Estimating US Long-Run Demand for Major Fossil Fuels

Transcription

1 Estimating US Long-Run Demand for Major Fossil Fuels Dragan Miljkovic and Lei Zhang North Dakota State University Department of Applied Economics 2016 IAEE Conference, Perth, Australia February 14-17, 2016

2 Background The US stands as one of the largest energy producers and consumers in the world. Natural gas, oil, and coal are considered pivotal energy resources in both the United States (US) and around the world. These fossil fuels represent 80% of all energy consumption in the United States. While it may seem obvious that a relationship exists among these three energy commodities, it may be difficult to accurately model it. Natural gas, oil, and coal are mined, delivered, and used in different ways which makes substitution difficult to show but logically plausible since they are major energy inputs for many industries.

3 Fossil Fuels Consumption in the United States,

4 Objective The purpose of this study is to examine the demand relationship these energy resources share in the US. Given US (relative) independence from world gas and coal markets, and growing independence from world oil markets (due to increased domestic production-fracking), the analysis will focus on US markets only. Moreover, this study further explores the substitutability between energy/fossil fuel inputs within the energy industry in the US.

5 Literature Review Pindyck s (1979) paper on global energy demand and pricing is one of the benchmarks in this literature. Among several contributions this paper made to the energy economics literature, it was the first paper that established and measured substitutability among three main fossil fuels in the United States: oil, natural gas, and coal. Uzava s partial elasticities of substitutions for the three fossil fuels reported in the paper have all been pretty small, and the signs inconsistent. Hence, other than hypothesizing the existence of a relationship among the three fossil fuels, the paper did not succeed in establishing a well-defined nature, i.e., the substitute or complement, of that relationship between the commodities.

6 Literature Review Krichene (2002) analyzed the world crude oil and natural gas output and prices during A simultaneous supply and demand model was used for the specification and identification of elasticities. Bachmeir and Griffin (2006) tested for market integration of the crude oil, coal, and natural gas markets in the United States using cointegration analysis. Their conclusion is rather different from Pindyck s in a sense that they found these three markets to be weakly integrated, i.e., there is not a primary energy market.

7 Literature Review Mjelde and Bessler (2009) also conducted an energy market integration study using the vector error correction model, and provided a comprehensive literature review on the issue. They showed that energy prices are cointegrated or linked. Yeboah et al. (2014) use a panel SUR model to estimate the substitution between different forms of energy in US electricity generation, while also providing a comprehensive literature review on this issue. The factor share equations for different forms of energy were derived from a translog cost function for 48 states from 1970 to The results from the empirical application suggest limited substitution potential for all the energy inputs. Further, natural gas was found to be the main substitute while wood and waste was a net compliment.

8 Data The seven variables used in the study are represented as follows: natural gas price (NGP), oil price (OILP), coal price (COALP), natural gas consumption (NGCSP), oil consumption (OILCSP), coal consumption (COALCSP), GDP per capita (GDPPC), plus oil shock dummy (OS#). The data collected is secondary data from the Energy Information Administration, Historical Statistics of the United States, and Bureau of Mines Minerals Yearbook. The annual data for all of the variables was collected for the years , and for the US market only.

9 Data The natural gas price is based on the wellhead price, the price of natural gas at the wellhead including all costs before shipment from the lease. Oil price is determined by the first purchase of crude oil from the property. Coal price is the price of coal purchased at the mine, less freight or shipping and insurance costs. All prices used and GDPPC are in nominal US dollars.

10 Data Consumption for each energy source is calculated by taking total production minus net exports. Natural gas production is a marketed production which is defined by the Energy Information Agency as, Gross withdrawals less gas used for repressuring, quantities vented and flared, and nonhydrocarbon gases removed in treating or processing operations. It includes all quantities of gas used in field and processing plant operations. Coal production is total coal production including all types: bituminous, subbituminous, lignite, and anthracite. Oil production is of crude oil.

11 Data Quantities for natural gas, oil, and coal respectively: million cubic feet, thousands of barrels, and short tons. Prices are dollars per thousand cubic feet, dollars per barrel, and dollars per short ton for natural gas, oil, and coal correspondingly. The oil shock dummies are used to capture large swings in oil and natural gas prices for the following years and corresponding reasons: 1920 supply shortage, 1921 gains in production from Texas, California, and Oklahoma, post war demand increased due to transition to automotive transportation, the end of the Korean War price controls, Suez Canal crisis, OPEC embargo, Iranian Revolution, Iran-Iraq War, First Persian Gulf War, East Asian financial crisis, US recession and 911 terrorist attacks, and The Great Recession (Hamilton, 2011).

12 Time-Series Tests Preceding Model Specification and Estimation Five steps to test the dynamic relationship among natural gas, oil, and coal variables: (1) test for unit roots to determine if the data is stationary or follows a random walk; (2) use cointegration techniques to identify long-run relationships; (3) test for causality among the variables using Granger causality test; (4) test for weak exogeneity to identify variables that are determined outside the system; and (5) use acyclic directed graphs to determine contemporaneous causality and in turn endogeneity or exogeneity.

13 Unit Root Tests Results By using the KPSS and ADF tests we conclude that the data does not have a unit root and is stationary after being differenced once, I(1). The null hypothesis of the KPSS test is that the series is stationary which makes accepting of the null harder and gives the test a higher power.

14 Cointegration Results The Johansen Cointegration test is used to find the number of cointegrating vectors among the variables (Johansen, 1991; Johansen & Juselius, 1994). Results from the maximal eigenvalue test and the trace tests indicate that two cointegrating vectors, at the 5% significance level, exist among the three energy prices and consumptions. This means that some or all six variables move in response to disequilibrium thus converging towards the long run equilibrium path/system.

15 Granger Causality Test Results The Granger causality tests (Granger, 1969), can help identify if variables that are assumed endogenous can be treated as exogenous. The results indicate that natural gas price Granger causes the coal price, coal price Granger causes the oil price, and oil price Granger causes the natural gas price. This cyclical causal flow makes it difficult to determine a meaningful causal relationship among the variables. Given these mixed results, the weak exogeneity test is used to test for exogeneity.

16 Weak Exogeneity Test Results According to Johansen and Juselius (1994), restriction to the cointegration vector can be used to detect structural relationships. Weak exogeneity of a variable can be tested to identify the effect it may have on the other variables in the long-run. Since there are two cointegrating vectors, the null hypothesis of weak exogeneity is accepted if the estimated coefficients of each variable in the two cointegrating vectors are both equal to zero. Based on the results, the null hypothesis of weak exogeneity is rejected for natural gas price, coal price, and coal consumption, while one can accept the null for oil price, oil consumption, and natural gas consumption. Hence, in the long-run, the results indicate that oil price, oil consumption, and natural gas consumption drive the prices of natural gas and coal, and coal consumption.

17 Directed Acyclic Graphs The weak exogeneity results are inconsistent with the Granger causality test, and fail to show unambiguously causation or endogeneity. To further test and confirm the causal structure of this market, directed acyclic graphs (DAGs) were used to show causality and exogeneity. Given that DAGs examine contemporaneous causal relationships rather than lagged relationships among variables, they complement the Granger causality and weak exogeneity tests. Directed acyclic graphs are visual representations of defined causal flows between and among a set of variables.

18 Directed Acyclic Graphs There are two algorithms that this study will focus on, PC and GES. GES graph fits the data better. TETRAD IV software used. The GES algorithm uses a different approach to create DAGs that use Bayesian posterior over alternative scores to search DAGs. The algorithm s first step is to begin with a DAG that has no edges connecting any of the variables. Then edges are added and/or directions reversed in a search across all possible DAGs to improve the Bayesian posterior score. Once a local maximum of the Bayesian score is found, which occurs when no edges or directions can be added, then edges are deleted or directions reversed as long as such actions improve the Bayesian posterior score (Chickering, 2002).

19 GES Graph-Coal consumption is the only variable that is completely endogenous in the system

20 Econometric Model Based on the Granger causality, weak exogeneity, and DAGs results one cannot say with certainty that the left hand side variables are endogenous, so a SUR model is appropriate rather than a VEC model. All of the variables except shock dummies and time trend are in log form so the estimated coefficients can be interpreted as elasticities. The oil shocks identified by Hamilton were mostly external shocks to the United States oil supply, and were then tested using Chow s breakpoint test to see if these shocks caused a structural change in the demand for natural gas, oil, and coal.

21 SUR Model Specification (1) ΔlnNGCSP t = β 1 + β 2 ΔlnGDPPC t + β 3 ΔlnNGCSP t 1 + β 4 ΔlnOILP t + β 5 ΔlnNGP t + β 6 ΔlnCOALP t + β 7 OS2 t + β 8 OS3 t + β 9 OS4 t + β 10 OS5 t + β 11 OS6 t + β 12 OS7 t + β 13 TREND t + ε 1,t (2) ΔlnOILCSP t = α 1 + α 2 ΔlnGDPPC t + α 3 ΔlnOILCSP t 1 + α 4 ΔlnOILP t + α 5 ΔlnNGP t + α 6 ΔlnCOALP t + α 7 OS2 t + α 8 OS3 t + α 9 OS4 t + α 10 OS5 t + α 11 OS6 t + α 12 TREND t + ε 2,t (3) ΔlnCOALCSP t = γ 1 + γ 2 ΔlnGDPPC t + γ 3 ΔlnCOALCSP t 1 + γ 4 ΔlnOILP t + γ 5 ΔlnNGP t + γ 6 ΔlnCOALP t + γ 7 ΔOS2 t + γ 8 TREND t + ε 3,t

22 SUR Results Independent Variables Intercept GDPPC NGCSP(-1) OILCSP(-1) COALCSP(-1) NGP OILP COALP OS2 OS3 OS4 OS5 OS6 OS7 TREND Dependent Variables NGCSP OILCSP COALCSP *** *** ** [ ] [ ] [ ] *** *** *** [ ] [ ] [ ] ** [ ] [ ] *** [ ] ** [ ] [ ] [ ] ** ** [ ] [ ] [ ] * [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] * [ ] [ ] ** ** [ ] [ ] [ ] *** *** [ ] [ ] [ ] R-squared Adj. R-squared Durbin H Statistic

23 SUR Results The SUR model demonstrated that there was low-level of substitutability among fossil fuels, but the small magnitude of the estimated coefficient indicates that natural gas, oil, and coal are more properly classified as independent goods than as substitutes of each other within the US market. The results found that lagged natural gas demand, GDP per capita, natural gas price, oil price, and the Iranian oil shock were all significant factors in determining natural gas demand over the past 95 years. Natural gas was found to be a normal good that is income inelastic, and also a weak substitute (or rather independent good) with oil.

24 SUR Results Results of the coal equation suggest complete independence among the three fossil fuels. The driving force of demand for coal has been income per capita suggesting how economic growth, as measured by income increase, directly and significantly drove the demand for coal. The previous period demand for coal also plays an important role in determining current period demand for coal suggesting an important role of storing in consumption of coal.

25 SUR Results The oil equation specification is not as good as the previous two equations. While a number of variables in this equation, such as income per capita, own price, price of coal (week substitutes), OPEC embargo, and Iranian oil crisis, are determined to have a significant impact on US demand for oil, the explanatory power of this equation is relatively low. A possible reason for it may be that the model did not include US oil imports although the US economy has been dependent, to some degree, on imported oil (Shafiee and Topal, 2008).

26 Future Research First, adding US oil imports to the oil equation could potentially improve the fit of the model. Second, a theoretical model based on the production economics theory of industry cost minimization which would include non-fuel input in energy production such as capital and labor, in addition to already present fuels inputs, could potentially provide a more comprehensive theoretical underpinning for an enlarged empirical model of the US demand for energy from fossil fuels. Yet, that would imply a common/single final good in consumption, which we have seen does not exist.

27 Questions?