The Relationship between Electricity Consumption and Economic Growth: Evidence from Azerbaijan

Research examines the relations between GDP in Manat and Dollar and total electric energy consumption (1995-2017) for the last 22 years in the Republic of Azerbaijan. Besides, the relations between the electric energy consumption and the growth of GDP in these sectors were analysed. Autoregressive distributed lag model was used as a research methodology. Stationary tests of variables (ADF, PP, and KPSS) and Pairwise Granger Causality Tests were done. Stability of models was examined. Eviews_9 econometric software program was used to establish graphics and do calculations. Having analysed the research, there is a positive correlation not only in GDP and electric energy consumption but also electric energy consumption and GDP in different sectors of economy. We recommend to save electric energy. result confirms that there is a relationship among electricity consumption, economic growth, population and electricity price Toda-Yamamoto causality test, VAR The results of this test show that there is bidirectional causality between energy consumption and economic growth. Findings of the study Gregory–Hansen test, VECM The results confirm the existence of a long-run relationship among the variables (between energy consumption, financial development, and economic growth). Find that there is a positive and statistically significant impact of financial development and economic growth on energy consumption in the long-run


INTRODUCTION
The roles that hydrocarbon resources, including oil products as energy carriers, play in the economy and in the life of people are undeniable (Muradov et al., 2019). Policy development, ensuring the growth and development of complex, open and non-linear economic systems, as well as the measurement and evaluation of its results are widely discussed topics in modern economics. It is not by chance that these topics occupy a special place in the reports of international organizations and in the diaries of scientific journals. "How are the priority directions of economic policy chosen?," "How are the needs of economic agents studied in incentives?" "How does economic policy affect the behavior of economic agents?," "What indicators can assess the effectiveness of regulatory measures?", "How can one measure and evaluate economic development and growth?" These and many other issues still remain "apples of discord" for economists.
Angus Deaton believes that economics can be a good tool for developing successful economic policies, but for this it needs, first of all, high-quality information. Nobel laureate, noting the fundamental importance of measurement in economics, argued that, as a means of correctly evaluating the results of economic policy, measurement could also become a source for new theoretical ideas.
Currently, in countries along with microeconomic statistics, microeconomic databases are also being developed. The creation, This Journal is licensed under a Creative Commons Attribution 4.0 International License preservation, systematization and processing of such scientific and labor-intensive microeconomic databases requires large financial resources. And the quality of macroeconomic data is reduced under the influence of the variability of monetary units of measurement. To eliminate the above deficiency, universal natural indicators are used, one of which is the electricity consumption.
Other studies show us that the nature of the relationship between electricity consumption and economic growth differs depending on the type of activity. There exists bidirectional causality between electricity consumption and real output in the services sector and unidirectional causality running from real output in the industrial sector to electricity consumption. However, there is no causal relationship between electricity consumption and real output in the agricultural sector (Ibrahiem, 2018).
In addition, it is believed that the living standard of the population and the economic development degree in the country also affect the relationship between electricity consumption and economic growth (Mahfoudh and Ben Amar, 2014). The results of these studies do not give us full grounds for adopting electricity consumption as an unequivocal indicator of economic growth.
Preponderance of industrial states is completely dependent on energy to fuel their economies. Besides, globalization has made the world to be so interconnected and interdependent that the energy industry is the biggest contributor of the climate change which doesn't affect a single country but have far wider implications (Vidadili et al., 2017).
Each country with its own economic structure, development level and security of natural energy resources, mechanisms for regulating energy markets, climatic, geographical and demographic conditions is a unique object of research. Conducting such studies in different countries can clarify the relationships between factors that influence the nature of the relationship between electricity consumption and economic growth. On this basis, the study of the relations between these indicators in Azerbaijan was adopted as the goal of this study.
We think that the results of our research are of scientific and practical importance in the following areas: 1. Research provides some empirical information about the relationship between energy consumption and economic growth in an energy-rich country; 2. Shows the behavior of energy consumption and growth in different sectors of the Azerbaijan economy; 3. The results can be used to predict the electricity demand throughout the economy, and by its branches in countries with similar conditions.

GENERAL CONDITIONS OF POWER SUPPLY IN AZERBAIJAN
Electricity production in Azerbaijan dates back to the first years of the 20 th century -from the time of the first oil boom. Subsequent industrialization processes and the possession of rich hydrocarbon resources led to an accelerated growth in demand for electricity. In the last 50 years alone, the production of electricity has almost doubled (Table 1).
İt can be seen that almost 90% of the total energy produced is accounted for by thermal power plants. Despite the country's huge potential for renewable and alternative energy sources, it was only in the last decade that solar and wind energy began to be used for production.
In Azerbaijan, production facilities, transmission lines and distribution of electricity are fully owned by the state and electricity tariffs are regulated by the state. The volume of production capacity is far ahead of the volume of domestic demand for electricity. Today, Azerbaijan is an exporter of electric energy to neighboring countries.
Despite the fact that all parts of Azerbaijan are fully and continuously supplied with electricity, we occupy one of the last places in the region in terms of the use of electricity per capita (Table 2).  1913 1920 1930 1940 1950 1960 1970 1980 1990  2000  2010  2017  Generation  110,8 122 503,9 1827 2924 6590 12027 15045 23152 18699 18710 24320,9  Fuel  1802 2894 4626 10893 13825 21399 17069 15003 20445,4  Hydro  24,3  29,5 1963 1022  1098  1658  1534   There is a strong correlation between electricity consumption and rich countries. There is also a correlation relationship between the lack of using modern service of energy and the people who work for 2 dollars per day Ömer and Bayrak (2017) 1990-2012 75 net energy-importing countries CADF

DOLS and FMOLS estimators
Based on the results of individual and group states, there is a positive and statistically significant relationship between electricity consumption and GDP in a long term. Thus, electricity consumption causes GDP growth Uyar and Gökçe (2017) Annual, 1985-2013 Vietnam, Indonesia, South Africa, Turkey and Argentina

Panel Quantile Regression
The influence of GDP on oil consumption is tremendously downsizing. However, hydroelectric stations impact on electricity consumption positively and surges significantly. Coal has no any influence on economic growth Kamal (2017) 2000-2011 Five south Asian countries Granger causality, VARM Cointegration test proves the positive relationship and balance between electricity consumption and GDP in a long term. The electricity consumption coefficient is 1.3%. It reveals that the increase of electricity consumption by 1% causes the growth of GDP by 1.31%. Thus, electricity consumption has a significant influence on economic development in South Asia Aslan (2014) 1980-2008 Turkey Granger causality There is a positive and statistically significant relationship between electricity consumption and GDP in Turkey Ibrahiem (2018) 1971-2013 Egypt Johansen cointegration approach, VECM Results prove the following relationships: 1. There is a double causal relationship between electricity consumption and real product in service sector 2. There is a single-direction causal relationship from real product to electricity consumption 3. There is no causal relationship between real product and electricity consumption in agrarian sector The research result confirms the cointegartion relationship between electricity consumption and economic growth and sets the double-direction causal relationship between electricity consumption and economic growth Ranjan et al. (2017) 1990-2012 BRICS countries The Pedroni (1999)(2000)(2001)(2002)(2003)(2004) Panel cointegration test, PECM There is no any strong relationship between GDP and electricity consumption. The growth of GDP is a key factor that causes the increase of electricity consumption in the studied states

1971-2014
Malaysia ARDL There is a cointegration relations between real GDP and electricity consumption. Electricity consumption influences positively on economic growth in a short term Lira and Mamofokeng (2016)

1982-2013
Uganda Granger causality test The result confirms the double-direction causal relationship between electricity consumption and economic growth in a long term Ozturk et al. (2019) 1970-2012 Denmark ARDL Granger causality test The result confirms the neutrality of the relationship between electricity consumption and economic growth in Denmark Bekareva et al. (2017Bekareva et al. ( ) 2000Bekareva et al. ( -2014 United States

Arellano-Bond method
The result confirms the positive relationship between renewable energy consumption and economic growth. Molem and Ndifor (2016)

1980-2014 Cameroon Generalised Method of Moments
The result confirms that there is a relationship among electricity consumption, economic growth, population and electricity price Mukhtarov et al., 2017Mukhtarov et al., 1990Mukhtarov et al., -2015 Azerbaijan Toda-Yamamoto causality test, VAR The results of this test show that there is bidirectional causality between energy consumption and economic growth. Findings of the study Mukhtarov et al., 2018Mukhtarov et al., 1992Mukhtarov et al., -2015 Azerbaijan Gregory-Hansen test, VECM The results confirm the existence of a long-run relationship among the variables (between energy consumption, financial development, and economic growth). Find that there is a positive and statistically significant impact of financial development and economic growth on energy consumption in the long-run The availability and relatively low tariffs of natural gas in all regions of Azerbaijan are the main argument for explaining this paradoxical situation.

LITERATURE REVIEW
The research of the relationships between energy consumption and economic growth has been a focal issue among scientists (Table 3). During research, a number of methods were employed. We can classify them as the following: for example, the relationships between energy consumption and economic growth has been analysed through autoregressive distributed lag (ARDL) method (Lefteris and Theologos, 2011;Ozturk and Ali, 2011;Ramazan et al., 2008;Nicholas, 2009;Fuinhas et al., 2012). However, other scientists researched the relationships between energy consumption and economic growth by Granger test (

ARDL Model
ARDL model was used for the research. Through this model, cointegration between electric energy and GDP was estimated. To be exact, research assessed the influence of total electric energy production to GDP and the impact of electric energy consumption in different fields to GDP in Azerbaijan Republic (A. Figure 1). The relations in long and short term were researched.

Unit Root Tests
It is essential to check the stationary of variables through Unit Root before the assessment of regression equations. Because, keeping stability between variables is important while assessing the dependency between two or more variables by using regression analysis. However, probability distribution for every time series in order to be stationary must be identical. Nevertheless, stationary of variables is not always desirable. For a long term or cointegration relation and assessment, the variables must be non-stationary in most methods. It is also required that the first difference should be stationary or I(1). It must be noted that if any time series variable is stationary with real values, then it can be considered I(0). If a variable is not I(0), then its first difference is calculated and its stationary is checked. In this case, if the variable is stationary, then it is considered I(1). A variable sometimes changes because of probability distribution. In that case, the variable becomes trendstationary. One can refer to modern econometric books regarding the stationary of changes and its effect in time series analysis (Hill et al., 2001;Heij et al., 2005;Asteriou and Hall, 2007). We can analyze them by applying three different unit root tests in order to get more reliable stationary test results: Augmented Dickey Fuller, Phillips−Perron (PP) and Kwiatkowski−Phillips−Schmidt−Shin (KPSS). The evaluation of these tests is done through E-Views 9. It must be noted that "unit root problem" or "variable is nonstationary" null hypothesis in unit root tests is checked. In KPSS test, "variable is stationary" hypothesis is taken and considered as stationary null hypothesis. If the variable is non-stationary without trend, and becomes stationary if trend is included, then the checked variable is considered "trend-stationary".

Test Cointegration
Cointegration test proves long-term relations and F−statistics is indicated to express it. Menatime, cointegration was identified by ECM model. In these models, GDP is dependent variable while electric energy consumption is an independent variable.

Diagnostic Test
This article will use Breusch Godfrey LM test (null hypothesis: "no serial correlation") in order to check subsequent correlation problem and use both Breusch−Pagan−Godfrey (null hypothesis: "no heteroskedasticity problem") and Autoregressive Conditional Hederoscedasticity test (ARCH) for obtaining more reliable outcomes for heteroskedasticity problem. During ARCH test, null hypothesis "no heteroskedasticity problem" theory is checked. Nonetheless, Ramsey RESET Test and Normality Test (Jarque-Bera) JB was checked. Null hypothesis rejection is acceptable for every five cases.
Statistical data encompasses 1995-2017. Data have been taken from Statistics Committee of the Republic of Azerbaijan.

Unit Root Test
Let's have a look at stationary of variables before identifying methods for evaluation. All stationary test results of variables for evaluation of both problems were given in the 0.005 ect (t-1) −0.03 ∆lmigdp (t-1) 0.23 ∆lecmii (t-1) 0.005 ect (t-1) −0.08 ∆lcgdp (t-1) 0.23 ∆liecc (t-1) 0.23  Table 3). It means that all above-mentioned methods are applicable. As mentioned above, during application process of ARDL cointegration method, one of the important issues while establishing a model is to identify optimum lag length. At this time, the most important factor is to eliminate the subsequent correlation problem in selected optimum model and keep the minimum of SBC information criteria value.

VAR Lag Order Selection Criteria
In order to determine optimal lag for ARDL model, VAR Lag Order Selection Criteria was employed and we got the below-mentioned results (Table 4). Table 5 illustrates whether cointegration relations between variables exist or not. Thus, there are cointegration relations among electric energy consumption per year (ECI) and GDP in manat (GDPM) and in dollar (GDPD), electric energy consumption of electric energy production entities (EPEITEC) and their GDP (EPEGDP), electric energy consumption in construction (IECC) and its GDP (CGDP), electric energy consumption in agriculture, hunting and forestry (ECAHFI) and their GDP (AHFGDP), electric energy consumption in other, commercial and public service entities (ECCPSI) and their GDP (CPSGDP) and electric energy consumption of people (ECPI) and People's income (PI). In other words, there are long-term relations. F-statistics factors are above the minimum indicators of 5% according to Narayan (2005)  However, there are no cointegration relations among electric energy consumption in industry (ECII) and its GDP (IGDP), electric energy consumption in mining (ECMII) and its GDP (MIGDP) and electric energy consumption in telecommunication, transport and warehouse (ECTWTI) and their GDP (TWTGDP).

Cointeq = lN -a × lM + c
According to the table, electric energy consumption causes the increase of GDP ( In general, there are valuable from economic standpoint. Except the equations refer to relations among the energy consumption in industry (ECII) and GDP and the energy consumption in transport, warehouse and telecommunication (ECTWTI) and their GDP. The main reason for this is other factors that play key roles in the augmentation of GDP.
Referring to A.Tables 4 and 5, we can mention that coefficients are 5% 1% and 0.1% significant.

Error Correction (Short Run) Model
This table reveals the results of short-term and ECM model. The results are in the following: There is a positive relation between GDP and electric energy consumption in all models. GDPD coefficient is significant at the level of 1% in correlation model between GDPD and total electric energy consumption (ECI). (model 1). Besides, 0.09* 0.13 0.12* 0.08* (EPEGDP) coefficient is significant at the level of 5% in the model between energy consumption in electric energy producing entities (EPEITEC) and their GDP (model 3). On the other hand, ect coefficient is negative (−) for all. According to the models, velocity to balance in a long term is 4% (model 1), 6% (model 2), 3% (model 3), 3% (model4), 8% (model 5), 16% (model 6), 3% (model 7), 2% (model 8), 4% (model 9), 9% (model 10) (Tables 7 and 7a). Although ect coefficients are insignificance in these models, their negativity substantiates the existence of cointegration relations proposed by Paseran and others (2001). Having positive relation in these models shows the role of electric energy and its consumption in the increase of GDP for new economic growth.
Some models for ARDL models (model 1-3 and 6) are 5% 1% and 0.1% significant. Regression equations are adequate. It also passes all the diagnostic tests against serial correlation (Durbin Watson test and Breusch-Godfrey test), heteroscedasticity (White Heteroskedasticity Test), and normality of errors (Jarque-Bera test). The Ramsey RESET test also suggests that the model is well specified. All the results of these tests are shown in Table 8 and 8a. The stability of the longrun coefficient is tested by the short-run dynamics. Once the ECM model given by equations (Table 7 and 7a) has been estimated, the cumulative sum of recursive residuals (CUSUM) and the CUSUM of square (CUSUMSQ) tests are applied to assess the parameter stability (Pesaran and Pesaran (1997). A. Figure 2 plot the results for CUSUM and CUSUMSQ tests. The results indicate the absence of any instability of the coefficients because the plot of the CUSUM and CUSUMSQ statistic fall inside the critical bands of the 5% confidence interval of parameter stability However, non-stability in model 2 and model 3 was observed (A. Figure 2).

CONCLUSION
Energy and especially electricity is one of the main factors of the development of society. From this perspective, energy and electricity consumption is essential in Azerbaijan too. We have achieved some results from the research that electricity consumption plays an important role in economic growth. Electricity is also important as an economic resource. Although the relationship between electricity consumption and GDP growth is not strong, we can mention the followings: there is a positive dependency between total electricity consumption and GDP in manat and dollar, as well as electricity consumption in electricity producing entities such as mining, construction, agriculture, hunting and forestry, commercial and public service and others and GDP in those sectors. Conversely, there is a negative dependency between electricity consumption in industry, transportation, warehouse and telecommunication and GDP. On the contrary, the opposite dependency was observed between electricity consumption of population and people's income. The positive income was also observed between electricity consumption and GDP according to ECM model results in a short term. Having a positive relationship in models shows that electricity consumption plays an important role in GDP growth.
So, the analysis has revealed that there is a weak relationships betwen either the electricity consumption and GDP in the Republic or in different sectors of economy. From this perspective, we recommend not to waste electricity consumption. ADF denotes the Augmented Dickey-Fuller single root system respectively. The maximum lag order is 3. The optimum lag order is selected based on the Shwarz criterion automatically; ***, ** and *indicate rejection of the null hypotheses at the 1%, 5% and 10% significance levels respectively. The critical values are taken from MacKinnon (Mackinnon, 1996). Assessment period: 1995-2017. S: Stationarity; N/S: No stationarity A. PP Phillips-Perron is single root system. The optimum lag order in PP test is selected based on the Newey-West criterion automatically; ***, ** and *indicate rejection of the null hypotheses at the 1%, 5% and 10% significance levels respectively. The critical values are taken from MacKinnon (Mackinnon, 1996). Assessment period: 1995-2017. S: Stationarity, N/S: No Stationarity A.  (Kwiatkowski et al.,1992) single root system. The optimum lag order in KPSS test is selected based on the Newey-West criterion automatically; ***, ** and *indicate rejection ofthe null hypotheses at the 1%, 5% and 10% significance levels respectively. The critical values are taken from Kwiatkowski-Phillips-Schmidt-Shin. Assessment period: 1995-2017. S: Stationarity, N/S: No Stationarity A.