MULTIPLE DETERMINANTS OF HOUSEHOLD NATURAL GAS DEMAND: A PANEL DATA ANALYSIS IN OECD COUNTRIES

Halim Tatli

Bingöl University, Faculty of Economics and Administrative Sciences, Bingöl 12100, Turkey

Keywords:Natural gas household demand Economic factors Climate and energy consumption Price and income elasticities OECD.

ARTICLE HISTORY: Received:27 September 2018 Revised:2 November 2018 Accepted:5 December 2018 Published:31 December 2018.

Contribution/ Originality:This study contributes to the current literature with a multivariate approach. Natural gas that demanded in the household sector is analyzed for OECD countries with a multivariate approach. In addition, new data sets are used in this study. The analysis is carried out with advanced econometric methods. The price and income elasticities of natural gas demand for OECD countries provide beneficial results for energy policy makers.

1. INTRODUCTION

Nowadays, natural gas is become more and more important in primary energy consumption on the global level. In 2016, the share of natural gas in global primary energy consumption is 31.8% (BP, 2017 ). Discovered natural gas reserves are also increased in parallel with the natural gas consumption. The natural gas reserve reached 186.6 trillion cubic meters in 2016, a 0.65% increase over 2015 reserve levels (BP, 2017 ). In the same year, the OECD countries, which own 9.5% of the world's total natural gas reserves, have 17.8 trillion cubic meters of natural gas reserves. Natural gas is  a source of energy  composed of hydrocarbons (methane (CH4), ethane (C2H6), propane (C3H8) and butane (C3H8)) and non- hydrocarbons gases (nitrogen (N2), carbon dioxide (CO2), helium (He), hydrogen sulfide (H2S) and water vapor) (Speight, 2017 ). Natural gas is used in many sectors including household, transportation, industry, energy production. The household natural gas consumption in Organization for Economic Cooperation and Development (OECD) countries has increased considerably in recent years. According to the International Energy Agency (IEA), household natural gas consumption increased from 288 746 million cubic meters to 306 719 million cubic meters over the period 1995-2014 (International Energy Agency, 2017 ). The notable increase in household natural gas consumption in OECD has been attributed to a few factors including generation relatively less carbon dioxide emissions than other fossil fuels (Apergis and Payne, 2010 ) the importance among alternative sources to reduce emissions of carbon dioxide (Shahbaz et al., 2013 ) and expansion of the gas network.

Many studies have investigated the relationship between natural gas demand and the effect of various factors including prices, income and population. In these studies are used different data type including cross-sectional data, time-series, aggregated data and cross-section time-series data. Cross-sectional studies include Leth-Petersen (2002 ) who use a cross-section data to estimate demand natural gas for Danish households. Brounen et al. (2012 ) use 2009 data for Netherlands to examine the extent to which individual characteristics and technical specifications of the dwelling variables can explain differences in households' behaviour. Other cross-section data studies include Brenton (1997 ); Fung et al. (1999 ) and Yoo et al. (2009 ).

Time series studies include Alves and da Silveira Bueno (2003 ); Erdogdu (2010 ); Payne et al. (2011 ); Madlener et al. (2011 ); Bernstein and Madlener (2011 ); Altinay and Yalta (2016 ) and Kani et al. (2014 ). Using time series data for Turkey, Cetin and Yüksel (2014 ) use Cointegration analysis with structural breaks to examine the long-term and short-term elasticities of demand for natural gas. Park and Zhao (2010 ) apply a time-varying cointegration model for the United States data for 1976 and 2008 to study the demand price and income elasticity of natural gas.

Cross-section time-series data include Bernstein and Griffin (2006 ); Alberini et al. (2011 ); Andersen et al. (2011 ); Bilgili (2014 ) and Harold et al. (2015 ). For example, Bilgili (2014 ) reports positive relationship between natural gas demand and income in eight OECD countries and finds that the income elasticity of natural gas is 1.333 which correspond to values of unit elastic. Moreover, Bilgili (2014 ) reports negative relationship between natural gas demand and prices and finds that the prices elasticity of natural gas is 1.126. Yu et al. (2014 ) have used 216 observations of Chinese's cities household data of natural gas consumption for the period of 2006-2009 and ran feasible generalised least squares approaches. Their estimation results have supplied price elasticity of -1.431 and significant income elasticity of 0.206. Another study using cross-section time-series data, Nilsen et al. (2005 ) use ordinary least squares, generalized least squares, seemingly unrelated regression and two-stage least  squares to examine the long-term and short-term elasticities of demand for natural gas for selected EU-12 countries, also addresses annual data for the period 1978- 2002. They reached long-run per capita natural gas price elasticities ranging from -1.541 to 1.844 and long run income elasticities ranging from 1.649 to 2.251. Yorucu and Bahramian (2015 ) uses cross-section time-series data for the years 2001 and 2012 and the panel dynamic OLS, panel fully modified OLS and traditional OLS methods to estimate the price elasticities of natural gas in the EU-12.  In a study conducted in the United Kingdom, Baker and Blundell (1991 ) estimated the gas and electricity demand function in the household sectors using data pooled from the family expenditure survey over the period 1972 to 1988. They have determined demand for natural gas to be as a rule price inelastic. In a recent study, Burke and Yang (2016 ) analyse the linkages among household natural gas demand, natural gas prices, income, population and climate of the 39 countries. They reached long-run natural gas price elasticities ranging from -0.50 to -0.68. Considering the studies directed about the natural gas demand in the literature, few study investigating the effect of temperature on the household natural gas demand is located. For that reason, in this study, we inspected the effect of the temperature on natural gas demand along with other factors including natural gas prices, per capita income, and population. For this reason, the effect of temperature on the household natural gas demand is not disregarded and the deviation caused by the neglected variable is reduced. Additionally, in order to expose the effect of price and income on natural gas demand more effectively, in the study, the price of natural gas and the per capita income data are in the purchasing power parity (PPP)¹ . Such use is not found in other studies. Aim of this study is to investigate relationship between natural gas demand and natural gas prices, per capita GDP, population, and climate for selected OECD countries.  The rest of the paper is organized as follows. The next section presents the data and methodology. The third section reports the empirical results. The last summarises the primary findings the paper.

2. DATA AND METHODOLOGY

The present study utilizes a balanced panel data set for the selected OECD countries in the period 2004-2015² . Since there are countries with missing data among the 33 countries, only 18 countries with complete data are used in this period. Annual data is utilized for 18 OECD countries which are Austria, Chile, Czech Republic, Germany, France, Hungary, Ireland, Korea, Mexico, Netherlands, Poland, Portugal, Slovak Republic, Spain, Switzerland, Turkey, the United Kingdom and the United States. Five variables are used in this study to build a natural gas consumption model for the selected OECD countries. The natural gas consumption (G) is utilized as the dependent variable measured in million cubic meters. Besides, four independent variables are utilized, namely, natural gas prices (P) measured US dollar/MWh in PPP, per capita gross domestic product (GDP) measured in constant 2011 international dollars in PPP, population (POP) measured in number of person, average annual temperature (TEMP) as an indicator climate measured Celsius. Except TEMP, all variables are in natural logs. Descriptive statistics and data sources for the variables used in this study are reported in Table 1.

Table-1. Descriptive statistics of variables used in the empirical study

Variable	Mean	Standard Deviation	Median	Min	Max
G	14215.6	30409.2	3934	206	143637
P	81.7	30.5	78.3	31	191.5
GDP	32839.9	12023.6	30528.8	15011.7	60944
POP	52314105	69857005	27874547	4070262	321000000
TEMP	10.81	3.39	9.8	6.3	21.6

Notes: The data for G is from the IEA (2017;2016a ) The data for P is from OECD/IEA (2016b ) The data source for the GDP, POP and TEMP are retrieved from the World Bank.

Table-2. Mean and standard deviation of the variables used in the study for the sampling (2004-2015).

Source: Has been compiled by using E-wievs software.

Table 2 shows the descriptive statistics on the variables for the sample of 18 countries included in this study. In this context, the country with the highest average consumption of natural gas in the analysis period is the United States and the country with the lowest natural gas consumption is Portugal. In addition, while the average price of natural gas is the highest in Chile, the US is the country with the lowest average natural gas price (Table 2).

Panel data analysis was conducted as the analysis method in this study. Panel data analysis models can be generally illustrated by Equation 1 (Tatoglu, 2013 ).

Where subscript i represents the units; subscript t represents time; βoit represents constant term; βkit represents the Kx1 dimensional vector of parameters; Xkit represents the value of k. independent variable for unit i. at time t, and Yit represents the value of the dependent variable for unit i. at time t. While estimating this model, it is assumed that the average of the error terms are zero, the variance is constant and normally distributed.

Three methods are used to estimate the pooled regression as the prediction method for both the cross-section and time series; Classical Model, Fixed Effects Model and Random Effects Model. The basic distinctive feature between them is that constant term is described in different ways.

According to Classical Model, the same constant term is used for all cross-section units in pooled regression model. Thus, both constant and slope coefficients are constant in terms of units and time in classic model. This model was presented by Equation 2.

As for Random effect model, differences belonging to the units are modeled into error term (Greene, 2010). In order to prevent a loss of degree of freedom encountered in fixed-effect models, changes by units or units and time are included into the model as a component of error model (Baltagi, 2005 ). As the unit effect is not constant in this kind of models and it is not within a constant parameter, but it is involved in the margin of error. Error term is indicated as qit ,

qit = uit + jit                                                                       (4)

in the panel data model of Equation 1. In Equation 4, uit indicates residual errors and jit indicates the unit error. In this context, random-effect model can be illustrated as seen in Equation 5.

According to Dahl (2012 ) the parameters which are predicted in a static model should be interpreted as intermediate-run estimates. In contrast, according to Pesaran and Smith (1995 ) the parameters which are estimated from cross-sectional variation based regressions can be interpreted as long-run parameters. In this study, the estimates have interpreted as long term coefficients by adopting Pesaran and Smith (1995 ) approach.

3. EMPIRICAL RESULTS

For the OECD countries, there is a possibility of the existence of cross-sectional dependence among its members. It is primarily required to test the cross-section dependence between the units in the variables. In order to decide if first generation or second generation tests have to be used to determine the stationary of the series³ .

The results of CDLM1, CDLM2 and CDLM-Adj tests are presented in Table 3. According to these tests, there is cross-section dependence in all variables for intercept model. Thus, it is required to test the stationary of series by second generation unit root tests.

Table-3. Results of Tests for Cross-Section Dependence

Variable	CDLM1	CDLM2	CDLM	CDLM-Adj
		Constant
lnG	561.491a (0.000)	22.323a (0.000)	5.624a (0.000)	21.505a (0.000)
lnP	1278.012a (0.000)	63.284a (0.000)	24.475 a (0.000)	62.465a (0.000)
lnGDP	853.091a (0.000)	38.993a (0.000)	23.119 a (0.000)	38.174 a (0.000)
lnPOP	1318.701a (0.000)	65.610a (0.000)	11.546 a (0.000)	64.791a (0.000)
TEMP	494.1481a (0.000)	18.473a (0.000)	15.830 a (0.000)	17.655 a (0.000)

Notes: a test values are significant at α= 0.01 level. Optimal lag length assumed to be 1. CDLM1  (Breusch and Pagan, 1980 ) CDLM2 (Pesaran, 2004 ) CDLM-Adj (Pesaran et al., 2008 ).

Stationary of the variables were tested by Bai and Ng (2004 ) PANIC attack tests taking into account cross-section dependence, and results are presented in Table 4. In stationary tests given in Table 3, it has been determined that all variables are stationary at the 0.01 significance level according to PANIC test. Therefore, least square method was used by estimating the study’s model.

Table-4. Unit Root Test Results

Variable	PANIC
	Constant
	Choi	MW
lnG	3.098* (0.001)	62.289* (0,004)
lnP	7.628* (0.000)	100.722* (0.000)
lnGDP	9.921* (0.000)	120.182* (0.000)
lnPOP	6.440* (0.000)	90.646* (0.000)
TEMP	1.944** (0.025)	52.496** (0.037)

Note: * and **  represent that series are stationary at the 1% and 5 % statistical significance level respectively Maximum number of factors assumed to be 2 in PANIC test.

While estimating with least square model, it is required to clarify the existence of constant and random effects in the model (Elmas et al., 2017 ). In this study, constant effects were tested by F test and random effects by LM test. In addition, Hausman (1978 ) test has been used in order to determine which one of these effects is significant. Results of F, LM and Hausman tests conducted for both models are presented in Table 5.

Table-5. F, LM and Hausman Test Results

Tests	Statistics	Probability
F Unit	1146.581	0.000*
F Time	1.242	0.236
FUnit-Time	462.99	0.000*
LMUnit	517.091	0.000*
LMTime	5.742	0.016*
LMUnit-Tüme	522.832	0.000*
Hausman	5.519	0.238

Note: *p<0.01

According to F test, it has been determined that constant unit and time effects was significant at 1% level (Table 4). According to LM test, it has been determined that random unit and time effects of the model is significant at %1 level (Table 5). Besides, the result of Hausman test shows that it would be more correct to use random effect model. Thus, model of the study has been estimated as one-way random effect. These results are presented in Table 6.

Table-6. Random Effects Regression Estimation Results

Variables	Coefficients	Standard Error	t- Statistics	Probability
lnP	-0.147	0.031	-4.668	0.000*
lnGDP	0.516	0.103	5.013	0.000*
lnPOP	1.103	0.138	7.993	0.000*
TEMP	-0.111	0.013	-8.563	0.000*
Constant	-14.088	2.412	-5.841	0.000*
	R2= 0.444    F= 42135*    F(Prob.)= 0.000

Note:* p < 0.01

Figure 1 shows the three-dimensional plot of the regression model and two-dimensional contour plot.  The highest lndg values are found where lngdp is highest. On the other hand, the lowest lndg values are found where lnp is highest. Also, as lngdp moves away from zero, lng values continuously increase, while as lnp moves away from zero, lng values decrease steadily. The surface plot shows how the lng variable is related to the lnp and lngdp variables according to equation 5. As a result, Figure 5 confirms the results of the econometric analysis.

Figure-1. Contour plot and surface plot of the regression

Source: Author created this figure based on the findings of the research.

According to the results of analysis; while lnP and TEMP affect lnG negatively, lnGDP and lnPOP effects lnG positively. But, as known generally, models are required not to have heteroscedasticity and autocorrelation problem so that these results estimated would not be misleading, namely, reliable. Therefore, LMh, LMp and LMp* tests were conducted in order to identify the existence of heteroscedasticity and autocorrelation in the models. Results of these tests are presented in the Table 7.

Table-7. Heteroscedasticity and Autocorrelation Tests Results

Tests	Statistics	Probability
LMh	52.874*	0
LMp	517.277*	0
LMp*	465.831*	0

According to result of LMh test, heteroscedasticity problem was determined at 1% significance level.  According to results of LMp and LMP tests, autocorrelation was determined at 1% significance level.  Thus, the models are required to be estimated again by resistant estimators. For this reason, models were estimated again by resistant estimator White Period resistant estimator which adjust both heteroscedasticity and autocorrelation problems. Table 8 displays the results of the White Period resistant estimator.

Table-8. Adjusted Random Effects Regression Estimation Results

Variables	Coefficients	Standard Error	t- Statistics	Probability
lnP	-0.146	0.051	-2.834	0.000*
lnGDP	0.515	0.256	2.013	0.005*
lnPOP	1.102	0.154	7.157	0.045**
TEMP	-0.111	0.011	-10.246	0.000*
Constant	-14.087	4.453	-3.163	0.001*
	R2= 0.444    F= 42.135*    F(Prob.)= 0.000

Note:* p < 0.01** and p < 0.05

In the analysis, coefficients of all factors affecting the residential natural gas demand were found to be statistically significant.
According to adjusted random effects regression estimation results; natural gas prices effects natural gas demand negatively at the 1% significance level. This result indicates the price elasticity of natural gas demand. Accordingly, in this period, price elasticity of residential of natural gas demand was predicted as -0.146. An increase of 1% of the natural gas prices rate decreases the natural gas demand approximately 0.15%. This finding overlaps with existing studies, such as Balestra and Nerlove (1966 ) for the United States, Yi (2000 ) for Swedish, Asche et al. (2008 ) for 12 European countries, Alberini et al. (2011 ) for  the United States, Andersen et al. (2011 ) for 13 OECD countries and Burke and Yang (2016 ) for a sample of 44 countries.

Per capita GDP effects natural gas demand positively at the 0.01 significance level. This finding indicates the income elasticity of demand. This finding result indicates that natural gas is a normal good. The income elasticity of household of natural gas demand was predicted as 0.515.  It can be said that an increase by %1 in the per capita GDP in the period of 2004-2015 in OECD countries leads to an increase by approximately 0.52% household natural gas consumption. This finding means that household demand for natural gas is less sensitive to per capita GDP. The findings about the relationship between natural gas demand and income are in line with the literature.  Eltony (1996 ) reports positive relationship between natural gas demand and income in six Golf Cooperation Council countries. Moreover, he finds that the income elasticity of natural gas is 0.48 in the long run.  Similarly, Catik and Deliktas (2016 ) provide similar evidence from the Turkey. They reported a long-run income elasticity of 0.748 using aggregated data. Similarly, Altinay and Yalta (2016 ) provide similar evidence from the Turkey. They report a long-run income elasticity of 0.512 using globally aggregated data, again similar to our estimates.

The population coefficient was found to be 1.103, which is positive and significant. It can be said that the long term population elasticity is near the unit elasticity. An increase by 1% in the population increases the household natural gas demand by 1.1% (p<0.05).

As expected, the climate influences negatively natural gas consumption. It has been observed that there is a negative relation between climate and natural gas demand at the 1% significance level. Because the average temperature is not converted into logarithms, the coefficient on average temperature is not elasticity.  Accordingly, one unit increase in the average temperature leads to a decrease of 11.1% in the household of natural gas consumption.  Namely, the results showed that for a 1 oC ambient temperature rise, the natural gas demand would decrease by 11.1%  oC in household sectors for OECD. Hence, this result means that relatively warmer countries consume less natural gas. This finding is consistent with Yu et al. (2014 ); Harold et al. (2015 ) and Altinay and Yalta (2016 ).

4. CONCLUSIONS

The aim of this study is to analyze the relationship between the household natural gas demand and price, income, population, and climate for the selected OECD countries for period 2004-2015. Given the predicted results of the household natural gas demand, all of the coefficients were determined to be statistically significant and their signs were found to be consistent with expectations of statistical economic theories.

Consequently of the analysis executed, it has found out that natural gas price significantly and negatively affected household natural gas demand. The real income per capita and population significantly and positively have affected household natural gas demand. Also, the average temperature significantly and negatively has affected household natural gas demand.

The price elasticity value of household natural gas demand was found to be smaller than one in for the selected OECD countries. Hence, considering the relationship between consumption and price elasticity, it can be suggested that an increase (or decrease) in the natural gas prices when price elasticity of the demand is less than one, demand in the household sector would decrease (or increase). As a result, it can be said that household natural gas demand in selected OECD countries is inelastic. This result shows that there are less substitution possibilities for natural gas in OECD household sector. This price elasticity value indicates that an increase in tax will decrease demand and therefore also intended tax revenue. Also, income elasticity natural gas of household demand was estimated to be between zero and one (0.515). Therefore, natural gas can be considered as a normal and compulsory goods in the household sector for OECD countries.

In the analysis, the most effective variable of household natural gas demand was found to be population. Taking into consideration the relationship between population and household natural gas demand, population is observed to be significantly increasing household natural gas demand. That is to say, population is found to have a strongly positive association with household natural gas demand in selected OECD countries.

Considering the relationship between climate and household natural gas demand, the average temperature is observed to be significantly decreasing household natural gas demand. According to this result, there is an adverse relationship between natural gas consumption and average temperature in household sector. Lower temperature values cause the natural gas to be used more for heating purposes. Also, households consume natural gas indirectly, via electricity generation. In this study; demand, price and revenue data are in the form of purchasing power parity. From this point of view, we believe that the estimated price and income elasticities of natural gas demand will be useful for energy policy makers.

Funding: This study received no specific financial support.

Competing Interests: The author declares that there are no conflicts of interests regarding the publication of this paper.

REFERENCES

Alberini, A., W. Gans and D. Velez-Lopez, 2011. Residential consumption of gas and electricity in the US: The role of prices and income. Energy Economics, 33(5): 870-881. Available at: https://doi.org/10.1016/j.eneco.2011.01.015.

Altinay, G. and A.T. Yalta, 2016. Estimating the evolution of elasticities of natural gas demand: The case of Istanbul, Turkey. Empirical Economics, 51(1): 201-220. Available at: https://doi.org/10.1007/s00181-015-1012-1.

Alves, D.C. and R.D.L. da Silveira Bueno, 2003. Short-run, long-run and cross elasticities of gasoline demand in Brazil. Energy Economics, 25(2): 191-199. Available at: https://doi.org/10.1016/s0140-9883(02)00108-1.

Andersen, T.B., O.B. Nilsen and R. Tveteras, 2011. How is demand for natural gas determined across European industrial sectors? Energy Policy, 39(9): 5499-5508. Available at: https://doi.org/10.1016/j.enpol.2011.05.012.

Apergis, N. and J.E. Payne, 2010. Natural gas consumption and economic growth: A panel investigation of 67 countries. Applied Energy, 87(8): 2759-2763. Available at: https://doi.org/10.1016/j.apenergy.2010.01.002.

Asche, F., O.B. Nilsen and R. Tveteras, 2008. Natural gas demand in the European household sector. The Energy Journal, 29(3): 27-46. Available at: https://doi.org/10.5547/issn0195-6574-ej-vol29-no3-2.

Bai, J. and S. Ng, 2004. A PANIC attack on unit roots and cointegration. Econometrica, 72(4): 191-221. Available at: https://doi.org/10.1111/j.1468-0262.2004.00528.x.

Baker, P. and R. Blundell, 1991. The microeconometric approach to modelling energy demand: Some results for UK households. Oxford Review of Economic Policy, 7(2): 54-76. Available at: https://doi.org/10.1093/oxrep/7.2.54.

Balestra, P. and M. Nerlove, 1966. Pooling cross section and time series data in the estimation of a dynamic model: The demand for natural gas. Econometrica, 34(3): 585–612. Available at: https://doi.org/10.2307/1909771.

Baltagi, B.H., 2005. A companion to econometric analysis of panel data. 3rd Edn., New York: John Wiley & Sons.

Bernstein, M.A. and J. Griffin, 2006. Regional differences in price-elasticity of demand for  energy. The Rand Corporation Technical Report.

Bernstein, R. and R. Madlener, 2011. Residential natural gas demand elasticities in OECD countries: An ARDL bounds testing approach. FCN Working Paper No. 15/2011.

Bilgili, F., 2014. Long run elasticities of demand for natural gas: OECD panel data evidence. Energy Sources, Part B: Economics, Planning, and Policy, 9(4): 334-341. Available at: https://doi.org/10.1080/15567249.2010.497793.

BP, 2017. Statistical review of world energy.

Brenton, P., 1997. Estimates of the demand for energy using cross-country consumption data. Applied Economics, 29(7): 851-859. Available at: https://doi.org/10.1080/000368497326507.

Breusch, T.S. and A.R. Pagan, 1980. The lagrange multiplier test and its applications to model specification in econometrics. The Review of Economic Studies, 47(1): 239-253. Available at: https://doi.org/10.2307/2297111.

Brounen, D., N. Kok and J.M. Quigley, 2012. Residential energy use and conservation: Economics and demographics. European Economic Review, 56(5): 931–945. Available at: https://doi.org/10.1016/j.euroecorev.2012.02.007.

Burke, P.J. and H. Yang, 2016. The price and income elasticities of natural gas demand: International evidence. Energy Economics, 59(C): 466-474. Available at: https://doi.org/10.1016/j.eneco.2016.08.025.

Catik, A.N. and E. Deliktas, 2016. Estimation of price and income elasticities of demand for oil, coal and natural gas in Turkey. Pamukkale University Journal of Social Sciences Institute, 25(50): 1-20. Available at: https://doi.org/10.5505/pausbed.2016.81905.

Cetin, T. and F.h. Yüksel, 2014. Empirical investigation on energy dependence-consumption nexus: Evidence from Turkish natural gas market. Applied Energy, 133(C): 243-251. Available at: https://doi.org/10.1016/j.apenergy.2014.07.062.

Dahl, C.A., 2012. Measuring global gasoline and diesel price and income elasticities. Energy Policy, 41(C): 2-13. Available at: https://doi.org/10.1016/j.enpol.2010.11.055.

Elmas, B., M. Polat and M. Baydaş, 2017. Innovation and company performance relationship: An investigation with panel data analysis in BIST stone and soil-based sector. The Macrotheme Review, 6(2): 33-48.

Eltony, M.N., 1996. Demand for gasoline in the GCC: An application of pooling and testing procedures. Energy Economics, 18(3): 203-209. Available at: https://doi.org/10.1016/0140-9883(96)00011-4.

Erdogdu, E., 2010. Natural gas demand in Turkey. Applied Energy, 87(1): 211-219. Available at: https://doi.org/10.1016/j.apenergy.2009.07.006.

Fung, A.S.-L., M. Aydinalp and V.I. Ugursal, 1999. Econometric models for major residential energy end-uses. CREEDAC-1999-04-05. Halifax, Nova Scotia, Canada: Canadian Residential Energy End-use Data and Analysis Centre.

Harold, J., S. Lyons and J. Cullinan, 2015. The determinants of residential gas demand in Ireland. Energy Economics, 51: 475–483. Available at: https://doi.org/10.1016/j.eneco.2015.08.015.

Hausman, J.A., 1978. Specification tests in econometrics. Econometrica, 46(6): 1251-1271. Available at: https://doi.org/10.2307/1913827.

IEA, 2016a. Natutal gas information statistics. Paris, France: International Energy Agency.

IEA, 2016b. Energy prices and taxes statistics. Paris, France: International Energy Agency.

International Energy Agency, 2017. Natural gas information statistics. France: International Energy Agency, Paris.

Kani, A.H., M. Abbasspour and Z. Abedi, 2014. Estimation of demand function for natural gas in Iran: Evidences based on smooth transition regression models. Economic Modelling, 36(C): 341-347. Available at: https://doi.org/10.1016/j.econmod.2013.10.007.

Leth-Petersen, S., 2002. Micro econometric modelling of household energy use: Testing for dependence between demand for electricity and natural gas. The Energy Journal, 23(4): 57-84. Available at: https://doi.org/10.5547/issn0195-6574-ej-vol23-no4-3.

Madlener, R., R. Bernstein and M. Gonzlez, 2011. Econometric estimation of energy demand elasticities, E. ON Energy Research Center Series, E. ON Energy Research Center, RWTH Aachen, 3(8).

Nilsen, O.B., F. Asche and R. Tveteras, 2005. Natural gas demand in the European household sector, Working Paper No: 44/05. Bergen, Norway: Institute for Research in Economics and Business Administration.

OECD, 2012. Eurostat-OECD methodological manual on purchasing power parities. Luxembourg: Publications Office of the European Union.

Park, S.Y. and G. Zhao, 2010. An estimation of US gasoline demand: A smooth time-varying cointegration approach. Energy Economics, 32(1): 110-120. Available at: https://doi.org/10.1016/j.eneco.2009.10.002.

Payne, J.E., D. Loomis and R. Wilson, 2011. Residential natural gas demand in Illinois: Evidence from the ARDL bounds testing approach. Journal of Regional Analysis & Policy, 41(2): 138. Available at: https://doi.org/10.1007/s00181-010-0381-8.

Pesaran, M.H., 2004. General diagnostic tests for cross section dependence in panels (No. 1240). Institute for the Study of Labor (IZA).

Pesaran, M.H. and R. Smith, 1995. Estimating long-run relationships from dynamic heterogeneous panels. Journal of Econometrics, 68(1): 79-113. Available at: https://doi.org/10.1016/0304-4076(94)01644-f.

Pesaran, M.H., A. Ullah and T. Yamagata, 2008. A bias-adjusted LM test of error cross-section independence. Econometrics Journal, 11(1): 105–127. Available at: https://doi.org/10.1111/j.1368-423X.2007.00227.x.

Shahbaz, M., H.H. Lean and A. Farooq, 2013. Natural gas consumption and economic growth in Pakistan. Renewable and Sustainable Energy Reviews, 18(C): 87-94. Available at: https://doi.org/10.1016/j.rser.2012.09.029.

Speight, J.G., 2017. Rules of thumb for petroleum engineers. John Wiley & Sons.

Tatoglu, F.Y., 2013. Panel data econometrics. Istanbul: Beta Publications.

Yi, F., 2000. Dynamic energy-demand models: A comparison. Energy Economics, 22(2): 285-297. Available at: https://doi.org/10.1016/s0140-9883(99)00042-0.

Yoo, S.-H., H.-J. Lim and S.-J. Kwak, 2009. Estimating the residential demand function for natural gas in Seoul with correction for sample selection bias. Applied Energy, 86(4): 460-465. Available at: https://doi.org/10.1016/j.apenergy.2008.08.023.

Yorucu, V. and P. Bahramian, 2015. Price modelling of natural gas for the EU-12 countries: Evidence from panel cointegration. Journal of Natural Gas Science and Engineering, 24: 464-472. Available at: https://doi.org/10.1016/j.jngse.2015.04.006.

Yu, Y., X. Zheng and Y. Han, 2014. On the demand for natural gas in urban China. Energy Policy, 70(C): 57-63. Available at: https://doi.org/10.1016/j.enpol.2014.03.032.

---