Volatility and Risk Sharing in European gas markets

Frank Asche†, Petter Osmundsen†, Marius Sikveland ††, and Ragnar Tveteras†

Address for correspondence: Professor Frank Asche University of Stavanger N-4036 Stavanger, Norway Email: [email protected]

†

Professor, University of Stavanger / Department of Industrial Economics.

††

Research fellow, University of Stavanger / Department of Industrial Economics.

Volatility and Risk Sharing in European gas markets

Abstract We investigate volatility of natural gas spot prices in UK, the Netherlands, and Belgium compared to long-term supply contract prices. Furthermore we show how the gas spot markets are correlated with the oil price and the long-term supply gas contracts. We find that when buying gas in the spot market, the buyer is exposed to considerably more price risk compared to buying by means of long-term supply contracts. We also provide empirical evidence that there has been a price shift in the relationship between the spot gas market and the contract gas market after 2003. After this point of time, shocks to the oil price create volatility in spot gas prices. Increased liquidity and maturity of the spot market coupled with higher capacity utilization in the gas infrastructure might have made the spot gas price more sensitive to shocks in the market for substitutes. An implication is that the long-term supply contracts have become even more important to gas buyers as a means for risk reduction.

Keywords: price volatility, gas spot markets, long term gas supply contracts

Introduction During the last decade, we have observed several changes in the ways natural gas is traded in Europe. Although long term contracts still dominate, we have in recent years observed the emergence and growth of several types of spot markets. Financial derivatives trading have emerged in several gas markets, both in continental Europe and the UK. The UK gas market was deregulated in the early 1990s, after British Gas was privatized in 1986. After the deregulation, gas in the UK is primarily traded in the Over The Counter Market. The National Balancing Point (NBP) gives the reference price for many forward transactions and for the International Petroleum Exchange futures contracts. The Interconnector - a pipeline connecting the UK market to Continental Europe - became operational in 1998, and brought

1

to the UK the European price of gas which is closely linked to oil products.

Recently, spot trading has also increased in continental Europe. Zeebrugge (ZEE) in Belgium is the arrival point for the Interconnector. The Title Transfer Facility (TTF) hub in the Netherlands is one of the most recently developed gas hubs in Europe. This hub, operational since 2003, is rapidly maturing into an important trading point for gas market participants in Northwest Europe.1 Due to these major market developments, price relationships between energy products might have changed, and the individual volatility processes might also have been altered.

If buying gas in the spot market is risky - in terms of price volatility and volume availability a risk averse buyer would be willing to pay a premium for risk reduction. At the same time, a risk averse seller would give a discount to avoid risk. Because of this, most of the European gas is sold by means of long-term supply contracts. In regulating the flexibility of contracting volumes and prices, the exporting and the importing companies have conflicting interests. Since gas storage is expensive and in limited supply, the importer would like to have flexibility with respect to volumes, thus being able to adjust to changes in downstream demand. Demand fluctuates, especially over the seasons, with a higher demand in winter than in summer. The exporters, on the other hand, have to sink large irreversible and often customer-specific investments in extraction, processing, and transportation facilities. Before doing so, they would like to have assurance against hold-up situations, i.e., assurance that they will be able to sell the gas for a reasonable price over a considerable period of time, thus securing a competitive return on their investments. Also, to exploit the extraction, processing and transportation capacity, the seller would prefer to deliver a stable gas stream at maximum capacity utilisation. As for price flexibility, the exporter would – before making large irreversible investments – prefer a specific price, a minimum price, or other types of price guarantees for the entire period of delivery. The buyers, on the other hand, would like the gas price to be responsive to the price of substitutes (such as oil products), so that they are able to resell the gas.

1

One reason for this is its relationship to the large Groeningen gas field, which plays a role as a swing supplier of natural gas in northern Europe. Because of its close proximity to many markets, it made large investments in pipelines unnecessary (Asche, Osmundsen and Tveteras, 2002).

2

The current prices on gas delivered according to the long term take-or-pay contracts is determined by a price formula. The formula links the current gas price to the price of relevant energy substitutes, thus continuously securing the buyer competitive terms. The price formula consists of two parts, a constant basis price (fixed term) and an escalation supplement linking the gas price to alternative forms of energy (variable term). Examples of alternative energy commodities used in pricing formulas for natural gas are light fuel oil, coal, and electricity. Usually a combination of alternatives is used for escalation purposes (weighted average). The basis price (which is not subject to subsequent price revision) reflects the parties’ evaluation of the value of the gas at the time of entering the contract. Each of the alternative energy commodities is assigned a certain weight in the escalation element, reflecting the competitive situation between natural gas and the substitute. The price change of each energy commodity is multiplied by an energy conversion factor, to make the substitute and natural gas commensurable. Thereafter, the individual escalation terms are multiplied by impact factors, i.e. the change in the price of the substitute is not fully reflected in the gas price. A typical price formula is given by P = P0 + ∑ α j ( AE j − AE j 0 ) EK AEj λ j ,

(1)

j

where P is the gas price, P0 is the basis price, α j is the weight in the escalation element for substitute j (with

∑α

j

= 1 ), ( AEj − AE j 0 ) is the price change for substitute j (actual minus

j

historic price), EKAEj is an energy conversion factor, and λj is the impact factor for price changes in substitute j.

Long-term contracts may also make competition in a market softer, yielding higher prices. Because of this, the US gas market was deregulated in the mid 1980s and long term contracts were dissolved (De Vany and Walls, 1993). In the EU, there has been a drive toward deregulation since the 1990s, but so far with somewhat limited effect. In continental Europe, long term gas contracts with prices linked to oil and other oil products still dominate. The recent experience of the Russians having gas sales as an integrated part of their foreign policy, has reduced the reluctance towards long-term gas sales contracts in alternative supplier countries, as such contracts are usually a necessary

3

condition for the development of new gas fields and transportation infrastructure. The oil companies need some insurance that they will recoup their large and irreversible investments.

There have been several studies on the link between different gas prices, and between gas and oil prices in Europe. Four studies are of particular interest in our setting. Asche, Osmundsen, and Tveteras (2002) investigate whether the German gas market is integrated. They examined time series of Norwegian, Dutch, and Russian gas export prices to Germany in 1990–1998, and found that the various border prices for gas to Germany move proportionally over time, indicating an integrated gas market. Silverstovs et al. (2005) studied monthly prices of pipeline gas prices in Europe and LNG import prices to Europe in the period 1993–2004. Their study also showed a high degree of market integration within Europe. Asche, Osmundsen, and Sandsmark (2006) studied the UK gas market in the period after deregulation of the UK gas market (1995) and before the opening of the Interconnector (1998). In this period the UK gas market had neither government price regulation nor a physical continental gas linkage. They used this period to explore if there was one integrated market for energy, i.e., if the relative price of natural gas to other energy commodities -such as oil and electricity - was stable. Their study indicated a highly integrated market where wholesale demand seemed to be for energy rather than a specific energy source. Panagiotidis and Rutledge (2007) examined the relationship between UK wholesale gas prices and the Brent oil price over the period 1996–2003 to investigate whether oil and gas prices decoupled after the deregulation of the UK gas market, and whether the link between the UK and the continent was caused by the Interconnector. They found that the cointegrating relationship was present throughout the sample period.

There are also several studies investigating the volatility of gas and gas futures prices. Mu (2007) examines how weather shocks impact prices in the US natural gas futures market. The empirical results show a significant weather effect on both the conditional mean and the conditional volatility of natural gas futures returns. The study finds that volatility is considerably higher on Mondays and the day when the natural gas storage report is released, i.e., information about market fundamentals is an important determinant of natural gas volatility. Pindyck (2004a,b) tests whether there is a significant trend in volatility and whether the demise of Enron increased volatility in natural gas and oil markets. A statistically significant and positive time trend is found for natural gas, but it is too small to have any 4

economic importance. The Enron event appeared to have no significant impact on natural gas volatility. Ewing et al. (2002) examine volatility transmission between two stock price indexes consisting of major companies in the oil and gas sector. Evidence is found of volatility persistence in both indexes and significant volatility transmission from the natural gas sector to oil sector but not vice versa.

To compare the natural gas spot markets in the UK, the Netherlands, and Belgium with longterm gas supply contracts - when it comes to price volatility and price structure - we first calculate the correlation coefficient between pairs of prices, and the standard deviation of individual price series. In recent years spot gas prices have increased considerably relative to long term gas contract prices, and we have seen increasing degrees of peak load pricing. To better understand the price risk in natural gas prices before and after the point of a potential structural shift in the relationship between spot and long term gas contract prices, we estimate a multivariate GARCH model with a BEKK parameterization (Engle and Kroner 1995) to capture the dynamics of the volatility process before and after this point of time. We do this for the two longest price series available at a daily frequency; the NBP gas spot price in the UK and the Brent blend crude oil price. This approach allows us to capture how shocks to either price series transmit to the other price series.

The paper is organized as follows: In the next section we provide a descriptive analysis of gas and oil prices, focusing on shifts in the mean and standard deviation of prices over time, and shift in price relationships. In Section 3 the econometric methodology is presented. We report the empirical results in Section 4, and Section 5 concludes.

2. Descriptive analysis Figure 1 shows the development of six different oil and gas prices; crude oil (Brent Blend), natural gas in the UK (NBP), in Belgium at Zeebrugge (ZEE), in the US at Henry Hub (HH), in the Netherlands (TTF), and the continental European oil linked contract gas price. We see that there is an upward trend in prices from 1996, and some prices seem to exhibit higher volatility in the later years.

5

Figure 1: Energy prices in USD/MMBtu

16 14 12 10 8 6 4 2 0 1990

1992 Brent

1994 NBP

1996 ZEE

1998 HH

TTF*

2000

2002

2004

2006

Continental gas price

From table 1 and figure 2 we can see that NBP has the highest level of volatility, whereas Zeebrugge (ZEE) has the highest level of volatility in continental Europe. The more liquid gas spot markets in Europe have higher volatility, the more mature deregulated market in the US has substantially lower volatility than the spot markets in Europe, but still much higher volatility than the oil price and the continental oil-linked contract gas price. We can divide Figure 1 into three distinct periods. The first period starts in October 96 and ends September 99. In this period we have relatively low gas prices, and low volatility. From 1999-2003 we see that volatility and price increase (with the exception of NBP, where the price is slightly lower). From 2003-2006 prices increase again, but volatility in Brent, oil-linked long term supply contract and Henry Hub (HH) decrease. Volatility in European spot gas markets increases in this period. On average, over the whole period, the continental gas price sells at a premium, and has significantly lower volatility.2 The price premium in the long term take-or-pay contracts can be explained by the value of flexibility inherent in swing facilities, which provide the buyer 2

As such it appears that the buyers pay a premium, and the sellers do not have to give a discount. Asche, Osmundsen and Tveteras (2002) compare import contract prices to Germany from the Netherlands, Norway and Russia and find that Russian gas sells at a discount. One explanation is that Russia, with the longest transport distance, opts for high capacity utilization and low levels (if any) of swing in the contracts.

6

with flexibility as to the volume and the timing of the gas take This is valuable to the buyer. In addition to the value of risk reduction with respect to lower price volatility, the contract contains a real option that allow them an arbitrage between spot and contract prices. The finding that contract prices have lower volatility is as expected, since they are based on monthly average oil product prices, including reference periods of several months and fairly high pass-through factors. All these elements work towards a non-volatile contract price. This is as intended; low price volatility was introduced for buyers that could not pass on all purchase price volatility to its end users.

Figure 2: Mean price and standard deviation

Mean price, USD/MMBtu

Mean-standard deviation 10.00

Brent

8.00 Oil-linked 6.00

HH

TTF

ZEE

NBP

4.00 2.00 0.00 0%

20 %

40 %

60 %

Standard deviation

80 %

100 %

Dated brent 96-99 NBP 96-99 HH 96-99 Oil linked 96-99 Dated brent 99-03 NBP 99-03 ZEE 99-03 HH 93-03 Oil linked 99-03 Dated brent 03-06 NBP 03-06 ZEE 03-06 HH 03-06 TTF 03-06 Oil linked 03-06

The relatively low volatility confirms the take-or-pay contract as a means for risk reduction. From the correlation matrix (Table 2) we see that the more liquid the market, the more independently it is priced from the crude oil price. Illiquid markets use to a larger extent the crude oil price as a price reference. What happens is that when a gas market is illiquid, gas is priced relative to other types of energy where market prices are available. The oil product indexed long term gas sales agreements set the price in line with the oil product market fluctuations. These fluctuations do not necessarily have anything to do with gas demand fluctuations (e.g., due to changes in temperature) that set spot prices. The correlation between NBP and Zeebrugge is 0.99. This is not surprising given that the two markets are connected

7

via the Interconnector. The co-existence of spot markets and oil product indexed contracts has led to an interesting phenomenon in the Continent: buyers have the opportunity to optimise between off-take under the contract and market purchase. This may explain why oil product prices - via contract prices - tend to have an impact on spot market prices, at least in the Continent. As gas buyers have considerable flexibility (swing) in the timing of their gas off-take, the gas contracts often represent the marginal source of gas supply, and consequently affect the overall price setting in the gas market.

Table 1: Mean, standard deviation, and coefficient of variation 1996-1999 Mean Std.dev Coefficient of variation

Brent 96-99 3.09 33 % 11 %

NBP 96-99 1.78 68 % 38 %

ZEE n/a n/a n/a

HH 96-99 2.25 52 % 23 %

TTF n/a n/a n/a

Continental gas price 96-99 2.33 11 % 5%

Brent 99-03 4.93 34 % 7%

NBP 99-03 2.62 54 % 21 %

ZEE 99-03 2.78 62 % 22 %

HH 93-03 3.95 60 % 15 %

TTF n/a n/a n/a

Continental gas price 99-03 3.34 17 % 5%

Brent 03-06 9.37 28 % 3%

NBP 03-06 6.18 88 % 14 %

ZEE 03-06 6.18 77 % 12 %

HH 03-06 6.66 59 % 9%

Brent 5.74 32 % 6%

NBP 3.46 69 % 20 %

ZEE n/a n/a n/a

HH 4.27 57 % 13 %

1999-2003 Mean Std.dev Coefficient of variation 2003-2006 Mean Std.dev Coefficient of variation

TTF 03-06 Continental gas price 03-06 5.66 5.71 42 % 12 % 7% 2%

1996-2006 Mean Std.dev Coefficient of variation

TTF n/a n/a n/a

Continental gas price 3.77 14 % 4%

We want to examine more closely the relationship between the continental gas price and the NBP gas price. In figure 3 we plot the percentage deviation of the NBP price from the continental gas price. From the figure we can clearly see how seasonal variations at NBP result in spot gas becoming more expensive than contract gas in winter. We can also observe something that looks like a change in the volatility around 2003.

8

Table 2: Correlation matrix Brent

NBP

ZEE

HH

TTF* Continental gas price

Brent

1

NBP

0.71

1

ZEE*

0.73

0.99

1

HH

0.69

0.72

0.71

1

TTF*

0.73

0.82

0.84

0.37

1

Continental gas price

0.93

0.74

0.76

0.67

0.80

1

Figure 3: Continental gas price and UK spot gas price spreads in percentages

80 % 60 % 40 % 20 % 0% 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 -20 % -40 % -60 % -80 % -100 % -120 % -140 % (Continental gas price-NBP)/Continental gas price

9

Average deviation

Historical cumulative distribution

Figure 4: Historical cumulative distribution of prices

1 0.8 0.6 0.4 0.2 0 1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

USD/MMBtu Brent

NBP

HH

Continental gas price

Based on the observed prices in the data period we plot the cumulative distribution for four energy prices in figure 4.3 If we interpret figure 4 as a probability distribution, we can see that NBP has highest probability of higher and lower gas prices. Although NBP has been cheaper than continental supply contract on average in the period investigated, NBP has a higher probability of high gas prices because of extreme seasonal variations and volatility.

3. Econometric approach We want to empirically investigate whether there has been a structural shift in the mean and volatility of spot and contract gas prices. A natural way to do this is through a GARCH model. The concept of conditional heteroskedasticity was introduced by Engle (1982), who proposed a model in which the conditional variance of a time series is a function of past shocks; the autoregressive conditional heteroskedastic (ARCH) model. Bollerslev (1986) extended the ARCH model and allowed for a more flexible lag structure. He introduced a conditional heteroskedasticity model that includes lags of the conditional variance (ht-1, ht-2,…,

3

Historical prices are sorted and assigned equal weight

10

ht-p) as regressors in the model for the conditional variance in addition to lags of the squared

error term (u t2−1 , u t2−2 ,..., u t2−q ) - the Generalized ARCH (GARCH) model. In a GARCH(p,q) model, ut is defined as: yt = µ + u t ,

(2)

where yt is return and µ the drift in returns, specified as

q

p

i =1

j =1

u t = ε t (α 0 + ∑ α i u t2−i + ∑ β j ht − j )1 / 2 ,

(3)

where εt ~NID(0, 1); p ≥ 0, q ≥ 0; a 0 > 0, ai ≥ 0, i = 1,..., q and β ≥ 0 , j = 1, 2,…, p. It follows from manipulation of the above equation that ht (the conditional variance of ut) is a function of lagged values of u t2 and ht:

q

p

i =1

j =1

ht = α 0 + ∑ α i u t2−i + ∑ β j ht − j

(4)

In the current setting, where we investigate several energy prices simultaneously, it could be relevant to model conditional variance in a system. If the volatility of one price series affects another, assuming exogeneity of one of the prices would not be correct. Various parameterizations of multivariate GARCH models have been proposed in the literature. The most popular include the VECH model introduced by Bollerslev el al. (1988), the BEKK model (Engle and Kroner, 1995) and the Dynamic Conditional Correlation (DCC) model (Engle 2002). We adopt the model introduced by Engle and Kroner (1995). The BEKK functional form ensures that the conditional covariance matrix is positive definite so that conditional variances are always non-negative, and also give more detailed output that allows us to show how shocks to either series, and level of conditional variance, will affect the other. The DCC model does not allow such a detailed investigation.

The BEKK parameterization for the bivariate GARCH(1,1) model can be written as:

11

∆y t = µ + u t

(5)

u t ~ N(0, H t )

(6)

H t = C ′C + A′ε t ε t′ A + B ′H t −1 B

(7)

where yt = ( p1,t , p 2,t )′ , µ is a vector of constants, H t is the conditional variance matrix, and N(0, H t ) is multivariate conditional normal density. In the bivariate case, C is a 2×2 lower triangular matrix with three parameters and B is a 2×2 square matrix of parameters. The latter matrix depicts the extent to which current levels of conditional variances are related to past conditional variances. A is also a 2×2 square matrix of parameters and measures the extent to which conditional variances are correlated with past squared errors (i.e. deviations from the mean). The elements of A capture the effects of shocks or events on volatility (conditional variance). In our case, the total number of estimated parameters is 13.

4. Econometric analysis As we can see in figures 1 and 3, the premium in the oil-linked contract in 2003 seems to have shifted, with spot prices increasing considerably relative to contract prices.4 The econometric analysis that follows is based on daily Brent and NBP prices. The Brent price is available at higher frequency compared to the Continental gas contracts, and by construction, these are highly related to the oil products. Therefore, Brent is used to represent the dynamics of this oil linked product. The price series from Zebrugge and TTF are so short that they do not contain enough observations before the potential shift in volatility. NBP is used to represent the spot gas market dynamics because of the high correlation between the different regional hubs. Before conducting the econometric analysis, the price series are transformed into natural logarithms.

4

To investigate whether this shift is statistically significant, we regressed the long term gas contract price on the NBP price and an intercept dummy. A test of the significance of the dummy gave a p-value of 0.02. Hence, there is evidence of a structural shift in the relationship between the two prices. The shift might be caused by the increased price risk in the UK spot market. The test was conducted with a heteroskedasticity and autocorrelation consistent covariance matrix.

12

Several studies have indicated that oil and natural gas prices are non-stationary (e.g., Serletis and Herbert, 1999; Asche, Osmundsen and Tveteras, 2002; Asche, Osmundsen and Sandsmark, 2006, Silverstovs et al., 2005; Panagiotidis and Rutledge, 2007). We investigate the time series properties of the data series investigated here with an Augmented Dickey Fuller test where the null hypothesis is that the data series are non-stationary5. The test indicates that both prices are stationary in first differences. The results are reported in the table below.

Table 3: Unit root tests

Variable

t-adf

lags

∆NBP

-18.86**

1

∆Brent

-36.50**

12

5% critical value for : –2.86. 1% critical value:-.3.44 ** indicates significance on the one percent level

We also tested for “ARCH effects” (Engle 1982) on both series. The test statistic for both series concludes that the there is evidence of ARCH effects. This implies that past levels of volatility can be used to predict current volatility.

To investigate whether the variance process of NBP and Brent returns has changed after the apparent break in the price relationship between contract gas and spot gas, we split our sample in two, where the first sample ends at the estimated breakpoint in the relationship between the two prices of interest. As before mentioned, there have been several changes when it comes to how natural gas is traded in Europe. Derivatives trading together with increased volumes of spot trading might have altered price relationships in the European gas market. We estimate a multivariate GARCH model with a BEKK parameterization for the NBP and the Brent return series. This allows us to show how shocks in either market influence the other. Following the recommendations of Engle and Kroner (1995), several iterations were performed with the simplex algorithm to obtain the initial conditions. Quasi-maximum likelihood estimates for the full bivariate BEKK representation of the GARCH(1.1) model is shown in the table

5

Lag length was chosen to minimize Akaike’s Information Criterion.

13

below. Standard errors are robust to departures from the maintained assumption of conditional normality (Bollerslev and Wooldridge, 1992).

Table 4: Estimation results

Parameter

1996-2003

Parameter

Estimate p-value Mean

µ NBP µ Brent

2003-2006 Estimate

p-value

0.0031

0.3884

Mean 0.0009

µ NBP µ Brent

0.5966

0.0002 0.7487 Covariance structure c NBP , NBP 0.0233** 0.0000 c Brent , NBP -0.0019 0.4028

0.0007 Covariance structure c NBP , NBP 0.0323**

0.0028

c Brent , NBP

0.0133**

0.0000

c Brent , Brent

0.0030

c Brent , Brent

0.0000

0.9994

a NBP , NBP

0.5427** 0.0000

a NBP , NBP

0.4507**

0.0000

a NBP , Brent

-0.0019

0.8010

a NBP , Brent

-0.0101

0.1279

a Brent , NBP

-0.0287

0.8574

a Brent , NBP

0.6780**

0.0091

a Brent , Brent

0.1886** 0.0046

a Brent , Brent

0.0867

0.1103

bNBP , NBP

0.8329** 0.0000

bNBP , NBP

0.8209**

0.0000

bNBP , Brent

0.0029

0.5371

bNBP , Brent

0.0114*

0.026

bBrent , NBP

0.0473

0.5878

bBrent , NBP

-1.4541**

0.0009

bBrent , Brent

0.9733** 0.0000

bBrent , Brent

0.7906**

0.0000

0.1873

0.3584

** indicate significance at the one percent level *indicate significance at the five percent level

Table 4 indicates that in the second subsample (1996-2003) shocks to the oil price create volatility in the gas price, as the parameter a Brent , NBP is statistically significant. Before 2003, this was not the case. Shocks in either market did not create volatility transmissions in the other market. A possible explanation is that when the gas market infrastructure has had little available capacity the recent years, the gas price is more sensitive to shocks also in the market for substitutes. The increased liquidity and maturity of the spot market might have contributed to this result. When the spot markets are no longer used mainly for balancing purposes of the long term contracts, a richer set of short- and long term expectations might be incorporated in the current gas price and hence influence might come from more sources of risk.

14

In the second subsample, the level of conditional variance of both series is related to previous levels of conditional variance, and causality goes in both directions (by the significance of

bNBP , Brent and bBrent , NBP ). We do not find this relationship in the first subsample. Oil price shocks do not affect oil price volatility in the second subsample. However, this is found in the first subsample (by a Brent , Brent ). To check whether the oil price has started receiving volatility transmissions from exogenous shocks, we included another variable in our model, the value weighted stock market index consisting of NYSE, AMEX, and NASDAQ from CRSP. We did not find any additional volatility transmission in this trivariate model. For the gas price, shocks to own price affect volatility in both periods ( a NBP , NBP ). Applying a Newey-West heteroskedasticity and autocorrelation consistent covariance matrix (Newey and West, 1987) indicates that the results have not been affected by heteroskedasticity and/or autocorrelation.

From our econometric and descriptive analysis we can clearly see that the long term continental supply contracts are more important than before when it comes to price risk reduction. Given that the spot gas prices continue to be highly volatile in the future, we might see that the risk premium in the contract prices will increase compared to historical levels. ???

5. Conclusions The more liquid gas spot markets in Europe have high price volatility. In the period investigated (1996-2006) the continental gas contracts sell at a premium compared to spot gas sales, and have lower volatility. The premium can be explained by arbitrage opportunities provided by swing facilities in the take-or-pay contracts and payment for risk reduction. The low volatility confirms the take-or-pay contract as a means for risk reduction. We observe that the more liquid the spot gas market, the more independently it is priced from Brent blend. Seasonal variations at NBP result in spot gas becoming more expensive than contract gas in the winter. Although NBP has been cheaper than the Continental gas contract on average, NBP also has a higher probability of high gas prices over long time intervals because of extreme seasonal variations and volatility. We also provide empirical evidence that there has been a price shift in the relationship between the spot gas market and the contract gas market after 2003. After this point of time, shocks to the oil price have started to create volatility in 15

spot gas prices. A possible explanation is that when the gas market infrastructure has had little available capacity in recent years, the gas price is more sensitive to shocks also in the market for substitutes. The increased liquidity and maturity of the spot market might have contributed to this result. When the spot markets are no longer used mainly for balancing purposes of the long term contracts, a richer set of short- and long term expectations might be incorporated in the current gas price and hence influence might also come from other sources of risk. As there has been a shift in the relationship between spot and contract gas prices, we might see a restructuring of the old oil linked contracts: more contracts may be linked to spot gas prices, or customers may have to pay a higher premium for the increased risk reduction. Whereas the former is affirmed by anecdotic evidence in the gas market, the latter remains to be confirmed.

16

References

Asche F., Osmundsen. P. and M. Sandsmark (2006). “The UK Market for Natural Gas, Oil and Electricity: Are the Prices Decoupled?” The Energy Journal vol. 27, No. 2. Asche F., Osmundsen. P. and R. Tveteras (2002). “European market integration for gas? Volume flexibility and political risk.” Energy Economics, vol 24, issue 3, p249 Bollerslev, T. (1986). “Generalised autoregressive conditional heteroscedasticity.” Journal of Econometrics, 51:307-327. Bollerslev, T., Engle, R.F., Wooldridge, J.M., (1988). “A capital asset pricing model with time varying covariances. Journal of Political Economy 96, 116–131. Bollerslev, T., and , J. M. Wooldridge (1992). "Quasi-maximum likelihood estimation and inference in dynamic models with time-varying covariances." Econometric Reviews, 11: 143172. De Vany, A., and W. D. Walls (1993). "Pipeline Access and Market Integration in the Natural Gas Industry: Evidence from Cointegration Tests." Energy Journal 14: 1-19. Engle, R. F. (1982). “Autoregressive conditional heteroscedasticity, with estimates of the variance of United Kingdom inflation”. Econometrica 50:987-1007. Engle, R. F. (2002). “Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models.” Journal of Business & Economic Statistics, 20: 339-350. Engle, R. F. and Kroner, K. F. (1995). “Multivariate simultaneous GARCH.” Econometric Theory, 11:122-150. Ewing, B.T., Malik, F., Ozfidan, O., (2002). “Volatility transmission in the oil and natural gas markets.“ Energy Economics 24, 525–538. Newey, W. and K. West (1987). “A simple positive-definite heteroskedasticity and autocorrelation consistent covariance matrix”. Econometrica, vol 55. 703-708. Panagiotidis, T., E. Rutledge (2007). “Oil and gas markets in the UK: Evidence from a cointegrating approach.” Energy Economics 29: 329–347 Pindyck, R., (2004a). “Volatility in natural gas and oil markets. “Journal of Energy and Development 30 (1), 1–19. Pindyck, R., (2004b). “Volatility and commodity price dynamics.” Journal of Futures Markets 24 (11), 1029–1047. Serletis A. and J. Herbert (1999). “The Message in North American Energy Prices.” Energy Economics 21: 471–483. 17

Silverstovs, B., G. L’Hegaret, A. Neumann, and C. Hirschhausen. (2005). “International Market Integration for Natural Gas? A Cointegration Analysis of Prices in Europe, North America, and Japan.” Energy Economics 27: 603–615. Xiaoyi Mu (2007). “Weather, storage, and natural gas price dynamics: Fundamentals and volatility.” Energy Economics 29 46–63.

18