡 consists of a vector with variables integrated of order one with 𝑟 cointegrating relationships of 𝑌𝑡 that are stationary. Lastly, Π represents
Π = 𝛾𝛽′ (5)
In our case an example of equations (4) and (5) could be as follows for France and Germany: ( ∆𝐹𝑟𝑎𝑛𝑐𝑒𝑡 ∆𝐺𝑒𝑟𝑚𝑎𝑛𝑦𝑡) = ( 𝛿𝐹 𝛿𝐺) + ( Γ11𝐹 Γ 12𝐹 Γ21𝐹 Γ22𝐹) ∗ ( ∆𝐹𝑟𝑎𝑛𝑐𝑒𝑡−1 ∆𝐺𝑒𝑟𝑚𝑎𝑛𝑦𝑡−1) + ( Π1 Π2) ∗ ∆𝐺𝑒𝑟𝑚𝑎𝑛𝑦𝑡−1+ ( 𝜀𝑡𝐹 𝜀𝑡𝐺) (6) where, Π1,2 = (𝛾𝛾1 2) ∗ ( 1 −𝛽) ′ (7)
The Johansen test, estimates equation (5) via maximum likelihood and imposes a restriction on 𝑟 in equation (5). It is proven that the maximum likelihood estimate for 𝛽 is equal to a 𝑘 × 𝑘 matrix of the 𝑟 number of eigenvalues. The final step in the Johansen procedure is to either do the trace test or the maximum eigenvalue test. The trace test assesses the smallest eigenvalues in the 𝑘 × 𝑘 matrix of eigenvalues and tests whether they are significantly different from zero. The alternative test is the maximum eigenvalue test where the hypotheses are 𝐻0: 𝑟 ≤ 𝑟0 versus 𝐻1: 𝑟 = 𝑟0+ 1. Both the trace test and the maximum eigenvalue test have as purpose to determine how many cointegrating relationships there are in the data. Remember that 𝑟 represents the number of cointegrating relationships. Both tests do not have a standard 𝜒2 -distribution, instead they follow an extension of the Dickey-Fuller distributions developed by Pesaran et al. (2000).
Let’s now summarize this section by walking through how we practically use the Johansen procedure. The estimated equations will be for example such as equations (6) and (7). We estimate equations (6) and (7) via maximum likelihood. This produces a matrix of eigenvalues which can than be tested by imposing restrictions on the rank (the number of cointegrating relationships). These tests are the trace test and the maximum eigenvalue test. If both tests reject the null hypothesis of no cointegrating rank, we accept the alternative hypothesis that there is a cointegrating rank present. If and only if a cointegrating rank is present, it is feasible to estimate a VECM. The VECM than provides the cointegrating parameter 𝛽 and the speed of adjustment 𝛾.
in the same uniform market (an integrated market). This may for example be the result of improved arbitrage opportunities that arbitrage away any price differences. The VECM we run if a cointegrating rank exists, then provides more details about the magnitude of the cointegration, e.g. how quickly the two price series respond to price changes of each other.
Another feature of the VECM is that it allows us to test whether the relative Law of One Price holds. The relative LOP holds perfectly if the cointegrating vector (1, −𝛽)′= (1, −1)′. We can test for this by running the VECM with the restriction that the cointegrating vector is (1, −1)′. Subsequently, a likelihood ratio test is performed which results in a test-statistic that has an asymptotic 𝜒2(1) distribution. If the test proves to be significant, it suggests that the model with the imposed restriction on 𝛽 is stable in the long-run. If the model with 𝛽 = 1 is stable, there is reason to believe that the relative LOP holds, or in other words, the relative price difference is constant in the long run.
6. Data
Our dataset contains nine variables plus the euro/dollar exchange rate which are retrieved from the Bloomberg database. The ‘tickers’ can be found in Appendix A3. In case of daily prices, this refers to trading days of oil or gas. Before analysis, all variables are transformed into natural logarithms. The descriptive statistics are reported in Table 1.
For the US we have two time series. The first series are LNG export prices of the US. Prices are reported monthly in US Dollars per thousand cubic feet. I converted these prices to US Dollars per Million British Thermal Units (MMBTU). As can be seen from Table 1 and Graph 2, the series is rather volatile. The second series are daily spot prices of natural gas at the HH. These prices are in US Dollars per MMBTU. For this series I calculated the monthly average of the daily spot prices, since the other monthly series we use are also monthly averages. The US LNG series is from January 2001 to December 2018, hence resulting in 216 observations. The HH series is from January 1999 to December 2018, hence resulting in 240 observations.
Table 1: Descriptive statistics
Variables Mean Max Min Std. Dev. Skewness Kurtosis N US LNG 7.380 16.92 3.462 3.454 0.963 2.802 216 Log 1.901 2.829 1.242 0.434 0.465 1.964 216 Japan LNG 8.945 18.07 2.750 4.352 0.598 2.091 240 Log 2.071 2.894 1.012 0.497 -0.027 1.992 240 France 9.371 18.40 4.184 3.256 0.207 1.963 123 Log 2.174 2.913 1.431 0.367 -0.240 1.869 123 Germany 9.451 18.61 4.139 3.206 0.196 2.013 122 Log 2.185 2.924 1.420 0.359 -0.271 1.956 122 HH 4.559 13.63 1.704 2.213 1.469 5.575 240 Log 1.417 2.613 0.533 0.440 0.365 2.583 240 Brent 11.48 24.65 1.929 5.803 0.361 1.994 240 Log 2.293 3.205 0.657 0.572 -0.441 2.337 240 WTI 11.00 25.21 2.200 5.048 0.315 2.154 240 Log 2.278 3.227 0.789 0.515 -0.472 2.411 240 Spain LNG 8.315 12.60 4.770 2.224 -0.128 1.597 104 Log 2.080 2.534 1.562 0.285 -0.374 1.728 104
Gas prices of France and Germany are very much alike which can be seen in Table 1 and Graph 1. We have daily day-ahead spot prices for France and Germany. We will use Germany, France and Spain LNG as a proxy for Europe. Prices are quoted daily in euros per megawatt hour and are converted to US Dollars per MMBTU. After converting the monthly averages are calculated. The French series is from August 2008 to October 2018 and the German series is from August 2008 to September 2018.
Unfortunately, there is only one LNG import price of a European country available. It is the LNG import price of Spain. The dataset is from Bloomberg with data provided by Poten & Partners, which is a very large LNG ship broker. This series contains the average import prices from Qatar, Australia and Nigeria to Spain. This dataset is the LNG Netback data including netback values, gas prices, regasification costs and shipping costs. Prices are reported monthly in US Dollars per MMBTU. Since Spain has the highest regassification capacity of Europe (67 bcm/y) we think that it serves as a good proxy for marginal European LNG import prices. The series is from May 2010 to December 2018.
7. Results
7.1 Unit roots
Since our data has to be nonstationary in order to perform the Johansen procedure we start with testing if our data is stationary. To be sure about the results we perform three tests. The Augmented Dickey-Fuller test (ADF), the Phillips-Perron test and the Kwiatkowski– Phillips–Schmidt–Shin test (KPSS). The ADF test and the PP test are unit root tests with the null hypothesis stating a unit root is present. The KPSS test is a test that looks at the variance of the random walk component. The null hypothesis of the KPSS assumes stationarity, note that this is opposite of the ADF and PP tests. All three tests require a lag length which is chosen via Akaike’s information criterion (AIC), Schwarz’s Bayesian information criterion (SBIC), and the Hannan and Quinn information criterion (HQIC). We test for both stationarity without a trend and with a trend. First-differences are calculated by subtracting the lagged value form the current value. The results are reported in Table 2.
The results are as expected, all data is nonstationary. Take a look at the ADF test for US LNG in levels with a constant. A t-statistic of -1.778 is reported. This implies that the null hypothesis of nonstationarity cannot be rejected, since the test is insignificant. Complementing this result is the ADF test result of US LNG in firsdifferences with a constant. The reported t-statistic is -10.114 and significant at 1%. In this case we reject the null hypothesis of nonstationarity and accept the alternative of stationarity. This is in line with theory as nonstationary data can be made stationary by first-differencing.
3 Augmented Dickey-Fuller test. H0: there is a unit root (nonstationarity), Ha: absence of a unit root (stationarity) 4 Phillips-Perron test. H0: there is a unit root (nonstationarity), Ha: absence of a unit root (stationarity)
5 Kwiatkowski–Phillips–Schmidt–Shin test. H0: stationarity, Ha: nonstationarity
Table 2: Unit root tests
Variables ADF3 PP4 KPSS5
Levels First-differences Levels First-differences Levels First-differences constant trend constant trend constant trend constant trend constant trend constant trend US LNG -1.778 -1.580 -10.114*** -10.137*** -2.624* -2.773 -20.111*** -20.086*** 1.50*** 0.757*** 0.057 0.033 Japan LNG -2.198 -2.386 -5.924*** -5.969*** -1.974 -1.735 -8.367*** -8.367*** 4.00*** 0.92*** 0.236 0.077 France -1.926 -1.769 -6.391*** -6.418*** -1.993 -1.892 -12.084*** -12.151*** 0.83*** 0.351*** 0.108 0.096 Germany -2.027 -1.939 -4.488*** -4.514*** -2.019 -1.829 -10.774*** -10.862*** 0.716** 0.286*** 0.141 0.112 HH -2.829* -3.189* -15.285*** -15.288*** -2.770* -3.086 -15.285*** -15.288*** 2.62*** 1.52*** 0.11 0.054 Brent -2.571* -2.211 -9.138*** -9.256*** -2.798* -2.240 -11.450*** -11.607*** 4.55*** 1.41*** 0.337 0.038 WTI -2.600* -2.386 -9.533*** -9.609*** -2.866 -2.487 -11.371*** -11.480*** 4.08*** 1.43*** 0.263 0.034 Spain LNG -0.939 -2.332 -4.028*** -4.083*** -0.889 -2.064 -11.927*** -11.979*** 1.61*** 0.312*** 0.204 0.124*
7.2 Bivariate gas-gas results
The results of the bivariate cointegration analysis with respect to the gas market can be found in Table 3. In all tables one, two or three asterisks report the significance levels of 10%, 5% and 1%, respectively. We will start with explaining how to interpret the results. The first term Trace reports the trace statistic for the trace test within the Johansen cointegration procedure. The null hypothesis of the trace test is that there is no cointegrating rank between the two variables, see equation (5) 𝑟=rank. If rejected, the alternative hypothesis is accepted which assumes there is a cointegrating rank between the two variables. The second term Max reports the max statistic of the maximum eigenvalue test within the Johansen cointegration procedure. This is an alternative test to the trace test and but has the same hypotheses as the trace test. For both tests the optimal lag length was selected using Akaike’s information criterion (AIC), Schwarz’s Bayesian information criterion (SBIC), and the Hannan and Quinn information criterion (HQIC).
In case the null hypothesis of no cointegrating rank is rejected and the alternative hypothesis is accepted, it is appropriate to estimate a Vector Error-Correction Model (VECM). The optimal lag length for the VECM was selected using the AIC, SBIC and HQIC. The VECM reports amongst others the speed of adjustment and the cointegrating vector. This brings us to the third term, 𝛽̂, which represents the estimated cointegrating vector (1, −𝛽)′, see equation (2). The vector is normalized to one for the time series in the row. The number in parentheses is the standard deviation. A 𝛽 of 1 would imply that the price difference between two series is always exactly the same, or in other words, that they are fully cointegrated.
Table 3: Bivariate Johansen test results gas-gas
Germany France US LNG HH Spain LNG
France Trace6 H0: r=0 73.36*** Max7 H0: r=0 69.27*** 𝛽̂8 1,-1.02 (0.01) LOP9 4.069** γ10 -1.046*** US LNG Trace H0: r=0 20.73*** 20.69*** Max H0: r=0 16.93** 16.79** 𝛽̂ 1,-1.91 (0.25) 1,-0.94 (0.25) LOP 0.096 0.043 γ -0.228*** -0.217*** HH Trace H0: r=0 15.88** 16.94** 14.40 Max H0: r=0 11.87 12.33 8.27 𝛽̂ LOP γ Spain LNG Trace H0: r=0 35.57*** 33.57*** 21.33*** 8.90 Max H0: r=0 34.48*** 32.54*** 20.65*** 7.79 𝛽̂ 1,-0.83 (0.05) 1,-0.82 (0.05) 1, -1.41 (0.20)11 LOP 6.235** 7.494*** 3.627* γ -0.292*** -0.287*** -0.370*** Japan LNG Trace H0: r=0 27.12*** 26.20*** 14.15 17.01** 24.69*** Max H0: r=0 22.55*** 21.93*** 11.22 14.05 24.11*** 𝛽̂ 1,-0.97 (0.09) 1,-0.93 (0.09) 1,-1.12 (0.06) LOP 0.041 0.355 3.452* γ -0.078*** -0.079*** -0.194***
*, ** and *** denote significance at 10%, 5% and 1%, respectively.
This table presents the results of the Johansen procedure. Trace and Max report the test statistics of whether there is a cointegrating rank present, if so, the price series are cointegrated. In case of cointegration, we estimate a VECM which provides the speed of adjustment and the cointegrating vector. The cointegrating vector is normalized to 1 for the price series in the row and 𝛽̂ corresponds to the price series in the column. The LOP is
a test where the VECM is ran with a restriction on the cointegrating vector.
6 Trace test statistic. Critical values 5% and 1% are 15.41 and 20.04, respectively.
7 Maximum eigenvalue test statistic. Critical values 5% and 1% are 14.07 and 18.63, respectively.
8 Unrestricted estimate of cointegrating vector, normalized at the row variable, with standard deviation reported
in parentheses.
9 Likelihood ratio test for relative Law of One Price with a 𝜒2(1) distribution.
10 Speed of adjustment.
The fifth term, 𝛾 represents the speed of adjustment (also: error-correction term) towards the long-run equilibrium, see equation (5). We have monthly data, so in our case it will represent the speed of adjustment between two months. This term should always be negative since the prices are converging towards a long-run equilibrium. A positive term would imply that the prices are diverging from the long-run equilibrium.
We will now look at France and Germany to illustrate the explanation above. The trace statistic is 73.36 and significant at the 1% level. The max statistic is 69.27 and significant at the 1% level. Both the trace and max statistic suggest to reject the null hypothesis, hence we accept the alternative hypothesis and assume there is a cointegrating rank between the two variables. The cointegrating vector, 𝛽̂, is (1, −1.02)′ and the standard deviation is 0.01. The cointegrating parameter 𝛽 is -1.02 which implies that a 1% change in the price of natural gas in Germany is met with a 1.02% change in the price of natural gas in France in the long-run. The test for the relative LOP reports a test-statistic of 4.069 which is significant at the 5% level. The null hypothesis is rejected and we accept the alternative hypothesis which suggests that the relative LOP holds. The speed of adjustment towards the long-run equilibrium, 𝛾, is -1.046. This means that the adjustment in the previous period was 104.6. In other words, a price change in Germany is met by a 104.6% change in the price of France, it overreacts since it is more than 100%.
we can conclude that the Gas Directives of the European Council have stimulated the opening of the market for gas successfully.
The bivariate results for the Henry Hub with Germany, France, US LNG, Spain LNG and Japan LNG are not as expected. For all five combinations there does not seem to be a cointegrating relationship as the null hypothesis is not rejected. Our hypothesis was based on the fact that the new arbitrage opportunities via LNG would have opened up the market. The separation of the US market with the rest of the world is however, in line with Li et al. (2014) who find a split between the US market and the European and Asian market. As can be seen from Graph 2, the prices at the Henry Hub are moving rather independently from oil and other gas prices after the introduction of shale gas in 2007. From 2007 onwards, the US became independent of gas imports which led to a decrease in prices. As can be seen from Graph 2, prices are consistently lower. Though the export price of American LNG rose to Japanese and European levels from 2007 onwards, there was still very little capacity to export via LNG, currently only 28 bcm. The lack of export capacity is the best candidate to explain why the markets are still separated from each other. In theory, prices should become integrated once arbitrage opportunities will no longer be constrained by capacity, but this is something the future will tell us.
The Japanese gas prices are found to be integrated with the three European prices. This is an expected result since Japan is almost fully dependent from gas imports and the prices of LNG imports seem to be comparable with the European prices due to the high levels of arbitrage. Furthermore, this result is similar to Siliverstovs et al. (2005) and Li et al. (2014) who have proven similar relationships. Li et al. (2014) however argue that this integration is driven by the oil-linked contracts instead of by arbitrage opportunities. Given that both Europe and Japan are net importers of LNG, the reasoning of Li et al. (2014) seems reasonable, because this implies that there is no direct trade between Japan and Europe. The second result is that the relative LOP does not seem to hold between Japan and Europe, which is again similar to Siliverstovs et al. (2005) and confirms our eighth hypothesis. Please take note that the relative LOP looks at the stability of price differences due to transporting costs, quality costs and so on. Within this definition there is one obvious candidate that might be the reason for the relative LOP not to hold between Europe and Japan. In Europe, transport costs are relatively stable due to the fact that still the largest amount of imports go through pipelines. Japan on the other hand is almost fully dependent on LNG imports, which have a very fluctuating transport cost due to boat size, boat efficiency and distance. We can therefore conjecture that transport costs may be a good candidate to explain why the relative LOP does not hold.
The last result from Table 3 is unexpected and rejects our seventh hypothesis. The result between Japan LNG and US LNG suggest that there is no cointegrating relationship present. This is odd since both Japan LNG and US LNG do have a cointegrating relationship with Spain LNG. Therefore we performed a multivariate Johansen test, the results are reported in Appendix A4. The results are surprising, because they suggest that there are two cointegrating ranks which implies that all three price series are cointegrated with each other. Moreover, the additional test suggests that US LNG should not be excluded. There is no clear explanation why the results in Table 3 suggest differently, but it may be due to the high deviations in the series of US LNG. Rationally thinking, it makes more sense that US LNG is integrated with Spain LNG and Japan LNG since we have argued that the market for LNG is global. Hence, prices should be fairly equal. We therefore conjecture that hypothesis seven is accepted, despite the contrasting result in Table 3.
Table 4: Bivariate Johansen test results for specific periods HH HH Japan LNG Period 08m1-15m12 16m1-18m12 Trace H0: r=0 9.96 31.44*** Max H0: r=0 9.45 31.44*** 𝛽̂ 1,-2.77 (0.36) LOP 3.420* γ -0.068*** Spain LNG Period 10m6-15m12 16m1-18m12 Trace H0: r=0 4.73 7.92 Max H0: r=0 4.71 6.56 𝛽̂ LOP γ US LNG Period 08m1-15m12 16m1-18m12 Trace H0: r=0 13.24 36.55*** Max H0: r=0 10.81 30.16*** 𝛽̂ 1,-0.40 (0.27) LOP 2.018 γ -0.585*** France France HH Period 08m9-15m12 16m1-18m12 Trace H0: r=0 7.75 12.58 Max H0: r=0 5.89 11.11 𝛽̂ LOP γ
*, ** and *** denote significance at 10%, 5% and 1%, respectively.
starts to increase from 2016 onwards. In 2016 LNG export capacity was 0.5 bcm, in 2017 it was 14.3 and in 2018 it was 26.5.
at the HH are slowly starting to move towards a more global uniform price. It must be noted that the results merely indicate a start of these events since the relationships are not that strong and only represent three years.
The other results in Table 4 show that prices of the HH are not integrated with Spain and France, which partly rejects hypothesis six. The separation of the HH with continental Europe can be explained as follows. The vast majority of gas consumed in Europe comes via pipelines, because LNG capacity is still limited. Prices are therefore mainly determined by pipeline prices which are obviously different than the LNG prices which are determined by many foreign suppliers on other continents. Assuming that the supply via LNG is inframarginal, it is not (yet) substantial enough to integrate the markets.
7.3 Bivariate oil-gas results
The results in Table 5 can be interpreted the same as in Table 3, but in Table 5 we focus on the relationship between gas and oil prices. Germany has been excluded from the analysis since France and Germany have proven to be virtually the same. We have included both the Brent crude and the West Texas Intermediate. Confirming Brown & Yücel (2008) we find that the gas prices seem to respond to changes in the oil prices and not vice versa. As expected, the two crudes have a high level of cointegration and the relative LOP holds.
When we look at the bivariate relationships of the two crudes and France, US LNG and Spain LNG we see that there is a cointegrating rank. This result confirms hypothesis two and is in line with many studies such as Panagiotidis & Rutledge (2007) and Hartley & Medlock (2014). We will focus on the Brent crude since this is the crude oil where the European gas prices are pegged to. The high cointegration between Brent and the natural gas prices proves that the gas-to-gas trade still has a limited influence on prices. It furthermore shows that the dominance of oil in the pricing formula of long-term contracts is still extant. The results of these six combinations are quite alike due to the fact that on the one hand France, Spain LNG and US LNG are integrated and on the other hand Brent and WTI are integrated.
Table 5: Bivariate Johansen oil test results WTI Brent France Trace H0: r=0 36.84*** 30.79*** Max H0: r=0 34.20*** 27.06*** Beta hat 1,-1.19 (0.12) 1,-1.01 (0.10) LOP 2.219 0.003 γ -0.256*** -0.266*** Japan LNG Trace H0: r=0 14.24 15.13 Max H0: r=0 9.47 11.20 Beta hat LOP γ US LNG Trace H0: r=0 25.41*** 30.20*** Max H0: r=0 21.14*** 26.31*** Beta hat 1,-0.90 (0.15) 1,-0.74 (0.11) LOP 0.269 3.311* γ -0.166*** -0.213*** Spain LNG Trace H0: r=0 21.32*** 21.33*** Max H0: r=0 19.04*** 20.25*** Beta hat 1, -0.97 (0.12) 1,-0.85 (0.07) LOP 0.049 2.527 γ -0.161*** -0.200*** HH Trace H0: r=0 16.49** 14.26 Max H0: r=0 11.18 9.44 Beta hat LOP γ Brent Trace H0: r=0 74.53*** Max H0: r=0 65.32*** Beta hat 1,-1.08 (0.03) LOP 6.016** γ -0.290***
*, ** and *** denote significance at 10%, 5% and 1%, respectively.
with the Brent or WTI. We would have liked to test for this relationship, since we would expect at least some sort of cointegration, but unfortunately there was no data available.
absolutely sure we, we also performed a multivariate Johansen test, the results are presented in the Appendix A4. The results in Table A4 indicate that there is only one cointegrating rank. Furthermore, the likelihood ratio test with HH excluded is significant at the 1%, indicating that HH may be dropped from the equation. The HH has traditionally been pegged to the WTI with long-term contracts, but since the introduction of shale gas, we argue that the national gas price within the US has decoupled from the WTI. This conjecture is now confirmed by our results.
Since Brown & Yücel (2008) and Caporin & Fontini (2017) do find a cointegrating relationship for their dataset we thought it was insightful to mimic their analysis. The results can be found in Appendix A5, and are in contradiction to the results of the other two authors. We find no cointegrating rank between the two variables, rejecting hypothesis three. We also find no cointegrating rank for the period after 2007, confirming hypothesis four. The difference between our results and the results in previous research may be the result of the fact that the HH deviates significantly from the oil-prices due to weather and storage conditions. These deviations are also reflected by the fact that in only one of the cases in Table A5 the cointegrating rank is found to be significant at the 1 % level. In addition to that, Brown & Yücel (2008) report that the relationship does not hold when they cut the observation period to July 2006.
8. Conclusion
In this paper we investigate whether the global natural gas market has evolved into a well integrated market with a uniform price or if it is still a regionally segregated market. We analyze this question by looking at the gas prices of Europe, the US and Japan. We state that for a market to be integrated, prices should have a stable long-run equilibrium relationship. If this is not the case, it means that there will remain large unstable price differences in the long-run. If price differences remain in the long-run, apparently there are other factors that influence the price. In this light, we also add two crude oil prices to our analysis. In addition to that, we test whether the relative LOP holds between two price series. If the relative LOP holds, it implies that the relative price differences, resulting from e.g. transport costs or quality differences, are stable. In order to test for all of this, we apply the Johansen cointegration test.
We hypothesized that the split between the market of the US and Europe/Japan would have disappeared, but we find that the markets are not integrated. After taking a smaller, more recent sample, we do find some evidence for cointegration between prices at the HH and Japan. We conjecture that the trade via LNG is not yet substantial enough to arbitrage the price differences away, but it is definitely starting to make a difference. Lastly, we find that the Brent and WTI are still important factors in determining the gas prices in Europe and the US. As expected, Japan shows no cointegration with these crudes, as Japanese gas prices traditionally depend on the JCC instead.
Our findings contain a policy implication. It seems that LNG trade on the margin is still too limited to have a substantial price influence. It may therefore be advisable to invest more in LNG import terminals, though the initial investment costs are substantial. Such investments will lead to lower gas prices, because LNG can be bought from the supplier with the lowest price, instead of being tied to what the pipeline has to offer. Especially for Europe, such policies are advisable. Europe is worried from a geopolitical perspective that it will become too dependent on Russia via its gas imports, and LNG trade may help reduce such troubles.
A limitation to this research is that we do not have gas prices from more European countries. Nevertheless, the gas market in Europe is well integrated which justifies our choice for France and Germany as proxy for Europe. Another limitation is that the LNG import price of Spain is an average of LNG imported from Qatar, Australia and Nigeria instead of an average from all import countries. Since Qatar and Australia are in the top 3 of LNG suppliers, it seems to be a good proxy for the LNG import prices of European countries.
