Changing cointegration relationships of oil and gas and
arbitrage opportunities across the Atlantic gas markets
An investigation of cointegration of oil and gas markets in the UK and the US and valuing arbitrage opportunities for LNG suppliers in the period from 1997 to 2012
Mark Casper Kranenborg
Master Thesis
Master of Science in Finance
Student number: s1701894 Date: 8 July 2013
Supervisor: L.J.R. Scholtens
Abstract
This research focuses on the arbitrage opportunities that arise in the Atlantic basin for LNG suppliers. It also investigates the cointegration relationships between the natural gas and oil markets in both the US and the UK during the period February 1997 to December 2012, incorporating structural breaks using the Bai and Perron framework. No evidence is found that gas prices permanently broke away from oil-linked prices. However, cointegration relationships between natural gas, LNG and oil change over time due to ‘regime shifts’. No evidence is found for an emerging global gas market. Despite these results, the arbitrage opportunities decreased significantly after 2008 for LNG suppliers who exploit arbitrage opportunities in the US.
JEL classification
N70 – Transport, International and Domestic Trade, Energy, Technology, and other Services: General, international or comparative.
P28 – Socialist systems and transitional economies: natural resources; energy; environment.
L95 – Industry studies: transportation and utilities: gas utilities; pipelines, water utilities.
Keywords
1. Introduction
This research investigates what happened to the physical arbitrage opportunities across the Atlantic in the period from 1997 to 2012 for liquefied natural gas (LNG) suppliers with an initial buyer in the UK and arbitrage potential in the US. The recent literature mentions that natural gas prices are decoupling from oil (Stern and Rogers, 2011). This decoupling influences the arbitrage opportunities for LNG suppliers in the Atlantic basin because a diminishing oil-link could have an effect on gas pricing differentials across the Atlantic. This investigation strives to assess the economic impact on LNG arbitrage of changing cointegration relations between oil and gas. Three main conclusions can be drawn.
First of all, the arbitrage opportunities for LNG suppliers trying to exploit arbitrage opportunities in the US diminished significantly in the period 2008 to 2012 in comparison with the period 1997 to 2008. The second finding of this research is that cointegration relationships are changing over time, which is probably caused by ‘regime shifts’ (Brigida, 2012). The third conclusion is that the natural gas and LNG prices across the Atlantic show no signs of cointegration from 2008 to 2012, which is an indication that a global gas market is still far from fully established.
Natural gas can be transported in a gaseous state through pipelines, or can be shipped overseas in a liquid state1, called LNG. LNG is currently a small fraction (around 2-3%) of the natural gas market (Maxwell and Zhu, 2011). Moreover, only 4.5% of total LNG trade can be considered arbitrage, so in relation to the whole natural gas market LNG arbitrage is a tiny fraction (Holleaux, 2007). The LNG market is, however, growing relatively fast (Rogers, 2010). One of the reasons for this growth is that more LNG suppliers are entering the market on different geographical locations. Since 2006, five new countries started exporting LNG and eight new countries started importing LNG (International Gas Union, 2011). At the moment, only eighteen countries produce LNG, which they transport to a limited amount of countries. Another reason of LNG growth is the increasing importance of gas2 (Khalilpour and Karimi, 2011). A third reason for increasing use of LNG is a reduction in costs indicating that LNG becomes more competitive with natural gas and relatively cheaper in relation to other energy sources. The costs to produce LNG can, generally, be split in four parts: production, liquefaction, transportation and regasification. According to Cornot-Gandolfine (2005), transportation costs dropped 20-30% since 2000 and liquefaction costs dropped 25-35% due to technical innovations during the period from 1990 to 2000.
Besides that the demand for LNG is growing, the LNG market shows an emerging spot market for LNG trade. In 2011, 25.5% of global LNG trade occurred in !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
1!!Natural gas and LNG are substitutes. Pipelines and marine shipping are also substitutes. Jensen
(2004) states that shorter distances tend to favor pipelines because pipeline costs rise linearly with distance. On the other hand, long distances favor marine shipping because shipping have high threshold costs, but the increase in costs do not relate as strong with distance as with pipelines (Jensen, 2004).!
2 Gas has a strategic advantage in comparison with oil and coal because gas has a lower carbon
the spot market, while this was only 16% in 2006 (International Gas Union, 2011). One of the drivers of this short-term trade of LNG are destination flexible contracts that emerged recently in the LNG market. Originally, long-term take-or-pay contracts3 with destination clauses were used mainly (Jensen, 2004). LNG contracts are nowadays diversifying in pricing formulations, quality and quantity4, and in addition have destination flexibility (Khalilpour and Karimi, 2011; Ikonnikova, 2009) Freedom for the LNG supplier to sell the gas to the buyer where the biggest profits are expected is called destination flexibility in this research. Hence, a LNG supplier with destination flexible contracts is not obliged to deliver the LNG to a predefined buyer or restricted to a certain market and has therefore more freedom to operate globally and link previously isolated gas markets (Jensen Associates, 2007).
Destination flexible contracts and short-term LNG trade lead to a ‘renaissance of LNG’ (Rogers, 2010). This renewed interest in LNG could change the arbitrage opportunities across the Atlantic for LNG suppliers. The question that is therefore researched here is what the impact of oil/gas cointegration and changing gas market characteristics is on physical LNG arbitrage.
In the next section, a thorough discussion of the literature about the cointegration relationships between natural gas and crude oil is provided. I also elaborate on a framework of LNG arbitrage (Zhuravleva, 2009) and discuss the literature that focuses on the potential profits from LNG arbitrage. Thereafter, the empirical model and parameters of LNG arbitrage are proposed, followed by a section concerning the used data and empirical results. The sixth and final chapter concludes this research.
!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
3 Take-or-pay contracts are contracts in which the buyer is restricted to take the cargo or pay a
predetermined penalty for declining it (Jensen, 2004).
4!If LNG demand is higher than expected, the buyer wants to buy additional LNG. However, when
2. Literature review
In this chapter an overview of the literature is given. This chapter is split up in six sections. In the first section, I clarify the concept of cointegration and discuss the literature concerning the cointegration of gas and oil prices in the US and the UK. In the second section, I discuss the literature concerning the cointegration of gas prices in the US and the UK. In the section thereafter arbitrage is defined as used in this research and a framework of LNG arbitrage is presented. The fourth section deals with the literature concerning LNG arbitrage. In the fifth section, I describe the relationships of this research with the existing literature. Finally, the hypotheses of this research and the underlying relationships of the hypotheses are connected in a conceptual framework.
2.1. Cointegration of gas and oil in the Atlantic
Cointegration means that two time series share a common stochastic trend (Brooks, 2008). Interpreted in economic terms, cointegration is an equilibrium relationship between two variables where at least one of the variables has a significant influence on the other (Dahl et al., 2011). Murray (1994) uses the dog owner and its dog as a metaphor for cointegration. A dog can move freely around the dog owner, but it will eventually end up somewhere near the dog owner, even though in the short run this is not necessarily the case. A dog owner and his dog are in equilibrium and cointegrated (Murray, 1994). Cointegration is a widely used concept to test for market interrelatedness in the literature about gas markets (Brooks, 2008). I interpret market interrelation as whether or not two time series ‘move together’ over time, hence whether or not a cointegration relationship between these time series occurs (Brooks, 2008). I now turn to the market interrelation of gas and oil.
There are basically three gas markets (Siliverstovs et al., 2005). The first gas market is located in Europe, which imports gas mainly from Russia, Norway and Algeria. In Europe, several gas hubs are established, ranging from mature and liquid hubs to illiquid and newly established ones. In the UK, the National Balancing Point (NBP), a virtual trading location for UK natural gas and the most liquid gas trading point in Europe, is widely used as a proxy for the gas price in the European market. I use the NBP price as the market proxy for the UK gas market as well5. The second gas market is located in North America that mainly obtains its gas from Mexico and Canada. The Henry Hub (HH), a physical distribution hub in the US, is considered the !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
5
Neumann et al. (2006) test the influence of the NBP on both the Zeebrugge and Bunde gas hub and find that the relationship between Zeebrugge and the NBP is rather strong, while the relationship between the NBP and Bunde remains non-cointegrated. Neumann (2009) claims that these
market price proxy for natural gas in this region6. The third market is the Japan/South Korea market that mainly imports from Indonesia, Australia and the Middle East (Siliverstovs et al., 2005). I am not considering this Asian market further in this paper, mainly for the reason that LNG arbitrage in this region is insignificant (Holleaux, 2007)7. The general view is that the three gas markets are only weakly interrelated: every gas market has its own pricing mechanisms and industry characteristics. This weakly interrelatedness is mainly due to a lack of pipeline connections between these markets and little trade capacity. With an increase in trading between the gas markets, these gas markets can entangle more, resulting in a global gas market.
Most of the prices for energy products are pegged to the oil price. For sure in the past, this was also the case for natural gas. Natural gas and crude oil are substitutes, but also complements and rivals in production. They are rivals because they are substitutes, thus an increase in oil prices affect the demand for gas. However, oil and gas are also complements because gas and oil is found together in oil wells (Ramberg, 2010). Therefore, the relationship between gas and oil is not surprising (Villar and Joutz, 2006). According to Villar and Joutz (2006), there are basically four mechanisms through which oil prices tend to influence gas prices:
1) An increase in crude oil prices motivates consumers to substitute natural gas for petroleum products in consumption, which put upward pressure on natural gas demand and prices.
2) Increases in crude oil prices resulting from an increase in crude oil demand may increase natural gas produced as a co-product of oil, which would tend to put downward pressure on natural gas prices.
3) An increase in crude oil prices resulting from an increase in crude oil demand may lead to increased costs of natural gas production and development, putting upward pressure on natural gas prices.
4) An increase in crude oil prices resulting from an increase in crude oil demand may lead to more drilling and development of natural gas projects, which would lead to increasing gas production and put downward pressure on natural gas prices.
More indirect mechanisms could also cause a link between gas and oil. Ramberg (2010) claims that portions of the manufacturing and electricity generation sectors use machinery with the ability to switch between oil production and natural gas as fuels. Another reason could be that natural gas is used as feedstock for oil refining or oil sands operations (Huntington, 2007). Based on all these mechanisms, it makes sense that there is a connection between oil and gas prices. However, because a lot of countries have liberalized their gas market, more and more gas is priced with gas-to-gas pricing mechanisms that rely on gas fundamentals instead of oil prices !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
6 Serletis and Rangel-Ruiz (2004) specifically test the influence of HH on other North American gas
prices after which they cannot deny that North American prices follow the same pricing pattern as the HH.
7 According to Holleaux (2007), the Atlantic basin attracted around 84% of the arbitrage business in
(Stern and Rogers, 2011). According to Kao and Wan (2009), these fundamentals are, among others, weather, temperature, gas storage levels, pipeline utilization and industry demand. Although gas prices are still priced regionally (unlike oil), it seems that gas markets shift more and more to these gas-to-gas pricing mechanisms. Negotiations of new LNG contracts with improved flexibility clauses are also a driver of this shift towards gas-to-gas prices (Kao and Wan, 2009). I now turn to the empirical findings on the relationship between oil and gas prices.
The literature concerning relationships between oil and gas prices focused solely on the US market during the 1990s. For an overview of the literature, see appendix A. The paper by Asche et al. (2001) was the first paper to investigate the cointegration relationship between oil and gas in Europe. Nowadays, more gas markets have been researched, a multitude of methods to do this are used in the literature and results are mixed. Below, some findings are elaborated on, starting with the findings in the UK.
The literature is ambiguous about the degree of cointegration between the UK gas and oil market. The main differences in the literature entail from the different time spans that are used by the researchers. Panagiotidis and Ruthledge (2007), Josse-Vásquez and Neumann (2006) and the papers by Asche et al (2001, 2002, 2006) find cointegration in the UK market for the period before 2005. Westgaard et al. (2011) verified this cointegration relationship in an empirical study on the relationship between the Gas oil and Brent crude oil futures contracts for the period 1994 to 2009. More interestingly, however, when the timespan was reduced from 2002 to 2009, the existence of cointegration seemed less plausible. This finding of Westgaard et al. (2011) is similar to the conclusions of Bencivenga and Sargenti (2009) who find no cointegration relationship between the NBP and Brent. Dahl et al. (2011) agree with Westgaard et al. (2011) that there was cointegration of oil and gas markets in the UK in the past, albeit nowadays this seems less plausible. Dahl et al. (2011) researched the cointegration of oil and natural gas in the UK using Brent oil and NBP gas and explicitly test for structural breaks. They argue that after a structural break in 2007 the evidence of cointegration is much weaker than in the period before 2007. Economic reasons for this break are that gas is used on a larger scale for electricity generation and spot markets of gas become more liquid (Dahl et al., 2011). By incorporating structural breaks, Dahl et al. (2011) compare the period before and after the structural break and can thus find changes in cointegration relationships. The paper of Dahl et al. (2011) is important for this research, because I incorporate the same structural break approach as Dahl et al. (2011) use, as is explained and justified in the methodology chapter. I incorporate structural breaks because this research tries to answer what the cointegration relationship between gas and oil is, when it changed and, when possible, try to find an explanation for this change. Using structural breaks, a distinction between before and after a structural break is explicitly taken into account and cointegration results for the sub periods can be compared.
a link between oil and gas prices (Vásquez Josse and Neumann, 2006; Dahl et al., 2011). In this research, I search for structural breaks that mark changing cointegration relationships. The remaining part of this section focuses on the literature that concerns the oil and gas cointegration relationships in the US.
The US gas market is more ‘mature’ than the European gas market in the sense that it adopted earlier gas-to-gas pricing mechanisms. However, Hartley et al. (2008) still find evidence for cointegration between oil and gas in the US market for the period 1990 to 2006. They claim, in line with the mechanisms provided by Villar and Joutz (2006), that this relation is indirect due to the fact that fuel oil and natural gas are substitutes as inputs for power generation8. Villar and Joutz (2006) also find evidence for cointegration between natural gas prices and oil prices in the US between West Texas Intermediate (WTI) crude oil and the HH, using a sample from 1989 to 2005. They find that oil prices influence the long-term development of natural gas prices, but vice versa, this does not hold. An argument for this might be that the oil market is much larger in size than the gas market and changes in gas prices do not affect oil prices that much, implying that oil is weakly exogenous in relationship with gas. Brown and Yücel (2008) argue that long-term movements in natural gas prices occur due to changes in crude oil prices, but that in the short-term, natural gas prices are mainly driven by other factors such as weather, seasonality and natural gas storage. In contrast to the foregoing mentioned research, Vásquez Josse and Neumann (2006) find no evidence for a stochastic trend between natural gas and Brent or WTI crude oil prices in the US in the period 1999 to 2005. Hence, they conclude that the US market is determined by gas-to-gas based prices. This is in line with Serletis and Rangel-Ruiz (2004), who investigated cointegration of the US market in the period 1991 to 2001. They argue that the ‘decoupling’ of oil and gas prices is a result of deregulation policies in the US. The literature about the US mentioned so far is mainly ambiguous about the cointegration relationships between oil and gas. Recently, however, two papers by Brigida (2012) and Ramberg and Parsons (2012) investigated ‘regime shifts’ in the cointegration relationships between the HH gas price and the WTI crude oil price, thereby explaining most of the ambiguous findings of the literature (Brigida, 2012). Regime shifts, caused for instance by technological innovations, result in changing cointegration relationships between oil and gas prices (Ramberg and Parsons, 2012). Besides that cointegration relationships are not always stable over time due to structural shifts, volatility of gas and oil prices is another important (and often neglected) factor in the discussion whether gas and oil prices are decoupled or have a cointegrated relationship according to Ramberg and Parsons (2012).
To summarize, most of the literature is ambiguous about the US and UK gas markets and the relationship of oil and gas prices. These contrasting findings result in two alternative perspectives on the market interrelatedness of oil and gas (Dahl et al. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
8!Villar and Joutz (2006) argue that the introduction of combined cycle gas turbines (turbines that can
2011). The first view is that oil and gas prices remain coupled because (1) high volatility of gas prices make producers insist on oil-linked contracts, (2) LNG contracts are oil-linked, (3) energy customers substitute towards cheaper energy sources and, (4) fear of manipulation of gas prices (Dahl et al., 2011; Stern and Rogers, 2011). In contrast, proponents of the opposite view argue that the oil and gas markets are separated or start to move independently because (1) nowadays more alternative energy inputs exist for electricity production, leading to a reduction of pricing weight of oil in gas prices, (2) increased gas-to-gas prices lead to gas prices linked to spot gas prices, and (3) increased liquidity of gas markets reduce the need for an oil link (Dahl et al., 2011). Based on arguments of the second perspective, hence that the cointegration of gas and oil is diminishing, the literature hints that the US gas market is less cointegrated to oil than the UK gas market.
This research strives to answer the question whether the natural gas price is still (as strongly) pegged to the oil price in the second half of the 2000s, compared to the period before, both for the US and the UK market. Supported by the findings of Westgaard et al. (2011), Dahl et al. (2011), Villar and Joutz (2006) and Serletis and Rangel-Ruiz (2004), the hypothesis is that the market separation perspective is correct, hence that natural gas is shifting away from oil-pegged prices to gas-to-gas prices. Moreover, in line with Dahl et al. (2012) and Ramberg and Parsons (2012), I test whether one of the reasons why the findings in the literature are mixed about cointegration relationships between oil and gas is because most of the literature did not incorporate structural breaks. Most of the literature implicitly assumes that the cointegration relationships of oil and gas prices are a stable relationship over the entire time period under study. Therefore, I test for structural breaks and adjust the data for the found structural breaks. Using this methodology, the goal is to show that structural breaks are the ‘missing’ factor in the literature. The market separation perspective results in the first set of hypotheses of this research and are called the ‘diminishing oil link hypotheses’, which can be found in the sixth section of this chapter.
2.2. Gas to gas cointegration across the Atlantic.
system for natural gas is already on its way. Moreover, the cointegration relationship between UK and US gas markets is what Neumann (2009) observes after 2003, when UK and US gas prices tend to move parallel. The observations of Neumann (2009) and Stern and Rogers (2011) lead to the, what I name, ‘global gas market hypotheses’, which can be found in the last section of this chapter.
When a global gas market is realized, it is not possible to make significant profits by selling gas in one market for a higher price than another market. This is the case because market reactions would diminish the price spread and thus decline the profits of arbitrage. However, because the literature suggests the cointegration of the UK and US gas markets is only limited, it seems that exploiting these price spreads is possible. Using LNG trade across the Atlantic basin for instance. In the next section, I discuss the concept of arbitrage in the setting of LNG and gas markets.
2.3. Arbitrage framework
The above discussion about the cointegration of gas prices suggests that there are arbitrage opportunities in the Atlantic gas market. But what is arbitrage exactly? Financial arbitrage is normally defined as “the simultaneous purchase and sale of the same, or essentially similar, security in two different markets for advantageously different prices” (Sharpe and Alexander, 1990). These securities can be a multiple of objects, either financial products such as stocks, but also physical objects such as gold or other commodities. The definition of Sharpe and Alexander (1990) implies that arbitrage is a riskless activity that does not require any capital (Shleifer and Vishny, 1997). LNG arbitrage, however, requires fixed capital and is not riskless due to the high volatility of gas prices and technical constraints such as lack of LNG storage (Holleaux, 2007). Another source of uncertainty that adds to the risky nature of LNG arbitrage is the fact that LNG is a physical commodity and not a financial product. Even though the “the simultaneous purchase and sale” of a LNG cargo is possible, it is not possible to deliver LNG simultaneously in the UK and the US. Exploiting arbitrage opportunities using LNG tanks is for foregoing reasons not similar to what is normally considered financial arbitrage in the literature (Holleaux, 2007). I define arbitrage in the LNG market similar to Zhuravleva (2009), who states that arbitrage is “a physical cargo diversion from one market to another, which offers a higher price”. Following from this definition, this research focuses on physical arbitrage instead of financial arbitrage. Once more, with LNG arbitrage it is impossible to simultaneously deliver the cargo to different places. Therefore, in this research, a time period of one month is considered the period in which physical arbitrage can take place. Diversion of cargo can be classified as arbitrage if the cargo was originally committed to a market and a buyer, but is transported and sold to another market that offers a higher price, within a month time.
!
Figure 1. Main prerequisites and barriers of LNG arbitrage (Zhuravleva, 2009).
US gas market are calculated in this research. Thereafter, an examination whether these spreads are sufficiently large to exploit arbitrage opportunities is executed. However, in order to make the decision to exploit the arbitrage opportunity, other prerequisites and barriers need to be considered.
Zhuravleva (2009) states that a second prerequisite of LNG arbitrage is that trading skills are required before LNG arbitrage can take place. The reason she mentions this explicitly is because the LNG market is complex and non-transparent: most information is scarce and hard to obtain because it is not publicly available. An example of non-transparency is that LNG is not traded as a commodity on exchanges. For the remaining part of this thesis, however, this prerequisite does not influence the results because it is assumed that arbitrageurs have all necessary information.
The third prerequisite of arbitrage, the ability of the trader to exploit the trade, mainly concerns with practical, limiting conditions to LNG arbitrage, such as cargo availability, shipping availability, regasification availability and scheduling constraints (Holleaux, 2007). These, mostly company specific, practical limitations are not taken into account in this thesis, even though they are important considerations in investigating arbitrage opportunities9. I leave these limitations out because these topics fall outside the financial scope of this research10 and the needed data, such as shipping availability is not publicly available. One point is however important for the remaining part of this research. Arbitrage is only possible with cargo that is unrestricted by contractual considerations and when the LNG supplier has the ability to choose freely to which market and buyer they ship a LNG vessel. As already mentioned in the introduction, this freedom of cargo destination is called ‘destination flexibility’ for the remaining part of the thesis. The remaining part of this paper focuses solely on LNG ships with destination flexibility and leaves LNG ships with destination clauses outside the scope.
The fourth prerequisite, the ability to profit from the arbitrage opportunity, is diminished by transaction costs of the arbitrage deal that comprise shipping costs, regasification reservation, quality corrections and other fees. I incorporate transaction costs in my model, as I discuss in the next chapter. To my knowledge, only five authors have tried to calculate the potential profits of LNG arbitrage. In the next section, these authors are examined.
2.4. Empirical findings of LNG arbitrage
The first publication that calculates the potential profits of arbitrage opportunities using LNG trade is Hayes (2006). Using Monte Carlo simulation, he finds that 57% of all future scenarios generate scenarios where the price spreads !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
9
For instance, Holleaux (2007) states that Use It Or Lose It (UIOLI) rules prevent traders doing last-minute cargo diversions that are based on price-signals because the capacity owner has to announce prior for using whether he will use a berthing slot.
10 For instance, there is not much public data on LNG contracts, so for this empirical research it cannot
minus the transportation costs are viable for arbitrage. Hayes (2006) also argues that when cargoes are diverted more (more arbitrage), this would increase the correlation of random changes in the US and UK gas prices (more cointegration), which would result in more narrow spreads between the UK and the US.
Suenaga (2007) updates the work of Hayes (2006) for the situation in the Pacific region where it is not possible to divert pipeline gas to the initial buyer when arbitrage deals are done. In his model, moreover, he incorporates the hedging possibilities of LNG with future contracts for an alternative fuel. Suenaga (2007) verifies the results of Hayes (2006) that LNG arbitrages have high volatility, but he cannot reject the possibility that LNG producers shift towards short-term trading because over 90% of the times, the returns with arbitrage opportunities are higher than without these possibilities. However, the model of Suenaga (2007) does not include transportation costs or other barriers to arbitrage profits such as regasification reservation, quality fees, etc., he solely bases the returns of LNG arbitrage on gas price differentials. To exploit the arbitrage opportunities that Suenaga (2007) shows, more factors need to be taken into account than pricing differences as the LNG arbitrage framework of Zhuravleva (2009) indicates. Suenaga (2007) nevertheless gives a valuable insight that arbitrage in the Pacific region is different from the Atlantic basin.
A third simulation study using a Monte Carlo method is the paper of Yepes Rodríquez (2008). He modeled the value of destination flexibility using a real options approach. Yepes Rodríquez’ destination flexibility value is what is named the arbitrage opportunity in this research: it is the profit that could be obtained when executing the arbitrage deal. He includes transportation costs but leaves out regasification reservation costs, resulting in his finding that the destination flexibility value exceeds 0.3 US$/MMBTU11. On average, price spreads between the UK and the US are 0.68 US$/MMBTU, so transportation costs only diminish the arbitrage potential by 50%12. According to Yepes Rodríquez (2008), even in the case of price convergence because of higher trade volumes between the US and UK markets, the option value of destination flexibility is likely to remain positive in the future. This is in line with Siliverstovs et al. (2005) and Hayes (2006), who both argue that arbitrage opportunities remain in the Atlantic basin. Neumann (2009) states that prices in the Atlantic basin in the LNG market actually converged during the period 1999 to 2009. The reason that arbitrage opportunities still remain is because LNG only accounts for a small fraction of the gas market, and therefore cannot solely connect the US and UK gas markets (Micola and Bunn, 2007).
Holleaux (2007) elaborates on the value of LNG arbitrage from both a historical and a forecasting perspective. Based on historical prices and market characteristics, arbitrage in the Atlantic could have been profitable in the period from 1999 to 2005. For the period after 2007, he claims that LNG arbitrage is a risky strategy that is very sensitive to market characteristics such as shipping availability. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
11 An MMBTU (Million British Termal Units) is a unit of energy, commonly used for gas.
12 This means that with a ship containing around 3x106 MMBTU (Holleaux, 2007), the extra profit can
For well-diversified players, however, Holleaux (2007) indicates that LNG arbitrage could be a good risk hedging strategy. Not only does he incorporate transport costs like Hayes (2006) does, he also argues that the booking of extra regasification facilities is a significant cost for the arbitrage opportunity. As discussed in the model specification in the next chapter, the reservation of regasification capacity on both sides of the Atlantic is accounted for as well in the arbitrage model. Holleaux (2007) assumes that transportation costs and regasification costs together constitute around 1 US$/MMBTU. Moreover, he argues that LNG suppliers do not want to divert their cargo to a different destination for less than 0.2 US$/MMBTU net margin due to the involved risk (Holleaux, 2007). This implies that the spread between the gas market prices, according to Holleaux (2007), should be at least 1.2 US$/MMBTU when arbitrage can be considered profitable.
The results of Yepes Rodríquez (2008) and Holleaux (2007) are in contrast with the findings of Dehnavi and Yegorov (2012). Dehnavi and Yegorov (2012) use a model that solely incorporates fixed shipping costs and argue that there were no favorable conditions for arbitrage in the Atlantic basin in the period 1996 to 2005. In the period 2006 to 2012, in contrast, the arbitrage opportunities are rather high, also for high transportation costs13, which is mainly the case because of the relatively low gas prices in the US. A possible reason why the gas price in the US is low is because the US subsidizes unconventional gas production, such as shale gas. A reason why the markets are not in equilibrium is probably because the US is not exporting domestically produced gas.14
To summarize, most of the literature finds that there are arbitrage opportunities for the period 1999 to 2005, with the exception of Dehnavi and Yegorov (2012). Using Monte Carlo simulations, Hayes (2006), Yepes Rodríquez (2008) and Holleaux (2007) find that arbitrage opportunities are likely to remain in the future, although LNG arbitrage is a rather risky activity.
In my opinion, three things can happen with the arbitrage opportunities in the Atlantic basin. First of all, the arbitrage opportunities could be constant over the investigated period, for example because gas markets are not cointegrated and gas prices do not react on LNG trade. The opportunities for LNG arbitrage can also increase. A reason for this could be that the costs of LNG transportation dropped significantly during the last decade (Maxwell and Zhu, 2011), so that, ceteris paribus, more room for arbitrage profits is left. Neumann (2009) argues that the gas prices in the Atlantic actually showed signs of price convergence during 1999 to 2009, so the gas price spread narrowed. This would indicate, ceteris paribus, that arbitrage !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
13 These high arbitrage opportunities are in contrast with the interrelation of Japan and the US. In the
Pacific region, only in the period 2009 to 2011 arbitrage opportunities existed on a small level, but for the rest of the period under study (2000 to 2012), the transportation costs would have offset the arbitrage profits.
14 The reasons why the US does not export domestically produced gas could be multiple. Because
opportunities diminished due to smaller price spreads. Hence, two conflicting effects can be revealed. Lower transaction costs make the costs of arbitrage lower, but narrowing price spreads are diminishing these arbitrage opportunities.
The ‘arbitrage value’ hypothesis of this research, in line with Neumann (2009) and Holleaux (2007) is that the potential profits of arbitrage decreased the US and the UK gas markets in the period from 2008 to 2012 in comparison with the period from 1997 to 2008, because of cointegration of gas prices across the Atlantic. This distinction in periods is chosen because a structural break between the HH and the NBP is found in November 2007.
2.5. Relationships with the literature
This research links with the literature in several ways. First, there is no paper in the literature that uses the period 2010 to 2012 for cointegration analysis of gas and oil markets in the UK. This investigation tries to fill this gap. In the US gas market, only Brigida (2012) investigates this period, but he uses weekly data, while daily data is used in this research. In both the US and UK markets, moreover, there are few authors that test for structural breaks in a cointegration framework. However, cointegration relationships change over time and the parameters of the time series are not necessary stable (Brooks, 2008). Authors investigating the US market who acknowledge that cointegration relationships are not rigid use the Chow breakpoint test (Villar and Joutz, 2006; Ramberg and Parsons, 2012). As I explain in the next chapter, the Chow breakpoint test is not the best method available to account for structural breaks when cointegration of oil and gas markets is investigated. Therefore, in line with Dahl et al. (2012), the Bai and Perron framework to test for structural breaks is used in this research.
2.6. Hypotheses and conceptual model
In this section, the hypotheses that are tested in this thesis are provided. The hypotheses already include structural breaks that are found using the Bai and Perron (2003) framework.
Diminishing oil link hypotheses
The diminishing oil link hypotheses are numbered D1 to D8 and can be seen below. The HH gas price is cointegrated with the WTI crude oil price in the period February 1997 to March 2006.
HD1
The HH gas price is cointegrated with the WTI crude oil price in the period April 2006 to December 2012.
HD2
The HH gas price is cointegrated with the Brent crude oil price in the period February 1997 to March 2006.
HD3
The HH gas price is cointegrated with the Brent crude oil price in April 2006 to December 2012.
HD4
The NBP gas price is cointegrated with the WTI crude oil price in the period February 1997 to December 2008.
HD5
The NBP gas price is cointegrated with the WTI crude oil price in the period January 2009 to December 2012.
HD6
The NBP gas price is cointegrated with the Brent crude oil price in the period February 1997 to December 2008.
HD7
The NBP gas price is cointegrated with the Brent crude oil price in the period January 2009 to December 2012.
HD8
Global gas market hypotheses
The global gas market hypotheses are numbered G1 to G2 and are provided below. The NBP gas price and HH gas price are cointegrated during the period February 1997 to November 2007
HG1
The NBP gas price and HH gas price are cointegrated during the period December 2007 to December 2012
Arbitrage value hypothesis
The value of arbitrage opportunities across the Atlantic for LNG suppliers with the UK as initial buyer of the cargo in the period 2008 to 2012 is significantly lower in
comparison with the period 1997 to 2007.
HA1
The relationships between the diminishing oil link hypotheses, the global gas market hypotheses and the arbitrage value hypothesis of this research can be seen in the conceptual model below.
Figure 2. The conceptual model with hypotheses incorporated. A – sign indicates a negative influence. A + sign indicates a positive influence.
If gas demand is rising and a growing energy source, the demand for LNG will also increase. If the market separation perspective is correct, natural gas is more and more priced according to gas fundamentals. This shift to gas fundamental pricing causes that the oil link of gas prices diminishes. When LNG is more and more used instead of pipeline gas, natural gas market shifts from a locally oriented market to a global market because LNG will connect the UK and the US market (although not totally (Micola and Bunn, 2007)). The global gas market results in a more global gas price. Due to the converging pricing differentials, the gas markets across the Atlantic shift to an equilibrium where the gas prices are more cointegrated. This causes that the arbitrage profits that can be obtained when exploiting arbitrage opportunities reduce.
Gas importance
LNG trade Diminishing oil link of gas prices
Global gas market
LNG arbitrage profits in the Atlantic
basin
+
3. Methodology
In this chapter, I present the methodology that is used in this research. In the first section, I explain the methodology that is used to test for cointegration. In this section, which unit root tests are used are explained, how to test for structural breaks and the Johansen cointegration methodology is provided. In the second section, a model is formulated to simulate the profits of LNG arbitrage. In the third section, the proposed parameter values are given. In the fourth section, I elaborate on the simplifications of the model.
3.1. Cointegration of (1) oil and gas, (2) gas across the Atlantic.
Nine cointegration tests are carried out. I first investigate whether the gas prices in the UK and the US are linked to the oil price. I do cointegration tests on the following pairs of time series: (1) HH&WTI, (2) HH&Brent, (3) NBP&WTI, (4) NBP&Brent, (5) LNGUS&WTI and (6) LNGUS&Brent. In addition, the transatlantic cointegration is considered by testing (7) NBP&HH and (8) NBP&LNGUS. Finally, I test whether the price at the HH is related to LNG imports in the US by investigating the relationship of (9) HH&LNGUS. As Yepes Rodríquez (2008) argues, there is no publicly available data on historical LNG prices in Europe, so it is not possible to test the influence of LNG prices on Brent crude oil prices or on the NBP. Siliverstovs et al. (2005), however, found that the LNG price is similar to the price of pipeline natural gas in Europe in the period 1993 to 2003. Accordingly, the price of LNG in Europe is the same as the NBP, also during the period 2004 to 201215. In line with the literature, prices are transformed to their natural logarithm.16 The reasons for this transformation are that skewness and kurtosis are reduced (Cuddinton and Wang, 2005).
To investigate the cointegration relationship between the above-specified variables, I proceed in three steps. First, two unit root tests to test for data stationarity are performed. Secondly, structural breaks are investigated. Thirdly, I test for cointegration using the Johansen (1995) framework. These three steps are considered in more detail below.
The first step of cointegration is to test for the stationarity of the data. Data is considered to be stationary when it has (1) a constant mean, (2) constant variance and (3) constant autocovariance (Brooks, 2008). When data contains a unit root, but is stationary in first differences, cointegration is an appropriate tool.
The literature mentioned in chapter two mainly uses the Augmented Dickey Fuller test17 to test for nonstationarity in the case of energy prices. However, the used !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
15 This assumption is a realistic assumption. I verified this at GasTerra, who use this assumption also.
Moreover, the LNG import prices in the UK and the NBP are almost identical, as seen by the UK customs import data that is publicly available.
16 Hendry and Juselius (2000) argue that cointegration in log levels also holds for cointegration in
levels. Therefore, transforming the price series to log levels does not have implications for the results.
17 Other unit root tests that are used by the literature can be found in Appendix A and are explained in
lag length estimations vary widely among the literature. Bencivenga and Sargenti (2006) base the lag length on the Schwarz Information Criterion (SIC), while both Vásquez Josse and Neumann (2006) and Villar and Joutz (2006) use the Akaiki Information Criterion (AIC). Serletis and Rangel-Ruiz (2004) argue that the AIC plus 2 is the best estimate for the lag, which is also used by Siliverstovs et al. (2005). Others do not mention their lag estimate at all (Brown and Yücel, 2008; Westgaard, 2011). Because of the difference used of lags I use lags based on both the AIC and the SIC. According to Culver and Papell (1997), ADF tests are sensitive to structural breaks. Because the data probably has structural breaks, I also use the Philips and Perron test (Philips and Perron, 1988) to test for unit roots, which gives robust estimates when the data has structural breaks (Saif Ghouri, 2006). I use the default specifications of Eviews for the PP unit root test. Hence the spectral estimation method is Bartlett kernel and the bandwith is the Newey-West bandwith. Even though the data probably does not have a drift, I perform the unit root tests for the general equation and also containing only an intercept. Using both approaches, I test for two types of non-stationarity: both the random walk model with a drift and a trend-stationary process (Brooks, 2008).
The second step of the cointegration analysis is to identify possible structural breaks. I explicitly test for structural breaks using the endogenous structural change model procedure of Bai and Perron (1998, 2003), as Dahl et al. (2011) propose. I specify this framework and the reasons for choosing this below.
Economic time series can have structural changes because market characteristics change, such as the legal framework or innovative production methods. Due to this changing market environment, the degree of cointegration can change as well. Structural breaks give a good indication where such a change occurs and thus the data can be adjusted accordingly to check whether the cointegration relationship has changed. Many methods have been produced to test for structural breaks. One of the best-known tests is the Chow structural break test (Chow, 1960), as used by Villar and Joutz (2006). The drawback of this test is that it can only test a single break. Moreover, the Chow test needs arbitrary input because the user a priori needs to define the break date, after which the Chow test tests whether it can reject the occurrence of a structural break. However, what we would like to identify are structural breaks without using any a priori, probably biased, information. For instance, it could be that the data has multiple breaks. According to Dahl et al. (2011), the framework of Bai and Perron (1998, 2003) is the only test that is able to “distinguish between periods with a cointegration relationship and periods when such a relationship is not present, and to allow for the possibility of more than one structural shift” (Dahl et al., 2011). Therefore, when investigating two time series without knowing beforehand where breaks (such as shocks) occur, the Bai and Perron framework is an appropriate method to test where and when structural breaks occur18.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
18 Although not available in EViews 7, the Bai-Perron framework is an add-in of EViews that links
The Bai and Perron framework consists of a linear regression model as can be seen in the following equation. The Bai and Perron approach is done in EViews by estimating the following equation:
(1) With t = Tj-1+1,…,Tj. In this model, yt is the dependent variable at time t (for instance, NBP gas price). xt and zt (qx1) are vectors of covariates, β and δj are the corresponding vectors of coefficients and µt is the disturbance term at time t (Bai and Perron, 1993). In this framework, the breakpoints are modeled as if they are unknown. The purpose of this framework is to estimate the unknown regression coefficients together with the break points when T observations are available. The only input this test needs is the trimming parameter. This parameter determines what part of the sample needs to be included in a period between structural breaks. For instance, with a trimming parameter of 0.1, the sample between two structural breaks should be at least 10% of the data. To test for robustness of the structural breaks, the structural break tests are executed four times, using trimming parameters of 0.05, 0.1, 0.15 and 0.2.
The next step is to split the data according to the estimated break dates. Only break dates that occur after 2004 are considered. This is done because for the period 1997 to 2004 the cointegration relationships are already investigated by the literature. Moreover, this research focuses especially on the changes in cointegration relationships that occur after 2004 because of the changing market characteristics. Hence, a simplifying assumption is that there are no structural breaks in the period 1997 to 2004 for all relationships. The month including the break date and the months before and after the break date are excluded from the data. The motivation for this is that structural breaks do not occur ‘over night’ but are the outcome of changes that occur during a period around the break date. Excluding this period of the data set makes sure that the structural change is in neither sub sample. By excluding the structural break period, a clear distinction is made between a period prior and a period after the structural break. In this way, it is possible to investigate whether the cointegration relationship has changed substantially.
Now that the stationarity of the data is investigated and the structural breaks are identified, the cointegration tests can be performed. There are many ways to test for cointegration, although two approaches dominate in the literature about gas and oil prices. The first one is the approach of Engle and Granger (1987), the other is the approach suggested by Johansen (1988, 1995). I use the Johansen Full Information Maximum Likelihood cointegration framework (Johansen, 1995). Using the Johansen framework, the cointegration vectors and the speed of adjustments of the cointegration relationships can be identified.
The Johansen framework is formed around a p-dimensional Vector Error Correction Model (VECM), as can be seen in the following equation:
yt = xt
'
β + zt
'
(2)
where Δ is the first difference operator, yt is the set of I(1) variables that are determined with unit root tests as discussed above, εt~n.i.i.d. (0,Σ), µ is a drift parameter, and Π is a (pxp) matrix of the form Π=αβ’, where α and β are (pxr) matrices of full rank, with β containing the r cointegrating vectors and α carrying the corresponding loadings in each of the r vectors. β is the cointegration vector and α is the speed of adjustment. The higher α, the faster price differences erode. The higher β, the more cointegrated the variables are.
3.2. Proposed model of transatlantic arbitrage
The model of transatlantic arbitrage that is given below differs from the literature in several ways. First of all, no Monte Carlo simulation is used as Yepes Rodríquez (2008), Suenaga (2007) and Holleaux (2007) do. The focus of this research is different, because it explores the value of arbitrage that occurred during the period 1997 to 2012. For this period price spreads are already given and, thus, there is no need for Monte Carlo simulation of future prices using parameters that are generated with historical data. The model I propose differs also from Dehnavi and Yegorov (2012), because I incorporate the regasification reservation costs.
It is not useful to use daily price spreads for the arbitrage value model. This is the case because it takes roughly a month to transport gas from Europe to the US, so it is not possible to exploit daily price spreads. Therefore, as an indication for price differentials, monthly spreads are used. In the month n, the average price of LNG in Europe and in US markets is denoted by PUK(n) and PUS(n). I assume that a shipment of LNG to the UK is the base case, for instance with a long-term contract between an LNG supplier and a buyer in the UK. The value of destination flexibility can be described by the following equation:
(3) Where C is the additional cost that needs to be taken into account when executing the arbitrage opportunity. These costs contain two major components, as is shown in the following equation:
C = ΔT (n) + Creservation (4)
Where ΔT(n) is the extra marine transportation costs that need to be paid (or obtained) by shipping the LNG to the alternative (US) market. Creservation is the additional cost of reserving a regasification facility. In line with Hayes (2006), the availability of regasification capacity is assumed to be costly. When prices are peaking and more people shift their gas supply to the high price market, regasification capacity is scarce
Δyt= ΓiΔyt−i+ ∏ yt−1+µ +εt, t = 1,...,T i=1
k−1
∑
(Hayes, 2006). In order to capture the full benefits of LNG arbitrage, regasification capacity should, therefore, be reserved. When a LNG supplier wants to operate with destination flexibility, regasification capacity has to be booked both in the US and the UK. This implies that the costs of booking regasification capacity are higher than in the scenario without destination flexibility.
The monthly prices of the UK and the US market are calculated by the arithmetic mean of the daily prices, as described in equation 5 and 6.
(5)
(6)
The value of free destination for a supplier is the discounted sum of v(n) over a period of time that the supplier produces and transport LNG ships. This can be described by the following equation.
V = v(n)⋅δn⋅ N = n=1 T
∑
max(PUS(n) − PUK(n) − C, 0)⋅δ n ⋅ N n=1 T∑
(7)Where N is the amount of LNG trades that are done for LNG arbitrage in month n and where δ is a discount factor that takes the time value of money into account.
3.3. Parameters of the arbitrage model
As mentioned in the section before, the model uses several parameters. In this section, I elaborate on these parameters and give the transportation costs, regasification costs and provide the discount factor of the arbitrage value model.
Transportation costs depend on the construction costs of the LNG vessels, the distance between the markets and the time that is needed to transport the LNG. According to Maxwell and Zhu (2011), construction costs for LNG vessels have decreased significantly over the last years because larger ships have been built that could capture some economies of scale. To my knowledge, unfortunately, there are no time series available for transportation costs in the LNG industry. Therefore, I calculate transportation costs, assuming a fixed size of LNG tankers. This assumption implies that the transportation costs are constant over the investigated period of 1997 to 2012. However, Maxwell and Zhu (2011) also argue that differences in prices occur mainly because of differences in shipping distances. The transportation costs in the arbitrage model are therefore the difference between the transportation costs to the initial buyer in Milford Haven in the UK and the transportation costs to Lake Charles in the US. These transportation costs are dependent on the location of the LNG
supplier. For LNG suppliers that need to cross extra distances in order to exploit the arbitrage option, such as Qatar, the transportation costs are higher because it takes longer to deliver the LNG. An LNG supplier in Trinidad and Tobago, in contrast, has lower transportation costs when he exploits the arbitrage opportunity because the US is closer than the UK for this supplier.
Three LNG suppliers are considered in this research. The first scenario is a LNG shipment from Qatar to the UK or the US. The second one is a shipment from Nigeria to either the US or the UK. The third scenario consists of a LNG supplier in Trindidad and Tobago. The liquefaction ports are located in Las Raffan, Bonny Island, and Port of Spain respectively. The regasifaction terminals that are used for this research are located in Lake Charles in the US and Milford Haven in the UK.
Table I. Transportation costs for three LNG suppliers.
Supplier-Buyer Distance (km)* Time (days)** Costs ($) *** Costs/MMBTU
($) **** Qatar – UK 20100 26 1,690,000 0.5026 Qatar – US 30400 38 2,470,000 0.7345 Nigeria – UK 13080 18 1,170,000 0.3479 Nigeria - US 16900 22 1,430,000 0.4252 T&T – UK 12500 17 1,105,000 0.3286 T&T - US 6740 10 650,000 0.1933
* Distances are obtained from Dehnavi and Yegorov (2012) and are the distances from the liquefaction plant to the regasification terminal and back.
** This time is the time it takes a ship to transport the LNG from the liquefaction plant to the regasifaction plant and back. Includes a day for loading and a day for unloading the LNG. Average speed of LNG tankers is assumed 19 knots.
*** It is assumed that short-term LNG tankers are available at a price of $65000 per day (Dehnavi and Yegorov, 2012).
**** It is assumed that a LNG tanker of 140000 m3 contains 3362800 MMBTU, with a conversion of
24.02 MMBTU/m3 LNG (IGU, 2012).
As can be seen from the table above, the transportation costs depend on the distance between the liquefaction and regasification location. When the liquefaction location is further away from the regasification location, the transportation costs are higher. This implies that for longer distances, arbitrage is more costly and, ceteris paribus, the profits of arbitrage opportunities are lower. The parameters that are used for the calculation of the arbitrage profits are:
ΔTQatar = 0.7345-0.5026 = 0.2319 $/MMBTU ΔTNigeria = 0.4252-0.3479 = 0.0773 $/MMBTU ΔTT&T = 0.1933-0.3286 = - 0.1353 $/MMBTU
In the model, I assume that on average, one ship a month can be used for LNG arbitrage19. Hence, N = 1 in equation 7.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
19 It is difficult to assess how much LNG trade can be classified as arbitrage because this information is
As explained before, the booking of a regasification facility is costly. Hayes (2006) states that the booking of a complete regasification facility comes at a cost of 75 million dollar per year or 6.25 million dollar per month. Because I assume that there is one ship available per month for arbitrage opportunities, this would amount to regasification costs of 1.85 US$/MMBTU. There is no need to reserve regasification capacity for every day of the month, however, so these costs are probably too high. According to Holleaux (2007), booking a regasification facility for arbitrage purposes cost about 0.9 US$/MMBTU. In his actually used model, however, he uses a price of 0.5 US$/MMBTU. Accordingly, this assumption of Holleaux (2007) that the extra regasification capacity that needs to be planned for arbitrage exploitation comes at a cost of 0.5 US$/MMBTU.
The discount factor δ takes the time value of money into account. Yepes Rodríquez (2008) chooses 0.95 as a reasonable estimate for the discount factor per year. A money amount N that is obtained at time 0 is worth δtN at time t, hence 0.95N at time 1. I use the same discount factor as Yepes Rodríquez (2008) because a risk-free discount rate of 0.95 is a reasonable trade off between weighting the first years of the data and the last years20. However, because my arbitrage value is calculated per month, the used discount factor per month is 0.951/12 = 0.995735.
To summarize, the table below contains the parameters that are used for the model of LNG arbitrage. Sensitivity analyses of the transport costs, the regasification fee, the discount factor and the number of cargoes are given in Appendix H. These sensitivity checks show that the regasification reservation cost and transportation costs have a linear relationship with the amount of arbitrage opportunities and the profits that can be obtained when exploiting these arbitrage opportunities. The discount factor also shows a linear relationship, but the arbitrage opportunities and profits of LNG arbitrage are less sensitive to the discount factor than to the transportation costs and regasification fees. The number of cargoes is insensitive to arbitrage opportunities of LNG arbitrage.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
20 A lower discount rate (<0.95) is more sensitive to the results for the market prices during the first
Table II. Parameter specification of the LNG arbitrage model Parameter Value ΔTQatar 0.2319 $/MMBTU ΔTNigeria 0.0773 $/MMBTU ΔTT&Ta - 0.1385 $/MMBTU Cgasification 0.5 $/MMBTU N 1 δ 0.995735
ΔTQatar are the extra transportation costs for a LNG supplier in Qatar to deliver the cargo to the US
instead to the UK.
ΔTnigeria are the extra transportation costs for a LNG supplier in Nigeria to deliver the cargo to the US
instead to the UK.
ΔTT&T are the extra transportation costs for a LNG supplier in Trinidad and Tobago to deliver the cargo
to the US instead to the UK.
N is the number of cargoes that is available for arbitrage purposes.
δ is the monthly discount factor to take the time value of money into account.
3.4. Simplifications of the arbitrage model
The model of LNG arbitrage contains many simplifications. I discuss some of them. Many simplifications can be related to the framework of Zhuravleva (2009), which is discussed in the literature section above.
LNG in the vessels evaporates due to heat leakage. When ships travel further in the case of arbitrage, these evaporation losses actually need to be taken into account. However, this is rather difficult because the regasified LNG of the vessels is used as fuel for the propulsion of the LNG vessels themselves, so it saves on other fuels (Dimopoulos and Frangopoulos, 2008). Moreover, boil off fractions of LNG are considered only 2-4% of total cargo per trip (Faruque Hasan et al., 2009). Because of the small influence on the potential profits of the LNG arbitrage, I assume that extra evaporation losses due to further distance transportations do not come with extra costs.
I leave out company specific circumstances, the third prerequisite of the framework of Zhuravleva (2009), as can be seen in figure 1. One of the things that need to be taken into account is whether ships can deliver the full load to the regasification facility (Dorigoni et al., 2008). At the moment, there are some LNG ships that are too big for certain docks, such as the Q-max ships of Qatar. Ships with a size of 150,000 m3 are used in the parameter estimations, which is not a problem for many shipping docks (Dorigoni et al., 2008).
I assume that the gas in the UK and the US is a homogeneous good and that pipeline gas and LNG are perfect substitutes. Quality interchangeability of natural gas and LNG is considered unproblematic in the UK and the US21. When a ship is sent to the arbitrage location (the US), it is assumed that the initial buyer gets gas from another source at the price of the NBP. The replacement energy, probably pipeline gas, comes not at significant extra costs for the LNG supplier. In reality, however, the !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
21!Although!there!are!differences!in!quality!characteristics!between!LNG!and!pipeline!gas,!such!as!
gas price could be higher because of storage availability, seasonality or other considerations. As Suenaga (2007) correctly argues, this assumption is not valid in Japan/South Korea because pipeline gas is almost not available there. In the Atlantic basin, however, the risk of non-available pipeline gas is considered negligible.
Another feature of LNG is that it is troublesome to deliver less than full cargoes: LNG shipping is a discrete process (Dehnavi and Yegorov, 2012). Unlike financial arbitrage, when it is possible to divert a part of the cargo, it is considered to be impossible to deliver, for instance, half a cargo to the US and the rest to the UK. If half a cargo could be transported to another terminal, regasification reservation probably offsets the arbitrage opportunities. Accordingly, this simplification does not have significant economic consequences for this investigation.
In line with Hayes (2006), it is assumed that the profits that are obtained with the LNG arbitrage are the total profits in the value chain of LNG trade, calculated as obtained solely by the LNG supplier. As Hayes (2006) argues, when alternative (contractual) structures are used, the profits and risks of arbitrage are divided among the participants of the arbitrage activity by a certain mechanism (Dehnavi and Yegorov, 2012; Zhuravleva, 2009). For an overview of these mechanisms appendix D can be consulted. This simplification does not influence the profits that can be obtained with LNG arbitrage; it only argues that in practice, the profits are probably shared among the participants in the LNG trade.
4. Data
In this chapter the data of the thesis is presented. The data and the source of the data that is used for this research is given in the first section. In the second section, characteristics of the data are given.
4.1. Sources
For this thesis, several data sources are used. I discuss them briefly.
Daily spot prices of the HH are extracted from the EIA database. The EIA is widely used in the literature as the provider of gas proxies in the US (Villar and Joutz, 2006, Siliverstovs et al., 2005, Hartley et al., 2008). For an overview of all the data sources of the literature, see appendix B. Daily spot prices of the NBP are obtained from Shell. Daily spot prices of NBP is used, among others, by Vásquez Josse and Neumann (2006), Neumann et al. (2007), Dahl et al. (2012) as a proxy for UK gas prices. NBP prices are originally given in British pounds, but are converted to US$/MMBTU by using monthly historical currency exchanges using the International Financial Statistics. Unavailable data on certain dates in a time series (f.i. Christmas days and national holidays in one of the countries) is excluded from the dataset. Both daily Brent and WTI crude oil spot prices are obtained with Datastream from the EIA. Import and export gas prices differ for natural gas by pipeline and LNG in the US and are extracted with Datastream from the EIA. Siliverstovs et al. (2005) found that LNG and pipeline prices do not differ significantly in Europe for the period up to 2004. For this reason, the price of LNG in Europe is set equal to the NBP. The timespan of all time series is 3 February 1997 to 31 December 2012.
4.2. Data descriptives
In this section, the descriptives of the data are presented.
Table III. Data descriptives of daily gas and oil proxies.
Variable* Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis
Log WTI 2.072 2.131 2.130 0.684 0.635 -0.190 1.857 ΔWTI 0.000 0.000 0.191 -0.171 0.025 -0.091 8.179 Log Brent 2.057 2.083 3.283 0.559 0.692 -0.122 1.862 ΔBrent 0.000 0.001 0.135 -0.136 0.023 -0.046 5.716 Log NBP 1.475 1.461 3.005 0.345 0.661 -0.062 1.822 ΔNBP 0.000 -0.001 0.478 -0.281 0.039 2.549 27.536 Log HH 1.443 1.456 2.734 0.489 0.491 0.113 2.236 ΔHH 0.000 0.000 0.284 -0.289 0.042 0.332 9.323 Log LNG 1.500 1.526 2.529 0.688 0.419 -0.014 2.058 ΔLNG 0.000 0.000 0.506 -0.570 0.031 -1.934 141.711
*Period for all data is 3 February 1997 – 31 December 2012.
All variables that contain Log in their name are the natural logarithms of the variables. All variables with Δ in their name are the first differences of the variables.
Number of observations for log variables is 4021. Number of observations for first differences is 4020. WTI: West Texas Intermediate crude oil, Brent: Brent crude oil, NBP: National Balancing Point natural gas price, HH: Henry Hub natural gas price, LNG: import price of LNG in the US. All data are expressed in $/MMBTU.
During the period from 1997 to 2012, the prices of Brent crude oil and WTI crude oil are roughly the same. The same holds for the NBP and HH gas prices. As indicated by Cuddington and Wang (2005), using log levels diminishes the skewness and kurtosis effects. Hence, for the remaining part of the research, in line with the literature, the log levels of the time series are used. A serious point of interest is that all time series of the data set are not distributed normally. The descriptives of the HH indicate that the HH shows excess kurtosis. The prices of the LNG import in the US also have some excess kurtosis. These signs of non-normality could bias the outcome of the research. Non-normality is verified by empirical distribution tests that are done on the time series, which can be found in Appendix C. Silvapulle and Podivinsky (2000) find that Johansen cointegration tests are robust for conditional heteroskedasticity. Hendry and Juselius (2001) argue that statistical inference is sensitive to parameter constancy, serially-correlated residuals and residual skewness. However, they acknowledge that the VAR estimations are rather robust for excess kurtosis and residual heteroskedasticity. As can be seen in table III, WTI, HH and Brent do not show much excess skewness. As displayed in Appendix C however, all the linear regressions that are used in this research show signs of autocorrelation in the residuals and heteroskedasticity when tested for log levels.
