Aggregate and Disaggregate Energy Consumption Relation with GDP: Evidence for Iran

Introduction

Islamic republic of Iran is one of the major energy rich countries in the Middle East and in the world. According to British Petroleum Statistical Review of World Energy 2014,Iran has about 157 billion barrels of oil (proved reserves) almost 9.3% world's total oil reserves as well as 1192.9 trillion cubic feet of gas (almost 18.2% of total world gas). By counting oil and gas reserves Iran holds about 27.5% of world's total energy and even can be the number one country of the world in hydrocarbon reserves. Iran is also one of the major energy consumer countries with respect to its population. This is one of the most important challenges of Iran's economy specifically in recent decades. Energy export income (oil and gas) forms country's major export income and every year major part of government budget is financed by this income, for this reason Iran's economy is vulnerable to energy shocks. Energy intensity in Iran is high in comparison with other oil exporting countries such as OPEC member countries; Fig (2) shows this issue. In recent decades, low prices for energy, population growth and growth of industrialization and urbanization has increased energy consumption tremendously. Fig (1) shows total energy consumption versus GDP in Iran. However the government in recent years wants to decrease energy consumption subsidies in order to reform energy consumption situation. Iran is planning to build nuclear power plants in order to its future needs. The above mentioned points highlight the importance of this study. The main goal of this paper is to examine the relationship of total energy consumption and its individual forms with GDP in Iran. This relationship has been tremendously a subject of ongoing debate in the energy economics literature .In this context we confront with four different hypotheses: Neutrality hypothesis, conservation hypothesis, growth hypothesis and feedback hypothesis. Neutrality hypothesis indicates no causality between energy consumption and economic growth in any directions. Conservation hypothesis indicates a unidirectional causality running from economic growth to energy consumption. Growth hypothesis indicates a unidirectional causality running from energy consumption to economic growth. Feedback hypothesis indicates a bi-directional causality between economic growth and energy consumption. Each hypothesis has its own importance in policy implications (Wei Zhang & Shuyun Yang, 2013).

If, for example, there exists unidirectional Granger causality running from income to energy, it may be implied that energy.

Fig. 1: Total Energy Consumption versus GDP in Iran in Logarithmic Form

Fig. 2: Energy Intensity - Total Primary Energy Consumption per Dollar of GDP (Btu per Year 2005 U.S. Dollars)

Conservation policies may be implemented with little adverse or no effects on economic growth. If unidirectional causality runs from energy consumption to income, reducing energy consumption could lead to a fall in income or employment. The finding of no causality in either direction, the so-called ‘neutrality hypothesis’ would imply that energy conservation policies do not affect economic growth (Asafu-Adjaye, 2000). This paper's contribution is twofold in comparison with past literature, investigation of individual forms of energy consumption and employing Hsiao’s (1981) methodology. The remainder of this paper is organized as follows. Section2 presents a review of literature.Section3 introduces the data and methodology.Section4 presents empirical findings and their analysis and the final section adds up the results.

2. Review of Literature

In this section we review some of the past studies about the relationship of energy consumption and GDP. Many studies have been done about causality relation between energy consumption and GDP in the energy economics literature. We can divide these studies for energy dependent and energy exporting countries or in another view we can divide them for developed and developing economies. The results of some of these studies are conflicting because of methodological and data differences.

Specifically in recent decade because of non-stationary variables researchers have employed cointegration and error correction models (ECM) (Mahmoud A. Al-Iriani , 2006; Asafu-Adjaye, 2000; Lise & Montfort, 2007; Yoo, 2005; Glasure, 2002) to get rid of spurious regression or Toda and Yamamoto (1995) methodology (Hondroyianais et al, 2002 ; Fatai et al, 2002; Wolde-Rufael, 2004; Lee, 2006; Ziramba, 2009; Bowden & Payne, 2009; Payne, 2009; Soytas & Sari, 2009; Wolde-Rufael, 2009; Tsani, 2010; Wolde-Rufael & Menyah, 2010) .In studies for a group of countries many researchers have used panel cointegration methods (Narayan et al, 2010; Lee, 2005; Soytas et al, 2001; Soytas & Sari, 2003).

In this section we focus on Asian and developing countries because of Iran's economic situation. One of the major studies about this issue goes back to the Kraft and Kraft (1978) study about the USA economy, in their analysis, the authors used data on gross energy inputs and GNP for the USA and found that causality runs from GNP to energy consumption. Cheng (1995) found a unidirectional causality running from economic growth to energy consumption in India. Masih & Masih (1996) used cointegration analysis to study this relationship in a group of six Asian countries and found cointegration between energy use and GDP in India, Pakistan, and Indonesia. No cointegration is found in the case of Malaysia, Singapore and the Philippines. The flow of causality is found to be running from energy to GDP in India and from GDP to energy in Pakistan and Indonesia. Glasure & Lee (1997) examined the causality issue between GDP and energy consumption for South Korea and Singapore. In their results, granger causality tests showed no causal relationship between energy and GDP for South Korea, but a causal relationship from energy to GDP for Singapore. Asafu-Adjaye (2000) investigated causal relationships between energy consumption and income for India, Indonesia, the Philippines and Thailand by cointegration methods. The results reveal that in the short-run, unidirectional Granger causality runs from energy to income for India and Indonesia, while bidirectional Granger causality runs from energy to income for Thailand and the Philippines. In the case of Thailand and the Philippines, energy, income and prices are mutually causal. The study results do not support the view that energy and income are neutral with respect to each other, with the exception of Indonesia and India.

Aqeel & Butt (2001) investigated the causal relationship between energy consumption and economic growth as well as between energy consumption and employment for Pakistan. Cointegration and Granger causality tests inferred that economic growth causes total energy consumption. Further independent investigation revealed unidirectional causality from economic growth to petroleum consumption and showed no causality between economic growth and gas consumption, but it did indicate unidirectional causality from electricity consumption to economic growth. Lee (2005) re-investigated the co-movement and the causality relationship between energy consumption and GDP in 18 developing countries. The empirical results provided clear support for a long-run cointegration relationship after allowing for the heterogeneous country effect. The evidence showed that there were long-run and short-run causalities running from energy consumption to GDP, but not vice versa. This result indicated that energy conservation might harm economic growth in developing countries regardless of being transitory or permanent. Mahmoud A. Al-Iriani (2006) investigated the causality relationship between GDP and energy consumption in the six countries of the Gulf Cooperation Council (GCC). Empirical results indicate a unidirectional causality running from GDP to energy consumption. Evidence shows no support for the hypothesis that energy consumption is the source of GDP growth in the GCC countries. Mehrara (2007) examined the causal relationship between the per capita energy consumption (PCEC) and the per capita GDP in a panel of 11 oil-exporting countries (Iran, Kuwait, Saudi Arabia, United Arab Emirates, Bahrain, Oman, Algeria, Nigeria, Mexico, Venezuela and Ecuador) by using panel unit root tests and panel cointegration analysis. The results showed a unidirectional strong causality from economic growth to energy consumption for the oil-exporting countries. Mozumder & Marathe (2007) examined the causal relationship between the per capita electricity consumption and the per capita GDP for Bangladesh using a cointegration and vector error correction model. Their results showed that there was unidirectional causality running from per capita GDP to per capita electricity consumption. Reynolds & Kolodziej (2008) examined the relationship between GDP and production of some energy sources for the former Soviet Union by Granger causality and found unidirectional causality from energy consumption to GDP and unidirectional causality from GDP to coal production and natural gas. Behmiri & Manso (2012) analyzed the relationship between oil consumption and economic growth for OECD countries and found that there is bi-directional causality between them.

3. Data and Methodology

Based on the type and goal of research researchers have used different approaches to investigate the energy consumption and real GDP linkage. In this paper we investigate total energy consumption (TEC, hereafter) and individual forms of energy including oil (OIL, hereafter), natural gas (NG, hereafter), coal (COAL, hereafter), renewable energies (REE, hereafter) and electricity (ELEC, hereafter) causality relation with GDP in Iran. The data were obtained from central bank of Iran time series database and energy balance sheet of Iran, from ministry of energy. All data are annual and cover the period 1974-2010. TEC and individual energy forms are in million barrels of oil equivalent and GDP is in constant price (billion rials).All variables are in natural logarithm form. In the first step we must identify integration degree of the variables. To this end we employ KPSS¹ unit root test. Table1 shows the results. As we can see all of the variables are stationary at 5% and 1% significance level. We included a constant and a time trend for variables except LnREE which trend was not statistically significance, but for other variables time trend and constant were statistically significance. The results can provide the information about the long-run causal relationship among the variables in the model because all of the variables are in level.

*.C.V indicates critical value (5% and 1% significance level)

**.C indicates an intercept and T indicates a time trend

***. The Kwiatkowski, Phillips, Schmidt, and Shin (1992)

Thanks to the stationary variables we can use conventional granger causality test. The variable X is said not to granger cause the variable Y if all the coefficients of lagged X are significantly zero (H0) because it implies that the history of X doesn't improve the prediction of Y and conversely the variable Y is said not to granger cause the variable X if all the coefficients of lagged Y are significantly zero (H0) .Equ.(1) and Equ.(2)  show this issue respectively.

Y_t= Y_t-j +   X_t-i +        H₀: Σ  H₁:             (1)           

X_t= X_t-i + Y_t-j +          H₀:  H₁: Σ          (2)          

One of the most important issues in this approach is identification of optimal lag length, because this method is very sensitive to lag length. For this reason we use Hsiao’s (1981) methodology, which combines the Akaike (1969) final prediction error (FPE) criterion with granger’s causality test to guide the selection of the appropriate lag specifications. This method consists from two steps: in the first step we run an equation for a variable with its past lags. In this step we discern appropriate number of lags based on minimum Final Prediction Error (FPE).For example in Equ.(3) we identify optimal number of lags for LnGDP based on FPE formula in this equation. In this formula T is total number of observations, m is number of lags for LnGDP and SSR (m, 0) is sum of squared residuals only with m lags. In the second step we add the other variable to the equation(for example LnTEC), and discern appropriate number of lags for the new variable(n) with identified lags of the first variable(m) and based on minimum FPE. In this step we calculate the FPE with new formula. Equ.(4) shows this formula, in this formula n is number of lags for new added variable, and also SSR(m,n) is sum of squared residuals based on m and n lags. Now we can judge about causality. If calculated FPE in the second equation or in the second step be smaller than first equation's FPE we can say there is a causality relation which runs from new added variable(for example LnTEC) to the first variable (for example LnGDP).We employ the same procedure for other variables.

LnY_t= λ+ LnY_t-j +    , FPE (m,0)= ( )( )                  (3)                    

LnY_t= µ+ LnY_t-j +   LnX_t-i +                                           (4)

FPE(m,n)= ( )( )                                                      

LnX_t= θ+ LnX_t-i +                                                                      (5)

LnX_t= δ+ LnX_t-i +   LnY_t-j +                                         (6)

4. Empirical Results

Table2 shows the results for LnGDP and LnTEC. According to this table second equation's FPE is smaller than first equation's FPE (0.003857 < 0.004036). This indicates a causality relation from total energy consumption to GDP (LnTEC  LnGDP).And also final prediction error of fourth equation is smaller than third (0.001869 < 0.0020844) and indicates a causality relation from LnGDP to LnTEC.

(LnGDP  LnTEC). According to the results of table2 we have bidirectional causality relation between total energy consumption and GDP.This issue supports the feedback hypothesis about Iran and indicates a high level of economic growth leads to high level of energy demand and vice versa. This result has conflict with Mohsen Mehrara (2007) findings about three oil exporting countries(Iran, Kuwait and Saudi Arabia).His findings show that there is a unidirectional long-run causality running from economic growth to energy consumption and energy conservation is a feasible policy with no damaging repercussions on economic growth for Iran. The difference of the results with some of previous studies may be attributed to the new method proposed by Hsiao (1981) and period of study.

Table2: Total Energy Consumption and GDP Causality Results

Table3 shows the results for other energy consumption forms with GDP.In this Table we show number of lags of GDP with m, and number of lags of other variables (LnOIL, LnNG, LnCOAL, LnREE and LnELEC) with n. As we can see there is a bidirectional causality relation between LnGDP and other variables except LnREE and LnCOAL. It seems that these results are not surprising because share of oil, natural gas and electricity is higher than coal and renewable energies in total energy consumption of Iran.

Table3: Hsiao’s (1981) Causality Test Results

5. Conclusions

In this paper we examined the relationship of total energy consumption and it's individual forms (oil, natural gas, coal, renewable energies, electricity) with real GDP in Iran over the period 1974-2010.We employed the Hsiao's (1981) methodology because of stationary variables and also choosing appropriate number of lags for the model. Empirical findings show there is bidirectional causality relation between total energy consumption and its three individual forms (oil, natural gas and electricity) with GDP. Therefore we can accept the feedback hypothesis for GDP and these forms of energy. There is not any evidence for causality relation between GDP and two other energy consumption forms including coal and renewable energies.