University of Amsterdam Faculty of Economics and Business BSc Thesis Economics
Price Elasticity of Natural Gas
Empirical Analysis on Supply and Demand in the U.K.
June 2014
Abstract
In the next decade, the natural gas market in the United Kingdom will be facing far reaching structural changes (e.g. strong expected decline in domestic production, higher dependency on imports and increasing carbon taxes). The potential implications of such changes are important to all market participants and price elasticity plays an important role in modeling the effects of such implications. This paper examines the short-‐run price elasticity in the natural gas market in the U.K. – using Instrumental Variable regression on a daily aggregated dataset from 2011 until 2013. Previous analyses on short run price elasticity generally show a short run inelastic price for supply and demand, partially attributable to rising short term infrastructure for producers and substantial substitution costs for consumers in the short run. The estimates in this thesis are in line with estimates from previous studies that were mostly conducted on gas markets outside the U.K.
Author Ramon Hoebink (10000430) Supervisor Lucyna Górnicka
Contents
1
Introduction ... 3
2
Literature review ... 4
3
The British natural gas market ... 6
National Grid ... 6
3.1
Efficiency and liquidity ... 7
3.2
Supply and demand ... 7
3.3 4
Methodology ... 9
Price elasticity ... 9
4.1
Data ... 10
4.2
Description of variables ... 10
4.3
Limitations ... 11
4.4
Regression model ... 12
4.5
Endogeneity ... 13
4.6 4.6.1
Instrumental variables ... 13
Seasonality and other biases ... 14
4.7 4.7.1
Weather variable ... 15
4.7.2
Dummy variables ... 15
5
Results and analysis ... 16
Test of instrument strength ... 16
5.1
Test of endogeneity ... 17
5.2
Test of normality of residuals ... 18
5.3
Test of heteroscedasticity and (log) linearity ... 19
5.4
Test of autocorrelation ... 19
5.5
Second stage regression ... 21
5.6 6
Conclusion and discussion ... 22
Validity of the model ... 22
6.1
Elasticity ... 22
6.2
Future research ... 23
6.3 7
References ... 24
1 Introduction
The United Kingdom has been a large producer of natural gas since the mid-‐late 1960’s when the first discoveries were made. From the late 1970s to the early 2000s, the U.K. was a major exporter of oil and gas. Supplies from the North and Irish seas peaked in 1999, however, production has since fallen by more than half (Figure 1-‐1). In 2010, total gas production was 59.8 billion standard cubic meters (bscm) whereas total gas demand was 99 bscm, making the UK a net importer of natural gas (IEA, 2012).
The government forecasts this decline in production to continue to 38.2 bscm by 2016; and projects import dependence to increase from around 41 percent in 2010 to more than 65 percent by 2025. Given the U.K. as one of the largest consumers of natural gas in Europe and the expected future growth in consumption mainly attributed to the use of natural gas in electricity generation. Natural gas as a source of primary energy in the United Kingdom accounts for 42% of total primary energy supply in 2010, a share that is greater than that in North America (28%) and the whole of Europe and Eurasia (34%),(Heather, 2010). Electricity is increasingly generated with natural gas power plants. In 1995, 13 percent of total generated electricity came from gas-‐fired power plants and this share rose to 39 percent in 2010 (Figure 1-‐2).
Figure 1-‐1. Production, consumption & reserves Figure 1-‐2. Electricity generation shares in the U.K.
Data source: U.S. Energy Information Administration (EIA) Data source: Digest of United Kingdom Energy Statistics
Participants in a natural gas market that is facing such structural changes, would need a thorough understanding of the dynamics of demand and supply in order to adapt their strategies. Within an economic framework, the key concept of natural gas demand and supply movers lies in the estimation of price elasticity, which is valuable information for consumers, producers and governments. Accordingly, analysis of elasticity of natural gas demand and supply can improve our understanding of natural gas markets. From a regulator’s perspective, elasticities of supply and demand can deliver useful information in order to develop the right regulatory framework for economic policies. Large consumers such as power generating companies might want to adapt their relative dependence on natural gas as a fuel for electricity generation when price elasticity of supply or demand is changing. Lastly, the price elasticity could affect the length of gas contracts offered by suppliers. According to Neuhoff and von Hirschhausen (2005), suppliers’ preferences for
0 200 400 600 800 1,000 1,200 0 20 40 60 80 100 120 91 93 95 97 99 01 03 05 07 09 11 13 Re ser ve s (b scm) Pr od uc tion , c on su m ption (b scm)
Proven reserves Production Consumption 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1990 1995 2000 2005 2010 Non-‐thermal Renewables & Storage Nuclear CCGT Conventional Thermal & Other
long-‐term contracts depend on the difference between short run and long run price elasticities of demand. If the long-‐run elasticity is significantly higher than the short run elasticity, the suppliers are said to prefer long-‐term contracts.
The objectives of this thesis are (1) to review and describe the dynamics of demand and supply in the British natural gas market, (2) to estimate recent short-‐run price elasticity of demand and supply of natural gas in the U.K. using Instrumental Variable regression on a simultaneous equation, (3) to analyze and discuss potential implications of the results. In this thesis, an overview of existing relevant literature is given in chapter 2, followed by a brief description of the British natural gas market in chapter 3. The methodology and the results are presented in chapter 4 and 5, respectively. The overall validity of the model, the potential implications of the results and suggestions for future research are discussed in chapter 6.
2 Literature review
Although the demand side of the natural gas market has been thoroughly researched, the supply side has often been ignored and left with mostly outdated estimations. Few studies on elasticity treat price as simultaneously determined by supply and demand conditions. Furthermore, most studies had a focus on either the U.S. market or the worldwide market, and no relevant studies were found with a focus on the U.K. market. Dahl (2007), has found over 1900 references on demand elasticity of which 177 were conducted on the natural gas market. Most of these studies have used annual, quarterly or monthly datasets that do not take into account the daily price volatility; hence, the analysis in this thesis differs in time, location and methodology from the existing analyses. This literature review attempts to briefly describe the findings in some of the existing studies on short run price elasticity. Krichene (2002) is one of the most recent contributors to the list of existing gas price elasticity estimations. The model that he employs is somewhat similar to the constructed model in this thesis in the sense that: (i) he makes use of a simultaneous equation model on demand and supply that is regressed using TSLS and (ii) he makes a distinction between short-‐run and long-‐run price elasticity. However, his data is gathered from worldwide supply and demand of natural gas from 1918 to 1999. In addition, he applied an Error Correction Model for potentially more accurate results. He found that short-‐run demand price elasticity was negative and very inelastic: -‐0.08 (non-‐significant) in 1918–1999, -‐0.39 (significant) in 1918–1973 and -‐0.01 (non-‐significant) in 1973–1999. Short-‐run price elasticities of supply were also low and negative, but significant: -‐0.14, -‐0.73 and -‐0.10, respectively. He concluded that natural gas supply was determined by existing production capacity, hence natural gas supply did not immediately react to changes in prices. The fact that the elasticity was negative could have been a consequence from (1) the nature of the demand curve (knowing the inelastic nature of demand, producers restrain output on purpose in order to stimulate price increases) or (2) that it indicates a short-‐run downward
supply curve arising from economies of scale in the industry. In the ECM, short-‐run price elasticities of demand were also low but significant: -‐0.06 in 1918–1999, -‐0.15 in 1918– 1973 and 0.04 in 1973–1999. The supply function showed low and negative short-‐run price elasticities: -‐0.59, -‐0.56 and 0.06, respectively. Concluding, his findings show low price elasticity in the short run for both supply and demand.
In contrast to the equilibrium model applied in this paper, Barret (1992) used a disequilibrium model in which the quantity exchanged was set equal to the minimum of quantity supplied or demanded with quantity supplied as a function of the natural gas price and reserves. The disequilibrium model yielded a short run natural gas supply elasticity that varied from 0.02 to 0.15. His equilibrium model showed lower values with a supply elasticity of 0.014. His data was gathered from the U.S. gas market between 1960 and 1990. Taylor (1977) reviewed a number of studies that also used aggregated data and an average price. He concluded that there is strong evidence that demand responds to prices, but that the actual magnitude of the elasticities is uncertain. Again, this is also confirmed by the many different outcomes on price elasticities in the existing literature. Taylor determined that the price elasticity of demand equals -‐0.15 for the short-‐run and more elastic than -‐1 for the long run.
Erikson and Spann (1971) used a different approach in which they estimated the price responsiveness on new discoveries of natural gas. By using variables such as wildcat well drilling, the success ratio and the average discovery size, they estimated the price elasticity of natural gas discoveries at +0.69. Although this price elasticity is not (directly) comparable to the price elasticity of supply of this paper, their analysis does provide us with insights on gas price determinants that might be hidden in our error term.
In a significant study on price elasticities of demand, Dahl (1993) concludes that, despite the use of various sophisticated models, and improved econometric technics on aggregated and disaggregated data, it appears that demand elasticity estimates will vary in every different (study or analysis). Aggregated demand for natural gas on a static model suggests a price elasticity of -‐0.27. Studies on aggregate and household data suggest that demand is price inelastic; whereas a comparative study making a distinction between households in the interstate and intrastate natural gas markets finds an inelastic response. Price elasticities across the industrial, electrical generation and commercial sectors show a rather wide variation in values. All average statistics on aggregate data suggest an inelastic price response; however, industry estimates tend to suggest an elastic response. Analyses that tend to ignore gas availability into consideration do result in rather high variations across regions. In summary, most of the studies reviewed above estimate inelastic short-‐run supply and demand of natural gas. The application of different models across different markets and time periods, amongst others, are reasons that help explain the differences in estimations.
3 The British natural gas market
The liberalization of the natural gas market in the United Kingdom has been a long process that started in the early 1980s, and reached a state of maturity only a few years ago. This was much later than United States’ “Henry Hub” gas market, but much earlier than the gas markets in continental Europe, which are still in the process of deregulation. The U.K. process of change was initiated by Prime Minister Thatcher's policy of privatizing state-‐ owned enterprises, rather than by a desire for liberalization per se.
Before the privatization of British Gas Corporation (BGC) in 1982, BGC had a monopoly on buying all gas produced in Britain, and was the sole gas supplier in Britain. The Oil and Gas Enterprise Act of 1982 was supposed to open access to the British’ Gas pipeline system to third parties. However, trade with third parties only began after the Gas Act of 1986, which included the licensing regime for gas transporters, shippers and suppliers. Customers with an off-‐take of more than 25,000 therms per year were now able to buy from other suppliers. In 1992, the gas supply tariff monopoly was removed and the tariff threshold was lowered to 2,500 therms; and by 1994, 45 percent of the customers above the 2,500 therms threshold were buying from competing gas companies (Eni, 2014). The publication of the Network Code in 1996 -‐ a framework agreement that defined the rules governing third party access to the gas transmission system, the deregulation and transformation was in its final stage.
The number of participants in the British wholesale gas market increased from less than 15 in 1995 to more than 80 by 2010. Counter-‐parties on the natural gas market now consist of banks, investment/pension funds, gas producers, utility companies and proprietary traders. The gas market U.K. thus has developed to the most liquid gas trading point in Europe where the ratio between traded and physical deliveries is more than 10 (Heather, 2010).
National Grid
3.1
The natural gas produced from the wells in the North Sea is transported and distributed through a high-‐pressure transmission network called the National Transmission System (NTS), operated by National Grid. The NTS supplies over 60 directly connected customers (large industrial clients and power plants) and 12 Local Distribution Zones (supplying small and medium consumers). The small users include domestic and business customers, but also the 16 independent gas transporters.
The gas that flows through the network of pipelines needs to be balanced at all times in order to keep a constant pressure of natural gas in the NTS primarily for safety reasons but also to ensure a balanced demand and supply. Suppliers and shippers are responsible for contract gas volumes and network capacity to meet consumer demand; whiles National Grid is responsible for ensuring both the availability of network capacity to meet anticipated transportation requirements and balance in the market.
The balancing takes place at the National Balancing Point (NBP) -‐ a virtual point where shippers nominate their buys and sells. The Shippers are in effect the “wholesalers” of
natural gas, buying from producers and selling to the suppliers. The primary objective of NBP was to balance the system, however, it quickly evolved as a trading point as well. Traders had confidence in buying and selling gas on a standardized basis at the most liquid point in the UK’s high pressure transmission system. The NBP has become an important component of British over-‐the-‐counter (OTC) market
Efficiency and liquidity
3.2
Efficiency and liquidity in a market are important aspects of price determination. In order to correctly estimate price elasticity, it is important to understand whether the markets are efficient. Demand and supply should intersect without restrictions because such restrictions might affect the estimates of price elasticity. The Efficient Market Hypothesis states that prices fully reflect all information available to the market (Fama, 1970). Market efficiency is attained in a competitive market through the price mechanism, which Hayek (1945) considers as the most efficient instrument to aggregate the asymmetrically dispersed information of market participants. Only when new information becomes available will prices change. A liquid and efficient market should facilitate the processing of information into valid price signals.
The NBP is the most liquid trading hub (Heather, 2010) with the highest informational efficiency in Europe (Nick, 2013). There is a very transparent volume and price information, which enables the participation of many in the market. Gas is brought onto the national grid physically at System Entry Points, but is traded most commonly at the NBP. According to the IEA (2007), over 50 percent of the gas consumed in the UK had been traded. The other 50 percent of gas not traded is sold and purchased on longer-‐term contracts.
An often used metric to estimate liquidity in a certain market is the so-‐called churn rate, which is a measure of the number of times a ‘parcel’ of the commodity is traded and re-‐ traded. Markets are said to have reached maturity when the trading churn is in excess of 10. According to Heather (2010), The NBP churn rate reached 17 for the U.K in 2010, lower than that of the Henry Hub in the U.S., but higher than that of its respective European counterparts (4 for Zeebrugge in Belgium, 3 for TTF in the Netherlands and 2.5 for NCG in Germany).
Supply and demand
3.3
Figure 3-‐1 shows the components that make up total supply, which is defined as all natural gas that flows into the NTS system. Gas flows originate from the following sources: gas production fields, withdrawals from storage and imports.
As shown in Figure 3-‐1, total gas supply is higher in winter months than in summer months. Although gas production seems to be a less volatile component than import or storage withdrawal, it still shows a seasonal pattern with lower deliveries during summer months.
Figure 3-‐2 shows the components that make up total demand, which is defined as all natural gas that flows out of the NTS system. Components of total gas demand includes include: Local Distribution Zones (LDZ), NTS power stations, exports, storage injection and large industrial clients who directly buy from the NTS.
Figure 3-‐1. Monthly natural gas supply per source (2013)
Data source: National Grid (2013)
Figure 3-‐2. Monthly natural gas demand per off-‐taker (2013)
Data source: National Grid (2013)
In 2010, the U.K.’s working storage capacity was 4.4 bscm in (IEA, 2012). Due to declining production and increasing import dependence, storage has become more important as a mean to provide flexibility. Furthermore, the deregulation of gas markets has meant that storage facilities are now available for commercial use in addition to operational use, and so gas storage now allows traders to exploit seasonal variations in the market price for natural gas (Breslin, 2008). Storage injection and withdrawals are important components of supply and demand. From April until November, storage levels tend to increase due to more injections than withdrawals; hence, production and imports are higher than consumption
0 2 4 6 8 10 12
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Su pp ly (b scm ) Storage withdrawal Import (LNG) Import (Pipeline) Production 0 2 4 6 8 10 12
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
D em an d (b scm) Industrial Storage injection Export (Pipeline) Power stations LDZ
(including exports). This pattern changes in December when storage levels tend to decrease sharply because consumption including exports is now much higher than production and imports (Figure 3-‐3). Furthermore, we can see that prices tend to peak when gas storages are emptied out.
Figure 3-‐3. Daily NTS storage level and natural gas price
Data source: National Grid (2013)
4 Methodology
Price elasticity
4.1
Price elasticity is a measure that shows the responsiveness of the quantity of a good or service to a change in its price. It gives the percentage change in quantity in response to a one percent change in price (all other variables held constant). The ceteris paribus condition makes price elasticity a partial elasticity. Graphically speaking, it is represented by the slope of the demand or supply curve.
The law of demand states that an increase in price of a good will be accompanied with a decrease in demand of that good, resulting in a negative price elasticity of demand. The law
of supply states that suppliers will supply a larger quantity of a good at higher prices of that
good. As a result of this supply curves are upwards sloping i.e. its price elasticity is positive. These economic theorems only apply for the so-‐called ‘normal’ goods and services.
𝐸 =
!"!! !!
=
!(!"# !) !(!"# !)
Given the elasticity of demand as an important variation on the concept of demand, the formula above will apply. Outcomes > |1| are said to be elastic, meaning that a 1% change in the price leads to a more than 1% change in the quantity. Outcomes < |1| are said to be inelastic, meaning that a 1% change in the price leads to a less than 1% change in the quantity.
Elasticity can be measured either (1) over an interval along the demand and supply curve or (2) at a specific point on the demand and supply curve. Since the change in price is measured over a period of three years and thus can be relatively large, an interval measure was chosen for the purpose of this analysis. In general, elasticity varies along the demand and supply curve and across different time periods; however, the model in this paper estimates only one aggregate price elasticity for 2011 until 2013 based on a price range from 1.26 to 3.61 pence.
Data
4.2
In order to construct a valid regression and to analyze the price elasticity; data on supply, demand, price, and other variables were gathered. The study relies predominantly on data produced by the National Grid and the APX power index. These are reliable data sources that provide prices quantities and other variables, on a daily basis that ranges back to October 2010. For this analysis, only full year daily data were included from 2011 until 2013. The limited time range of the data – three years -‐ are likely to capture only short run price effects but this enables us to ignore the structural changes in the gas market that long-‐ run models usually suffer from.
Description of variables
4.3
𝑄!!= 𝑄
!!= 𝑄! Quantity of natural gas (mscm per day)
𝑝𝑟𝑖𝑐𝑒! Volume weighted average price of all natural gas trades on the OCM (pence per kWh)
𝑝𝑜𝑤𝑒𝑟! Volume weighted average price of electricity (pence per kWh)
𝑤𝑒𝑎𝑡ℎ𝑒𝑟! Composite weather variable (CWV)
𝑠𝑢𝑏𝑡𝑒𝑟𝑚𝑖𝑛𝑎𝑙! Total domestic natural gas production that flows onto the
National Transmission System (mscm per day)
𝑖𝑛𝑡𝑒𝑟𝑒𝑥𝑝𝑜𝑟𝑡𝑠! Exports of natural gas through pipelines (mscm per day) 𝑖𝑛𝑡𝑒𝑟𝑖𝑚𝑝𝑜𝑟𝑡! Imports of natural gas through pipelines (mscm per day)
𝑠𝑡𝑜𝑟𝑜𝑢𝑡! Withdrawals from natural gas storages (mscm per day)
𝑚𝑜𝑛𝑡ℎ!,! Dummy variable for each month: 1. if month f
0. otherwise
𝑤𝑒𝑒𝑘𝑒𝑛𝑑! Dummy variable for weekend: 1. if weekday
0. if weekend
Limitations
4.4
The elasticity estimated covers a time period of three years (from 2011 until 2013) with data gathered on a daily basis. One could wish that the data set was longer, however, it is restricted by the availability of the dataset provided by National Grid. Despite the limited time range, the dataset is large enough to estimate a significant price elasticity of demand and supply.
The nature of the variables included in the model and the daily data set are likely to only capture short run price effects. Given that the analysis only covers three years, it is likely to be less sensitive to structural changes in the gas market. Long run time series often suffer from too many structural changes to capture long run adjustments (Dahl, 1993).
The use of aggregated data does not distinguish between regional or sectorial differences in supply and demand, which could potentially result in biased estimates in the aggregated model. The level of aggregation in the data has been found to make a difference in the estimated results of demand models, with models estimated at finer levels of aggregation performing better than their more aggregated counterparts (Bohi and Zimmerman, 1984). Gas contracts play an important role in the dynamics of demand and supply in the natural gas market. This analysis makes no distinction between price discrimination among different groups of consumers (residential, industrial, etc.) or different gas contracts (short, medium or long-‐term), or different gas pricing mechanisms. For example, the U.K. has developed from a market with long-‐term contracts into a more liquid spot market. Long-‐ term contracts engage market participants on quantity and price for several years, that do not allow any demand-‐side response to price signals; however spot markets offer more
storin 1096 16.23498 14.92321 0 82.83 storout 1096 14.13661 20.38604 0 108.62 interimport 1096 22.27796 18.19531 0 114.1 interexport 1096 31.12066 15.85995 11.9 76.85 subterminal 1096 166.8908 38.31594 70.04 252.53 weather 1096 10.07867 4.591595 -.94 15.97 power 1096 52.5588 7.40372 33.48 92.56 price 1096 2.085689 .2632622 1.2552 3.6059 quantity 1096 243.7327 63.40287 112.15 416.56 Variable Obs Mean Std. Dev. Min Max
flexibility but also more volatility, suggesting that the short run price in the spot market would need to be more elastic than in a long-‐term contract market. This analysis does not control for the share of long-‐term gas contracts in trades and deliveries.
Finally, the volume weighted average price of all natural gas trades excludes taxes levied on consumers and therefore does not reflect the end price paid by consumers. Nevertheless the British “NBP” price is the most important component of end-‐user prices, which are set by the suppliers (IEA, 2012). This analysis restricts the scope to the aggregate high-‐pressure natural gas market in the U.K.
Regression model
4.5
The aim of this paper is to estimate price elasticities by making use of a simultaneous demand and supply equation that assumes that the natural gas quantity results as a function of pricing and other exogenous variables.
In order to estimate the proposed elasticities of supply we assume that the quantity of natural gas supplied to the market 𝑄!! , results as a function of own price 𝑝𝑟𝑖𝑐𝑒! gas production ( 𝑠𝑢𝑏𝑡𝑒𝑟𝑚𝑖𝑛𝑎𝑙!) , pipeline imports (𝑖𝑛𝑡𝑒𝑟𝑖𝑚𝑝𝑜𝑟𝑡!) and storage injections
(𝑠𝑡𝑜𝑟𝑖𝑛!), corrected for monthly variances 𝑚𝑜𝑛𝑡ℎ!,! .
For the proposed elasticities of demand a comparable equation is constructed in which the quantity demanded 𝑄!! results as a function of price 𝑝𝑟𝑖𝑐𝑒
! , weather circumstances
(𝑤𝑒𝑎𝑡ℎ𝑒𝑟!), pipeline export (𝑖𝑛𝑡𝑒𝑟𝑒𝑥𝑝𝑜𝑟𝑡!), storage injections (𝑠𝑡𝑜𝑟𝑖𝑛!), corrected for
structural differences between weekday and weekends with a dummy variable (𝑤𝑒𝑒𝑘𝑒𝑛𝑑!) The data for the quantity of gas supplied and demanded were gathered from the National Grid and measured on a daily basis. As previously discussed, supply and demand does not equal domestic production and consumption because natural gas can be transported for import/export or stored for future consumption. These factors complicate the construction of the model, and the use of right variables and valid instruments. The key assumption for the model remains: demand should equal supply. The data on supply and demand shows that balancing is generally accurate with little day-‐to-‐day differences, which are likely inefficiencies in the balancing mechanism or measurement errors. For the analysis on the simultaneous equations model, supply and demand must be exactly the same, in order have a valid simultaneous regression model. For the purpose of this analysis, the differences are ignored and the delivered quantity (demand) is used as the dependent variable for both equations.
In this analysis, a semi log form, in which only price and quantity are logged, was used to estimate price elasticity. Although linear forms are generally preferred (Bohi and Zimmerman, 1984) for the reason that price elasticity does not need to be constant at all price levels, the preferred model for this analysis is a semi-‐log model since it directly
measures the parameter estimates. Furthermore, I expect that the relationship between price and quantity of natural gas could better be analyzed in percentage terms.
The equations below have their own interpretations and are separately regressed but are eventually linked, because the observed quantity is determined by the intersection of demand and supply.
Demand: ln 𝑄!! = 𝛽!+ 𝛽!ln 𝑝𝑟𝑖𝑐𝑒! + 𝛽!𝑤𝑒𝑎𝑡ℎ𝑒𝑟!+ 𝛽!𝑖𝑛𝑡𝑒𝑟𝑒𝑥𝑝𝑜𝑟𝑡!+ 𝛽!𝑠𝑡𝑜𝑟𝑖𝑛!+ 𝛽! 𝑤𝑒𝑒𝑘𝑒𝑛𝑑! + 𝑢! Supply: ln 𝑄!! = 𝛾!+ 𝛾!ln 𝑝𝑟𝑖𝑐𝑒! + 𝛾!𝑠𝑢𝑏𝑡𝑒𝑟𝑚𝑖𝑛𝑎𝑙 𝑝!+ 𝛾!𝑖𝑛𝑡𝑒𝑟𝑖𝑚𝑝𝑜𝑟𝑡!+ 𝛾!𝑠𝑡𝑜𝑟𝑜𝑢𝑡! + 𝛾!𝑚𝑜𝑛𝑡ℎ!,!+ 𝑢!
Endogeneity
4.6
The simultaneous nature of the demand and supply model makes the variable price in both equations endogenous. Quantity (supply and demand) is jointly determined by price through the equilibrium mechanism and therefore will cause the variable to correlate with the error term, hence making the regression analysis inaccurate.
In order to solve this endogeneity problem, instrumental variables were found by making use of intuition, economic reasoning and statistical technics. The instrumental variable should only capture the effects on quantities with shifts in price. Two conditions needs to be met: first, the instrument must be exogenous, hence the covariance of the instrument and the error term must equal zero. This cannot be tested but should follow fundamental economic logic. Second, the instrument must correlate with the endogenous variable price, hence the covariance of price and the instrument is different from zero. This will be tested in the first stage regression of TSLS. However, a significant correlation is not sufficient and substantive reasoning based on economic logic is still needed.
4.6.1 Instrumental variables
In the supply equation, two instruments are used for the endogenous variable for price: weather conditions (𝑤𝑒𝑎𝑡ℎ𝑒𝑟!) and the price of electricity (𝑝𝑜𝑤𝑒𝑟!).
The weather is a strong influence on the demand for natural gas and thus correlates with prices (see section on seasonality). In order to apply 𝑤𝑒𝑎𝑡ℎ𝑒𝑟! as an instrumental variable,
it is required that it does not correlate with supply, or with any other exogenous variables in the supply equation. The gas production itself does not depend on daily weather conditions (note, however, that extreme weather circumstances could potentially affect supply by causing disruptions; for the purpose of this analysis this has not been taken into account). Besides gas production from the wells, total supply also depends on gas withdrawn from storage and gas imports. These quantities could potentially correlate with
weather conditions, however, this is again an indirect correlation through demand and price that consequently affects supply from import and storage withdrawal.
In 2008, approximately 39 percent of the electricity produced in the U.K. was generated with natural gas; hence a higher gas price is likely to result in higher electricity prices as well (Heather, 2010). Furthermore it is unlikely that electricity price correlates with gas supply other than through the gas price simply because electricity costs are not the major operational costs for gas producers. That notwithstanding, electricity price might affect import/export and storage levels indirectly, but this is happening through the demand side of natural gas.
𝑺𝒖𝒑𝒑𝒍𝒚: ln 𝑝𝑟𝑖𝑐𝑒! = 𝜋!+ 𝜋!𝑤𝑒𝑎𝑡ℎ𝑒𝑟!+ 𝜋!𝑝𝑜𝑤𝑒𝑟!+ 𝑣!
In the demand equation, price is replaced by the variable for the daily U.K. gas production that flows into the NTS (𝑠𝑢𝑏𝑡𝑒𝑟𝑚𝑖𝑛𝑎𝑙!). The production variable does influence supply and
thereby the price; but it does not affect demand through variables other than price. Inversely, demand also does not affect gas production through variables other than price. Since daily production makes up a large share of daily quantity supplied, a decrease in production would likely affect prices.
𝑫𝒆𝒎𝒂𝒏𝒅: ln 𝑝𝑟𝑖𝑐𝑒! = 𝜋!+ 𝜋! subterminal + 𝑣!
Seasonality and other biases
4.7
Seasonality affects supply and demand quantities of natural gas. In this model, three variables are included that control for seasonal related effects. For example, natural gas demand in the U.K. peaks in the winter, subsequently decreases over the spring and summer months, and begins to rise in autumn. According to National Grid, average daily gas demand ranges from 250 to 300 mscm, whiles on an average winter day gas demand is 350 to 400 mscm. In Figure 4-‐1 total gas demand and average monthly temperature are graphed and the two lines seem to be negatively correlated.
Figure 4-‐1. Monthly total demand and average temperature (2013)
Data source: National Grid (2013)
0 2 4 6 8 10 12 14 16 18 0 2,000 4,000 6,000 8,000 10,000 12,000
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
°C
m
scm
In this regression model, three variables are used to control for seasonal effects: a weather variable and a dummy variable for weekends in the demand equation; and a dummy variable for months in the supply equation.
4.7.1 Weather variable
Due to the high rate of gas usage for heating, total gas demand will vary according to temperature fluctuations. As 2010 was a relatively cold year (1.1 degrees Celsius cooler than 2009), residential gas demand was 17 percent higher than the previous year (IEA, 2012). Thus in order to assure that the increase in fuel consumption is not due to a fall in temperature, a control variable is used. According to National Grid, temperature explains most of the variation in LDZ demand, which is an important component of total demand. To get a more precise fit, a variable was constructed by the National Grid that takes into account not only temperature, but also wind speed, effective temperature and pseudo seasonal normal effective temperature. This Composite Weather Variable (CWV) is measured from data that is extracted from all LDZ weather stations across the UK. In Figure 4-‐2, the regression of demand on CWV indicates negative linear correlation.
Figure 4-‐2. Daily demand regressed on CWV using OLS (2011 – 2013)
Data source: National Grid (2013)
4.7.2 Dummy variables
The dummy variable 𝑚𝑜𝑛𝑡ℎ!,! was included in the supply equation to control for supply
differences in each month. The month variable was not included in the demand equation for the reason that the weather variable was already controlling for seasonal bias.
From 2011 until 2013, the average daily gas demand on weekdays equaled 247.3 mscm and in weekends 234.9 mscm. In order to control for these structural differences across the week, the dummy variable 𝑤𝑒𝑒𝑘𝑒𝑛𝑑! was included in the demand equation.
Other bias may result from variables that are missing in the model that affects natural gas demand. If these variables are correlated with price, the affects will be attributed to price by the estimating model with the direction of the bias uncertain; depending on the relationships between the included and excluded variables.
y = -‐12.427x + 368.98 R² = 0.8099 0 100 200 300 400 500 600 0 2 4 6 8 10 12 14 16 D em an d ( m scm )
5 Results and analysis
In this chapter the results of several tests are presented that include analyses on the validity of the model (instruments’ strength, endogeneity, normality, heteroscedasticity, linearity and autocorrelation) as well as an analysis on the second stage regression that enable estimating the price elasticity.
Test of instrument strength
5.1
The results of the first stage regressions for both supply and demand are presented below. To check for instrument relevance (strength of instruments), the F-‐statistic was calculated for the null hypothesis that the coefficients of the variables used as instruments are all zeros in the first-‐stage regression. If the F statistic is not significant, then the additional instruments have no significant explanatory power for ln 𝑞𝑢𝑎𝑛𝑡𝑖𝑡𝑦! after controlling for
the effect of ln 𝑝𝑟𝑖𝑐𝑒! . The higher the F-‐statistic, the stronger the instruments and an F-‐ statistic less than 10 indicates possibly weak instruments. The F-‐statistics of supply and demand both meet this condition at approximately 94.8 and 97.0, respectively, and do not indicate potential weak instruments.
Furthermore, all instrumental variables are significantly different from zero at a 5% significance level (even at a significance level of 0.1%); hence there is a significant correlation with the endogenous variable 𝑝𝑟𝑖𝑐𝑒! .
The R2 is obtained from fitting the first-‐stage regression by OLS. Higher values indicate
stronger instruments, and instrumental-‐variables estimators exhibit less bias when the instruments are strongly correlated with the endogenous variable. The R2 values from the
first stage regression in Figure 5-‐1 give some idea of the adequacy of the instrument. A very low value indicates a large loss in precision in estimating the elasticity, while a very high value indicates such strong collinearity that might cast doubt on whether the instrument is really exogenous.
Figure 5-‐1. First stage regression of supply (left); and demand (right)
_cons .6470688 .0423421 15.28 0.000 .5639867 .730151 power .0055716 .0003784 14.72 0.000 .0048291 .0063141 weather -.0176882 .0019823 -8.92 0.000 -.0215779 -.0137985 nov .0391445 .0122148 3.20 0.001 .0151771 .0631118 oct .0840787 .0149839 5.61 0.000 .0546779 .1134795 sep .0913226 .0189569 4.82 0.000 .054126 .1285192 aug .0941836 .0205548 4.58 0.000 .0538517 .1345154 jul .113726 .0199285 5.71 0.000 .074623 .1528291 jun .0845523 .0187207 4.52 0.000 .0478193 .1212853 may .0780283 .0165895 4.70 0.000 .045477 .1105796 apr .0491188 .0137624 3.57 0.000 .0221147 .0761229 mar .035189 .0120579 2.92 0.004 .0115295 .0588485 feb -.0363873 .0124157 -2.93 0.003 -.0607489 -.0120257 jan -.0801045 .0118573 -6.76 0.000 -.1033705 -.0568385 interimport .0020356 .0001971 10.33 0.000 .0016488 .0024224 storout .0004431 .0002015 2.20 0.028 .0000477 .0008384 subterminal -.0007912 .0001431 -5.53 0.000 -.0010719 -.0005105 lnprice Coef. Std. Err. t P>|t| [95% Conf. Interval] Root MSE = 0.0785 Adj R-squared = 0.5762 R-squared = 0.5824 Prob > F = 0.0000 F( 16, 1079) = 94.04 Number of obs = 1096 First-stage regressions _cons 1.180094 .0368519 32.02 0.000 1.107786 1.252403 subterminal -.0014256 .0001583 -9.01 0.000 -.0017362 -.0011151 weekend .0277036 .0070797 3.91 0.000 .0138122 .041595 interexport -.0018632 .0002344 -7.95 0.000 -.0023232 -.0014033 storin .0006548 .0002667 2.46 0.014 .0001315 .0011781 weather -.0185463 .0016257 -11.41 0.000 -.0217362 -.0153565 lnprice Coef. Std. Err. t P>|t| [95% Conf. Interval] Root MSE = 0.1005 Adj R-squared = 0.3047 R-squared = 0.3079 Prob > F = 0.0000 F( 5, 1090) = 96.99 Number of obs = 1096 First-stage regressions
However, it can be misleading to only take into account R2. If ln 𝑝𝑟𝑖𝑐𝑒! were strongly
correlated with the exogenous variables, but only weakly correlated with the additional instruments, then these statistics could be large even though a weak instrument problem is present. The partial R2 statistic measures the correlation between ln 𝑝𝑟𝑖𝑐𝑒! and the
additional instruments after eliminating the effect of the exogenous variables. Unlike the R2
statistics, the partial R2 statistic will not be overestimated because of strong correlation
between the endogenous variable ln 𝑝𝑟𝑖𝑐𝑒! and the other exogenous variables. The R2
statistic for supply and demand equal 0.58 and 0.31, respectively; while the partial R2 for
supply and demand equal 0.21 and 0.07, respectively (Figure 5-‐2). These partial statistics are significantly lower, but still validate the use of the variables as instruments.
Figure 5-‐2 also contains the nominal 5% Wald test which defines a set of instruments to be weak if a Wald test at the 5% level can have an actual rejection rate of no more than 10%, 15%, 20%, or 25%. Assuming we can accept a rejection rate of at most 10%, we can reject the null hypothesis that the instruments are weak, because the minimum eigenvalue statistics from both demand and supply are larger than the critical values obtained from the Wald test.
Figure 5-‐2. Additional statistics from first stage regression of supply (left); and demand (right)
From the performed tests above we can conclude that the variable ln 𝑝𝑟𝑖𝑐𝑒! is indeed endogenous and that the variables 𝑠𝑢𝑏𝑡𝑒𝑟𝑚𝑖𝑛𝑎𝑙!, 𝑝𝑜𝑤𝑒𝑟! and 𝑤𝑒𝑎𝑡ℎ𝑒𝑟! are strong
instruments that are unlikely to correlate with the error terms in the second stage regression.
Test of endogeneity
5.2
Since the instrumental variables are uncorrelated with 𝑢!, ln 𝑝𝑟𝑖𝑐𝑒! is correlated with 𝑢!,
only if 𝑣! is correlated with 𝑢!. To test whether this is the case, an endogeneity test from
Durbin and Wu-‐Hausman was performed. Under the null hypothesis of both tests, the variables under consideration are exogenous. The results are presented in Figure 5-‐3. Both tests are highly significant, thus allow for rejecting the null hypothesis of exogeneity and for treating the variable ln 𝑝𝑟𝑖𝑐𝑒! as endogenous.
LIML Size of nominal 5% Wald test 8.68 5.33 4.42 3.92 2SLS Size of nominal 5% Wald test 19.93 11.59 8.75 7.25 10% 15% 20% 25% 2SLS relative bias (not available) 5% 10% 20% 30% Ho: Instruments are weak # of excluded instruments: 2 Critical Values # of endogenous regressors: 1 Minimum eigenvalue statistic = 147.589
lnprice 0.5824 0.5762 0.2148 147.589 0.0000 Variable R-sq. R-sq. R-sq. F(2,1079) Prob > F Adjusted Partial First-stage regression summary statistics
LIML Size of nominal 5% Wald test 16.38 8.96 6.66 5.53 2SLS Size of nominal 5% Wald test 16.38 8.96 6.66 5.53 10% 15% 20% 25% 2SLS relative bias (not available) 5% 10% 20% 30% Ho: Instruments are weak # of excluded instruments: 1 Critical Values # of endogenous regressors: 1 Minimum eigenvalue statistic = 81.1544
lnprice 0.3079 0.3047 0.0693 81.1544 0.0000 Variable R-sq. R-sq. R-sq. F(1,1090) Prob > F Adjusted Partial First-stage regression summary statistics