Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Energy Consumption and Economic Growth Linkage: Global Evidence from Symmetric and Asymmetric Simulations

Energy Consumption and Economic Growth Linkage: Global Evidence from Symmetric and Asymmetric... IntroductionA plethora of work has been done across geography and it is high time we examined the relationship between economic growth and energy consumption (EC). Although comprehensive studies on the phenomena have been done using linear regression, they failed to account for the data’s hidden cointegrating properties, which lead to spurious, unreliable and myopic outcomes. This empirical irregularity has culminated in the misalignment of energy policies with growth and development objectives almost undesirable. Past studies investigating energy use and economic growth through conventional econometric methods had their own limitations. This paper applies the recent non-linear autoregressive distributed lag (ARDL) procedure of Pesaran (2014) to illuminate and clarify an essential concept that has remained unclear for a long time. This paper examines the asymmetric relationship between energy use and economic growth using panel data for all the countries where data is available as a whole sample and in Brazil, Russia, India, China, and South Africa (BRICS), the Next Eleven, Big Four in Western Europe, Asia-Pacific region, Group of Seven, and South Asian Association for Regional Cooperation (SAARC) region as a sub-sample analysis. The previous studies, in particular, are country-specific, which utilised the non-linear ARDL methodology and/or non-linear co-integration in time-series (Shin et al. 2014; Bayramoglu, Yildirim 2017; Ndoricimpa 2017; Tugcu, Topcu 2018). We specifically use the Pooled Mean Group (PMG) estimator of the dynamic heterogeneous panel (Pesaran et al. 1999) to account for heterogeneity related to country-specific effects of energy use on economic growth but in a non-linear way. Short-run and long-run coefficients of PMG estimates respectively vary across cross-sections in the short run and are constrained to be the same in the long run. Though Huang et al. (2008) believe that EC and economic growth are related only in the long run, Ouedraogo (2013) maintains that it may lead to a false conclusion if we test only the long-run relationship between the two. Therefore, the examination of the dynamic relationship between the two is a must.We considered precisely 85 countries in this study. However, they are not represented geographically. Therefore, we believe a separate analysis for BRICS, the Next Eleven, Big Four in Western Europe, Asia-Pacific region, Group of Seven and SAARC region as a sub-sample analysis will produce outcomes most inclined to the development objectives of each region. To the best of our knowledge, studies in this field used linear panel data methods to examine the subject relationship. Some panel data studies are conducted on a similar subject in African countries while utilising non-linear threshold models, non-linear Granger causality tests, and panel smooth transition vector error correction models (Huang et al. 2008; Omay et al. 2014). Nonetheless, these studies did not account for asymmetries by decomposing energy use and economic growth in their positive or negative components. Moreover, the framework of threshold models does not consider asymmetries and/or shocks; therefore, the econometric method we choose for this study deals with all the limitations and can produce more meaningful results that can rejuvenate policies towards the realisation of growth and development objectives.To test for co-integration between the positive and negative energy use components and economic growth, we used the newly developed hidden panel co-integration test of Hatemi-J (2018). This test can be considered an equivalent to Shin et al.’s (2014) hidden co-integration test on time series. There are several advantages of using Hatemi-J’s (2018) test. To name a few, it can test for hidden co-integration in a panel data framework and separate the variables in the equations for the positive and negative components to test for the presence of co-integration.There is a conflicting set of findings on nature, direction, magnitude, and the drivers of economic growth in the context of energy use and economic growth nexus in the literature. These varying results can be attributed to various sets of specifications/explanatory variables and estimation techniques. The examination of modelling and estimation techniques applied in these studies clearly indicates that much of the literature ignores the asymmetries in interest variables. Economic series adapt to different regimes such as financial crises and unexpected changes in economic policy. This analysis has severe implications for the individual economies in general and the group of economies in particular.Review of literatureEC and economic growth have been a topical issue for the past three decades holding to the increasing energy demand and economic expansion in many developed and developing economies. However, the relationship between both variables has yielded mixed and controversial results. There is still no consensus on which drives which. As a result, four known hypotheses (growth hypothesis, conservation hypothesis, feedback hypothesis, and neutrality hypothesis) have been tested in previous studies. The growth hypothesis posits that EC drives economic growth, while the conservation hypothesis assumes the opposite. The feedback hypothesis argues for bidirectional causality between both variables, unlike the neutrality hypothesis, suggesting no causality between both. Numerous studies (Bayat et al. 2017; Fotourehchi 2017; Hasanov et al. 2017; Lechthaler 2017; Obradovic, Lojanica 2017; Atems, Hotaling 2018; Kumari, Sharma 2018) discovered that EC drives economic growth which is in line with the growth hypothesis mentioned earlier. These studies emphasised the vital role of EC in manufacturing and other production processes, which led to the creation of goods for local consumption and exportation, and hence economic growth. However, the studies mentioned above did not bother much about reverse causality. Studies (like Burakov, Freidin 2017; Kyophilavong et al. 2017; Tatlı 2017; Chen, Fang 2018; Kirikkaleli et al. 2018; Nyasha et al. 2018; Rahman, Velayutham 2020) were able to show that economic growth is capable of increasing EC. This is in line with the conservation hypothesis. According to these studies, economic growth necessitates more energy demand for home appliances, factory machines, waste management, etc. Now, it is also possible that the direction of causality is mutual. That is, EC drives economic growth, and vice versa. This was the major argument of Lin and Wang (2019) for China, Saad and Taleb (2018) for the EU countries, Bazarcheh Shabestari (2018) for Sweden, Zafar et al. (2019) for Asia countries, and Hasan et al. (2017) for Bangladesh. Moreover, Bah and Azam (2017) have a case for the neutrality hypothesis in South Africa.Besides, most of the existing studies have explored the relationship between both variables within a linear framework (see, for instance, Gozgor 2018; Gozgor et al. 2018; Mensah et al. 2019; Wang et al. 2019; Zafar et al. 2019; Ali et al. 2020; Le et al. 2020). There are a few studies that examined a similar relationship, but within a non-linear framework via various techniques, such as non-linear models (Baz et al. 2019; Awodumi, Adewuyi 2020; Ghosh, Kanjilal 2020; Munir et al. 2020), a panel threshold model (Chen et al. 2020; Wang, Wang 2020) and quantile regression (Shahbaz et al. 2018; Zhou et al. 2019).Munir et al. (2020) explored the direction of non-linear causality between economic growth and ASEAN countries’ EC from 1980 to 2016. They discovered a unidirectional causality from economic growth to EC in Malaysia, Thailand, and Indonesia; unidirectional causality from EC to economic growth in Singapore, and feedback causality between the Philippines’ variables. Ivanovski et al. (2020) investigated the impact of renewable and non-renewable energy on OECD (Organisation of Economic Co-operation and Development) countries’ economic growth from 1990 to 2015 via a nonparametric method. Their findings reveal that the latter has no adverse effect on economic growth. Also, both energy sources advance economic growth in non-OECD countries. The authors reiterated the need for the transition to renewables to consolidate growth and uphold environmental sustainability. Awodumi and Adewuyi (2020) examined the effect of EC on economic growth and emissions for oil-exporting countries in Africa from 1980 to 2015. Their findings suggest that EC exacts an asymmetric impact on both economic growth and CO2 emissions. However, the response of both variables to an increase in EC was mixed. Chen et al. (2020) applied a threshold model to examine the link between EC and economic growth for 103 countries from 1995 to 2015. They reported that EC had no significant/positive effect on growth in developed/OECD countries, respectively. However, the consumption of energy above/below a certain threshold will have a significant positive/negative effect on developing countries’ growth.Ali et al. (2020) examined the impact of EC and urbanisation on Nigeria’s economic growth after considering structural breaks. The FMOLS (fully modified ordinary least square) and DOLS (dynamic ordinary least square) results confirmed that EC added to economic growth, while urbanisation inhibited growth in Nigeria. Further findings revealed a one-way causality from urbanisation to economic growth and EC. This contradicts Nathaniel and Bekun’s (2020) earlier study which discovered no direction of causality; supporting the neutrality hypothesis, for Nigeria, Wang and Wang (2020) investigated the non-linear effect of EC on the OECD countries’ economic growth. They discovered that EC has a positive effect on economic growth. Nonetheless, this positive impact changes at different threshold levels.In summary, the relationship between EC and economic growth remains inconclusive with discrepancies in findings emanating from regional differences, peculiarities about countries, estimation techniques, and choice of variables. The majority of existing studies assumed a linear relationship between both, but recent studies have focused on nonlinearity because most economic variables do not exhibit a linear relationship. This study adds to this argument by examining the relationship between both variables in a non-linear framework for different blocs, regions, and organisations.Materials and methodsDataTwo different variables (real GDP per capita and EC) are used in this study. The time period, consistent with data availability, spans 1971 to 2014. This study has been carried out for BRICS, Next Eleven countries, the Economic Cooperation Organization (ECO), the Big Four in Western Europe, Asia-Pacific region, Group of Seven, Arab League, SAARC region, and World sample. EC is measured in kg of oil equivalent per capita, while real GDP per capita, our proxy for economic growth, is measured in constant 2010 USD. The data for both variables were obtained from the World Development Indicator (WDI 2020). These two variables are widely used in the energy-economic growth literature (see Ali et al. 2020; Meo et al. 2021; Nathaniel 2020). In all, 836 observations were used for conducting the relevant analysis. See Table A1 in Appendix for the list of countries adopted in this study.MethodsThe globalisation of the world’s economy has promoted interconnectivity and the transfer of shocks from one economy to another. Therefore, interdependence among countries is possible. The study proceeds with Pesaran’s (2004) cross-sectional dependence (CD) test. The CD test is of utmost importance to ensure unbiased estimates and estimator efficiency. We further applied the Im-Pesaran-Shin (CIPS) test to ascertain the stationarity properties of the variables. The CIPS test is robust amidst CD. Once variables are stationary, a co-integration test is required to ascertain the possibility of a long-run relationship. In this study, the Westerlund (2007) test investigates the presence of co-integration among the variables. Just like the CIPS test, the Westerlund (2007) test is also robust amidst CD and performs better than the traditional first-generation co-integration tests, including the Kao and Chiang (1999) and Pedroni (2000) co-integration tests.Unlike most previous studies, we examine the asymmetric impact of EC on economic growth (GR) via the non-linear autoregressive distributive lag (NARDL) model. The primary model for this study is given as:(1)GRt=φ0+φ1(ECt)+μt{GR_t} = {\varphi _0} + {\varphi _1}\left( {{EC_t}} \right) + {\mu _t}The NARDL model of Shin et al. (2014) can estimate short and long-run dynamics even when the sample size is small (Pesaran et al. 2001; Narayan, Narayan 2005; Meo et al. 2020). The non-linear form of the model is given as:(2)GRt=φ0+φ1(ECt+)+φ2(ECt−)G{R_t} = {\varphi _0} + {\varphi _1}\left( {EC_t^ + } \right) + {\varphi _2}\left( {EC_t^ - } \right)where ϕ0, ϕ1 and ϕ2 are the long-run parameters. The asymmetric effects of EC are incorporated by positive change ECt+EC_t^ + and negative changes ECt−EC_t^ - respectively, whereas ECt+EC_t^ + and ECt−EC_t^ - are the partial sums of positive and negative changes in EC, respectively. Now, to obtain the short-run effect, Eq. (1) can be re-specified as:(3)ΔGRt=∑k=1mθ1kΔGRt−k+∑k=1mθ2kΔECt−k+∂1GRt−1+∂1ECt−1+μt\Delta G{R_t} = \sum\nolimits_{k = 1}^m {{\theta _{1k}}} \Delta G{R_{t - k}} + \sum\nolimits_{k = 1}^m {{\theta _{2k}}} \Delta E{C_{t - k}} + {\partial _1}G{R_{t - 1}} + {\partial _1}E{C_{t - 1}} + {\mu _t}Eq. (3) is the specification that encompasses both long-run and short-run coefficients. Δ is the short-run coefficient while ∂1 and ∂2 are the long-run coefficients. We specify Eq. (4) to derive the asymmetric cointegrating.(4)γt=β+xt++β−xt−+μt{\gamma _t} = {\beta ^ + }x_t^ + + {\beta ^ - }x_t^ - + {\mu _t}where xt are the long term parameters of kx1 vector of regressors decomposed as (5)xt=xt++xt−{x_t} = x_t^ + + x_t^ - where xt+x_t^ + and xt−x_t^ - are the explanatory variables, decomposed into a partial sum of positive and negative changes. Eqs. (6) and (7) are the partial sums of positive and negative changes in EC.(6)EC+=∑i=1tΔECi+=∑i=1tmin(ΔECi,0)E{C^ + } = \sum\nolimits_{i = 1}^t {\Delta EC_i^ + = } \sum\nolimits_{i = 1}^t {\min \left( {\Delta E{C_i},0} \right)} (7)EC−=∑i=1tΔECi−=∑i=1tmin(ΔECi,0)E{C^ - } = \sum\nolimits_{i = 1}^t {\Delta EC_i^ - = } \sum\nolimits_{i = 1}^t {\min \left( {\Delta E{C_i},0} \right)} Replacing EC in Eq. (3) by EC+ and EC− gives us the asymmetric ARDL model:(8)ΔGRt=∂0+∑k=1mϕkΔGRt−k+∑k=1mϕkΔECt−k++∑k=1mϕkΔECt−k−+∂1GRt−1+∂2ECt−1++∂3ECt−1−+μt\matrix{ {\Delta G{R_t} = {\partial _0} + \sum\nolimits_{k = 1}^m {{\phi _k}\Delta G{R_{t - k}} + } \sum\nolimits_{k = 1}^m {{\phi _k}\Delta EC_{t - k}^ + + } } \cr {\sum\nolimits_{k = 1}^m {{\phi _k}\Delta EC_{t - k}^ - + {\partial _1}G{R_{t - 1}}} + {\partial _2}EC_{t - 1}^ + + {\partial _3}EC_{t - 1}^ - + {\mu _t}} \cr } From Eq. (8), the long-run impact of both positive and negative changes in EC are −∂2 / ∂1 and −∂3 / ∂1. One huge advantage of the NARDL technique is in its ability to accommodate variables with various levels of stationarity whether I (0) or I (1) or I (0) and I (1)Results and discussionsThe descriptive statistics of the regressors used in the NARDL model of this paper have been presented in Table 1 with mean, standard deviation, minimum, maximum, and coefficient of variation.Table 1Descriptive statistics.VariableObsMeanStd. dev.MinMaxCoef of variationWhole sampleGDPG3,4311.7154.945−64.9953.972.88EU3,4312,130.8862,369.31699.7118,178.1410.011BRICSGDPG1146.8234613.996938−7.388113.638330.585EU114842.0321559.7145279.3802,236.7300.664Next ElevenGDPG4182.494.55−30.1822.121.82EU418958.1930.1599.705,248.50.97ECOGDPG1141.406.05−30.1822.144.32EU1141,064.1680.6302.542,889.100.63Big Four in Western EuropeGDPG1521.661.78−5.465.871.07EU1524,053.8341.33,172.84,658.30.048Asia-Pacific regionGDPG5703.473.47−14.3513.641EU570944.231,122.899.714,083.81.189Group of SevenGDPG2661.662.10−7.946.331.26EU2664,770.71,970.92,203.88,455.50.41Arab LeagueGDPG4181.1485.10−25.6111.834.44EU4181,811.71,903.7235.186,871.81.05SAARC regionGDPG1903.012.35−7.3990.78EU190334.7119.699.71636.50.35BRICS – Brazil, Russia, India, China, and South Africa; ECO – Economic Cooperation Organization; SAARC – South Asian Association for Regional Cooperation; GDPG – GDP growth ; Obs – Observations.Source: own elaboration.In Table 1, we report the summary statistics that contain the mean (averages), observations, standard deviation, the minimum (Min) values as well as their maximums (Max). In our joint sample analysis, the coefficient of variation measures the heterogeneity of regressors. With a large coefficient of variation in the whole sample, we established the heterogeneity of regressors with GDP per capita having the more severe sample variances than energy use.Across the sample period, the averages across varying samples were considered within the study. In the whole model, the averages for energy use are large compared to GDP per capita averages. Apart from the fact that this statistical revelation points to ample energy use among the sample frame, the structural properties of data permit drawing credible inferences in a broad regional specific categorisation but not without controlling for the heterogeneity of regressors (Lewbel 2012). Across the BRICS, Next Eleven, Big Four in Western Europe, Asia-Pacific region, Group of Seven, and SAARC region economic classifications, the heterogeneity of regressors seems relatively weak and the averages of energy use seem predominate their large variances arrangement. Nonetheless, the Arab League shows heterogeneity well above the whole sample, and this implies its heterogeneity properties are greater than those observed in the entire sample. With the data extensive and fraught with the heterogeneity of regressors across samples and a pointer to nonlinearities among regressors, we proceed to test for common factor restrictions leading to the verification of the presence of cross-sectional dependence.The regional EC pattern and their growth-inducing strategies are almost identical with notable variances (Ahmed, Wahid 2011). These homogenous growth stabilisation approaches, induced by varying EC, ensure that economic indices are fraught with interdependence among cross-sections. It then becomes imperative to test for CD and consider controlling such disturbances to reach conclusions that are non-spurious (Pesaran 2015). We tried the null of no CD using the Pesaran (2004) CD test. In Table 2, Pesaran (2004) confirms the existence of CD with probability values <1%. The presence of CD calls for estimation procedures that control such disturbances. We proceed to estimate the second-generation panel stationarity test of CIPS of Pesaran (2004) because of its high-level precisions and capacity to control CD in panel data econometrics. In consonance with Phillips and Sul (2003), we followed this path to avert spurious outcomes when CD is not controlled.Table 2CD test results (Pesaran’s test 2004).VariablesCD-testP-valueAverage joint TMean ÏMean abs (Ï)GDP growth per capita21.4170.00037.950.060.18Energy use102.570.00037.950.270.54Notes: Under the null hypothesis of cross-sectional independence, CD ~ N (0, 1) P-values close to zero indicate data are correlated across panel groups.CD – cross-sectional dependence.Source: own elaboration.Beyond the descriptive revelation of heteroskedasticity and nonlinear properties among the regressors, it becomes apt and imperative to adopt the leading stationarity test to set a clear line of thought on the stationarity characteristics of the data set to reach non-spurious, policy and theory consistent empirical outcomes. Using the cross-sectionally augmented IPS test of Pesaran (2007) that considers CD when estimating its series, we found impressive stationarity outcomes. Table 3 reports the second generation panel stationarity test from the CIPS test of Pesaran (2007). We found regressors to be stationary at first difference across all the series I (1). These empirical outcomes establish an apparent need to uncover the covariance characteristics of the data. We proceed to estimate the Durbin-Hausman co-integration (Westerlund 2007) test to establish long-run cointegrating relations because of its relative sensitivity to CD.Table 3Unit root test results in the presence of cross-section dependence (Pesaran’s test 2007).VariablesLevelFirst differenceCIPS (−1)CIPS (−2)CIPS (−3)Critical value (5%)CIPS (−1)CIPS (−2)CIPS (−3)Critical value (5%)No intercept no trendGDP growth−1.231−3.761−3.774−1.52−5.963−5.935−5.934−1.52Energy use−1.134−1.267−1.268−1.52−5.097−4.940−4.883−1.52Only interceptGDP growth−1.107−4.107−4.107−2.08−5.977−5.977−5.977−2.08Energy use−1.606−1.606−1.606−2.08−5.272−5.272−5.272−2.08Linear trendGDP growth−2.48−4.470−4.453−2.56−6.206−6.161−6.181−2.65Energy use−1.836−1.936−1.893−2.56−5.537−5.461−5.410−2.65Notes: H0 (homogeneous non-stationary): bi = 0 for all i.Source: own elaboration.We prefer the Westerlund (2007) co-integration test to the Pedroni (1999) test due to its relative sensitivity to CD and slope homogeneity of regressors. Westerlund (2007) produces consistent estimates under the mild assumption. Since it disregards lag information about integrating series orders, it is widely applicable in a broader context. By permitting spatial correlation of CD, the Westerlund (2007) test controls for unobserved heterogeneity of regressors that characterises panel data estimation. To reach meaningful outcomes in the presence of cross-section dependence issues, the Westerlund (2007) test computes P-values robustly based on bootstrap iterations. Through the error correction component, cointegrating properties are revealed in the individual sample and across samples. Westerlund (2008) went further by computing the nullity and otherwise of the cross-section unit using the weighted averages and associative t-stats.Further evidence is sought into the cointegrating properties when the Westerlund (2007) tests establish the nullity of co-integration across panels by pooling the estimates in a comprehensive panel framework. The result presented in Table 4 shows that variables in different samples do not tend towards their long-run equilibrating position even when CD exists. Our result is robust at a 1% level of significance. We proceed to estimate the asymmetric co-integration in tandem with Hatemi-J’s (2020) in Table 5.Table 4Panel co-integration test (Westerlund’s test 2007).StatisticsModel with interceptModel with trendValueZ-valueP-valueValueZ-valueP-valueBRICSGt−4.369−4.9970.000−4.010−3.5040.000Ga−23.489−5. 2010.000−24.688−3.2290.001Pt−6.332−3.8470.000−5.953−2.6230.004Pa−22.190−7.0070.000−22.953−3.9610.000Next ElevenGt−1.7280.1830.573−2.421−0.2220.412Ga−4.9981.3070.904−7.7462.0810.981Pt−0.5954.2311.000−4.9152.3910.992Pa−0.2422.9800.999−2.9423.2410.999ECOGt−0.2432.9590.999−2.0330.7100.761Ga−1.5991.7640.961−3.2512.2310.987Pt0.7703.2971.000−1.7242.2130.987Pa0.5871.8790.970−2.3061.8720.969Big Four in Western EuropeGt−2.295−1.1520.125−2.696−0.8120.208Ga−12.127−1.8320.034−11.7060.0900.536Pt0.6973.6131.000−3.0921.2950.902Pa0.8282.2790.989−3.7191.7010.956Asia-Pacific regionGt−2.365−2.5300.006−2.421−0.2610.397Ga−12.430−3.7620.000−13.242−0.7010.242Pt−9.287−3.7020.000−8.0280.1760.570Pa−11.345−6.2070.000−11.655−1.7200.043Group of SevenGt−2.633−2.5200.006−2.831−1.5130.065Ga−13.333−3.0090.001−12.795−0.3050.380Pt−2.2931.5460.939−3.8711.9640.975Pa−3.1470.6460.741−4.8771.7500.960Arab LeagueGt−1.6290.5480.708−1.7772.4050.992Ga−100.936−57.1450.000−67.851−27.2350.000Pt−2.2232.5930.995−1.6306.1471.000Pa−1.9011.7400.959−1.8063.8551.000SAARC regionGt−2.532−1.8770.030−2.579−0.5840.280Ga−14.997−3.2260.001−13.943−0.6350.263Pt−5.671−2.4480.007−4.5310.2200.587Pa−15.506−5.6790.000−14.606−2.0690.019World sampleGt−2.715−1.8070.035−2.3310.0770.530Ga−9.223−0.6620.254−9.1360.7330.768Pt−2.751−0.2450.403−2.0541.8350.967Pa−2.6940.5990.726−4.0681.3740.915BRICS – Brazil, Russia, India, China, and South Africa; ECO – Economic Cooperation Organization; SAARC – South Asian Association for Regional Cooperation.Source: own elaboration.Table 5Asymmetric co-integration results of Hatemi-J’s test (2020).VariablesKao test resultsIm et al. (2003) unit root test results on residualsDecision regarding residualsArab LeagueY+, X+−2.4916 (0.0003)I (0), StationaryY+, X−−2.7552 (0.000)I (0), StationaryY−, X−−2.7677 (0.000)I (0), StationaryY−, X+−2.6207 (0.000)I (0), StationaryAsia-Pacific regionY+, X+−2.6571 (0.000)I (0), StationaryY+, X−−2.9797 (0.000)I (0), StationaryY−, X−−2.3041 (0.003)I (0), StationaryY−, X+−2.4447 (0.0001)I (0), StationaryBig Four in Western EuropeY+, X+−4.0673 (0.000)I (0), StationaryY+, X−−3.2235 (0.000)I (0), StationaryY−, X−−2.5821 (0.012)I (0), StationaryY−, X+−2.6298 (0.010)I (0), StationaryECOY+, X+−3.5681 (0.0004)I (0), StationaryY+, X−−2.5875 (0.025)I (0), StationaryY−, X−−2.8706 (0.013)I (0), StationaryY−, X+−2.7547 (0.013)I (0), StationaryGroup of SevenY+, X+−3.2938 (0.000)I (0), StationaryY+, X−−3.3661 (0.000)I (0), StationaryY−, X−−2.5256 (0.002)I (0), StationaryY−, X+−2.5917 (0.001)I (0), StationaryNext ElevenY+, X+−1.9333 (0.081)I (0), StationaryY+, X−−1.7222 (0.301)I (0), Stationary at 10%Y−, X−−2.0643 (0.021)I (0), StationaryY−, X+−2.0375 (0.026)I (0), StationaryWorld sampleY+, X+−2.3252 (0.000)I (0), StationaryY+, X−−2.6464 (0.000)I (0), StationaryY−, X−−2.6051 (0.000)I (0), StationaryY−, X+−2.4566 (0.000)I (0), StationaryECO – Economic Cooperation Organization.Source: own elaboration.The results of no co-integration between energy use and economic growth across the various samples led to the verification of hidden co-integration to reveal the series’s negative and positive cointegrating properties of components. Hatemi-J’s (2018) test for asymmetric co-integration was employed for this purpose. It revealed (2020) the hidden panel cointegrating properties. The initial analysis presents false outcomes of no co-integration by hidden co-integration, but further examination reveals co-integration between negative and positive components. This hidden composition leading to false cointegrating outcomes could result from unobserved factors in the non-cointegrating relations causing negative and components series to vary differently from idiosyncratic influences. Establishing a clear line of thought on the heterogeneous impact of energy use and economic growth asymmetric relationship informs about the need for this approach. Heterogenous energy pricing, the various institutional setup that moderates policy and content of reforms towards change, trade and export strategies, and so on, influence the output elasticity of the model in a non-linear and different manner. Thus, establishing the asymmetric properties of the model’s parameter will produce results that are most inclined to each region’s development objectives. Hatemi (2020) assumes the first difference integration test stationary of the series before decomposing regressors into the positive and negative components. It is also important to note that the residual from the series has to be stationary to establish regressors’ collinearity. The result in Table 5 reveals nonlinearities in the co-integration relationship between energy use and economic growth for all samples with residuals stationary at the level.Table 6 presents the linear ARDL result for the model. We relied on the Mean Group estimation procedure because of the high precision and relevance in establishing a clear line of thought on the energy use-growth relations. The convergence term is negative and statistically significant at a 1% level. The convergence terms give the speed of adjustment from long-run tendencies to short-run equilibrating positions. We have found long-run and short-run behavioural movements of the data set in the BRICS and SAARC regions. The Wald test has revealed the asymmetric properties of the model and affirmed no short-run and long-run asymmetric relations. In the search for hidden asymmetric properties of the energy use induced growth model, we proceed to estimate the non-linear ARDL model in consonance with Shin et al. (2014).Table 6Linear ARDL results.ΔYMG (BRICS)PMG (SAARC)ECT−0.008* (0.043)−0.0064577 (0.389)Long-run effect−0.829* (0.000)−1.040631* (0.000)Short run effect0.054* (0.001)0.0496974* (0.006)_cons4.914* (0.000)2.895211* (0.000)Hausman testH0 = PMGH1 = MG7.51** (0.0234)0.21 (0.6462)Wald test (long-run asymmetry)1.75 (0.1855)1.09 (0.2971)Wald test (short-run asymmetry)1.26 (0.2616)0.42 (0.5146)Log likelihood−366.9499Notes: The P-values are given [* at 1%, ** at 5%].ARDL – autoregressive distributed lag; BRICS – Brazil, Russia, India, China, and South Africa; PMG – Pooled Mean Group; SAARC – South Asian Association for Regional Cooperation; ECT – error correction term.Source: own elaboration.To assign numerical weights to the asymmetric relationship between energy use and economic growth within a global analytic framework, we report the non-linear ARDL result in Table 7. Beyond linear relations in the energy use-growth dynamic associations, we rely on the PMG to establish the regressors’ independent capacities to predict variations in the response variables. Following Shin et al. (2014), we unravel the energy use-growth dynamic relations’ asymmetric properties in an idiosyncratic capacity.Table 7Nonlinear panel ARDL.VariablesArab LeagueAsia-Pacific regionBig Four in Western EuropeECOΔY+ΔY−ΔY+ΔY−ΔY+ΔY−ΔY+ΔY−ECT−0.585* (0.000)−0.409* (0.000)−0.520* (0.000)−0.455* (0.000)−0.415* (0.000)−0.577* (0.000)−0.260** (0.033)−0.598 (0.000)eut+eu_t^ + −0.0080* (0.000)0.0047* (0.001)−0.0114 (0.000)0.014* (0.000)−0.005** (0.030)−0.002* (0.992)0.069 (0.009)−0.025 (0.012)eut−eu_t^ - 0.003** (0.011)0.0054* (0.002)0.0242* (0.000)−0.028* (0.000)0.025* (0.000)−0.011* (0.000)0.220* (0.001)−0.065 (0.000)Δeut+\Delta eu_t^ + 0.0011 (0.267)−0.0006 (0.974)−0.046** (0.052)0.045** (0.045)0.001 (0.914)−0.007 (0.393)0.036 (0.061)−0.021 (0.078)Δeut−\Delta eu_t^ - 0.952 (0.267)0.051 (0.242)0.006 (0.874)0.001 (0.974)0.0019 (0.595)0.028 (0.929)0.018 (0.490)0.002 (0.894)Δeut−1+\Delta eu_{t - 1}^ + 0.005 (0.703)−0.006 (0.686)−0.018 (0.146)0.018 (0.084)0.004 (0.318)−0.002 (0.456)0.023 (0.166)−0.020 (0.217)Δeut−1−\Delta eu_{t - 1}^ - −0.103 (0.290)0.105 (0.269)0.060 (0.619)−0.038 (0.689)−0.002 (0.579)0.004 (0.238)−0.012 (0.445)0.013 (0.424)Constant2.328* (0.005)−6.688 (0.229)−0.296 (0.862)−3.550** (0.013)−1.984 (0.173)−1.281 (0.368)3.920 (0.725)−2.563 (0.783)Hausman (χ2)H0 = PMGH1 = MG3.17 (0.2054)1.55 (0.461)3.56 (0.613)2.32 (0.192)Wald test (long-run asymmetry)3.320** (0.0367)5.69* (0.002)5.61** (0.020)7.76* (0.005)Wald test (short-run asymmetry)0.031 (0.142)0.29 (0.370)0.91 (0.213)0.12 (0.730)Log likelihood−1176.886−1072.3−193.7512−258.049Notes: P values are given in [], where eu is energy use; *1%; **5%; ARDL – autoregressive distributed lag; ECO – Economic Cooperation Organization; ECT – error correction term.VariablesGroup of SevenNext ElevenWorld sampleΔY+ΔY−ΔY+ΔY−ΔY+ΔY−ECT−0.427* (0.000)−0.5535* (0.000)−0.458* (0.000)−0.413* (0.000)−0.494* (0.000)−0.446* (0.000)eut+eu_t^ + −0.006* (0.000)0.021** (0.036)0.034* (0.007)−0.036* (0.000)−0.001** (0.013)0.001* (0.003)eut−eu_t^ - 0.021* (0.000)−0.013* (0.000)0.440* (0.000)0.066* (0.000)0.015* (0.000)−0.011* (0.000)Δeut+\Delta eu_t^ + −0.061 (0.466)−0.049 (0.710)−1.114 (0.264)1.167 (0.242)−0.004 (0.577)0.005 (0.544)Δeut−\Delta eu_t^ - 0.026 (0.243)−0.002 (0.290)1.337** (0.026)−1.217** (0.036)−0.005 (0.480)0.005 (0.463)Δeut−1+\Delta eu_{t - 1}^ + 0.011 (0.992)−0.009 (0.627)0.945 (0.257)−0.955 (0.259)−0.010 (0.397)0.011 (0.380)Δeut−1−\Delta eu_{t - 1}^ - −0.001 (0.626)0.009 (0.396)−0.339 (0.640)0.436 (0.556)−0.003 (0.794)0.004 (0.747)Constant−1.813 (0.075)−1.394 (0.145)−296.64* (0.006)−71.781 (0.121)1.881744 (0.236)−5.050* (0.001)Hausman (χ2)H0 = PMGH1 = MG2.17 (0.250)2.55 (0.601)4.17 (0.542)Wald test (long-run asymmetry)4.05** (0.0443)7.76* (0.005)5.76** (0.051)Wald test (short-run asymmetry)0.07 (0.7917)0.12 (0.730)0.83 (0.305)Log likelihood−284.11111.42−3620.302P values are given in (), where eu is energy use; *1%; **5%.Source: own elaboration.Table 6 rejected the nullity of covariance among economic growth and positive and negative components of energy use. With the Wald test outcomes not statistically significant at any significance threshold, we failed to reject the null of asymmetry in the short run while confirming the presence of asymmetry in the long run across the Arab League, Asia-Pacific region, Big Four in Western Europe, ECO, Group of Seven, Next Eleven and World sample. Across the heterogeneous sample leading to a global framework, the error correction components are appropriately signed (negative) and statistically significant at a 1% level of significance. By intuitions, the convergence from long-run behaviours is gradually adjusted to the short-run equilibrating positions. The sign and magnitude of the convergence terms established asymmetric properties in the energy use-growth relationship in a broad-based sample. For clarity, positive changes in our response variable ΔY+ denote growth expansion while negative changes ΔY− denote growth contraction. Furthermore, positive values of the parameters estimate eu+ that are associated with the positive value of the response variables ΔY+ imply growth expansion and negative values of parameters eu− associated with positive response variables ΔY+ imply economic contractions. Nonetheless, a positive value of parameter estimates eu+ with the negative response variable ΔY− implies economic recovery and negative values of parameter eu− estimated in a negative response variable ΔY− imply advance in economic contractions.The result from Table 7 was analysed based on regional sample categorisation, and the interpretations were presented logically and sequentially. With economic growth expected to react differently to heterogenous positive and negative energy shocks and produce non-linear results that are most inclined to each region’s development objective, our result follows a stratified and sample-specific pattern to reach conclusions regionally generalisable. The asymmetric short-run relations are observed for the Arab League, Asia-Pacific region, and Next Eleven. The short-run asymmetric properties in these samples depict energy shocks that could be a result of their extensive energy dependence. There is no observed asymmetric relationship between energy use and growth in the Big Four in Western Europe, ECO, Group of Seven, and World sample with just two of their coefficients significant at established thresholds. Short-run asymmetric relations are a testament to variations in cross-sectional domains that defines a country or regional groupings. The variations in the asymmetric properties across the sample categorisation could result from the heterogeneous level of reliance on energy use by different samples observed in this study.To establish linearity in reverse causative structure in the energy use-growth relation across varying samples, we reported the Dumitrescu and Hurlin (2012) panel Granger test in Table 8. The Dumitrescu and Hurlin (2012) panel Granger test accounts for the endogeneity of the regressor by computing the probability values of the numerical coefficient matrix. Using a large bootstrap iterative process, we accounted for cross-sections of the country and regional dimensions of the data set. We confirmed stationarity at the first difference before engaging the Panel Causality test in consonance with contemporaneous Granger causality test literature. We rejected the null hypothesis of Real GDP per capita does not Granger-cause energy use in the Big Four countries in Western Europe, ECO, Group of Seven and SAARC regions. By rejecting the nullity of the no Granger cause relations in the real GDP-energy use, we established uni-directional causality from real GDP to energy use in the Big Four countries in Western Europe, ECO, Group of Seven, and SAARC regions. In other related but distinct findings, we also rejected the null of Energy use does not Granger-cause real GDP per capita in the Group of Seven and SAARC region. This implies that a one-way causality exists from energy use to real GDP per capita in the Group of Seven and SAARC region. In summary, a bilateral causality exists between energy use and real GDP per capita in the Group of Seven.Table 8The result of Dumitrescu and Hurlin’s (2012) panel causality test.Null hypothesisW-statZ-barP-valueLagArab LeagueReal GDP per capita does not Granger-cause energy use3.502375.134793.E−071Energy use does not Granger-cause real GDP per capita1.888561.738470.08211Asia-Pacific regionReal GDP per capita does not Granger-cause energy use2.580860.748830.45402Energy use does not Granger-cause real GDP per capita5.586115.829656.E−092Big Four in Western EuropeReal GDP per capita does not Granger-cause energy use3.369032.927180.0034*1Energy use does not Granger-cause real GDP per capita1.344340.357670.72061BRICSReal GDP per capita does not Granger-cause energy use0.83674−0.248120.80401Energy use does not Granger-cause real GDP per capita2.777341.884700.05951ECOReal GDP per capita does not Granger-cause energy use16.213616.65190.0000*1Energy use does not Granger-cause real GDP per capita2.497381.577010.11481Group of SevenReal GDP per capita does not Granger-cause energy use10.430615.72750.0000*1Energy use does not Granger-cause real GDP per capita2.988223.232980.0012*1Next ElevenReal GDP per capita does not Granger-cause energy use4.696907.648712.E−141Energy use does not Granger-cause real GDP per capita5.205408.718880.0000*1SAARCReal GDP per capita does not Granger-cause energy use1.02665−0.050860.95941Energy use does not Granger-cause real GDP per capita9.7771912.36510.0000*1World sampleReal GDP per capita does not Granger-cause energy use5.7498426.76880.0000*1Energy use does not Granger-cause real GDP per capita2.237456.709962.E−111Notes: We computed the probability values using large iterative bootstrap identifications; W-Stat statistic presents the averages from cross-sections of N standard individual Wald statistics of Granger non-causality tests. The Z-bar statistic is a standardised statistic.BRICS – Brazil, Russia, India, China, and South Africa; ECO – Economic Cooperation Organization; SAARC – South Asian Association for Regional Cooperation.Source: own elaboration.Conclusion and policy suggestionsThis paper investigates the following questions. Is the relationship between economic growth and EC asymmetric? What if we investigate the same connection across a different group of countries? Suppose the relationship between the variables of interest is asymmetric, does it prevail both in the short run and in the long run? Does the impact of EC on economic growth enhance or contract growth across a different group of countries and in the whole sample? These and other relevant questions are answered while examining the impact of EC on economic growth in a non-linear panel ARDL framework.Using data from 1977 to 2014 for 85 countries (as a whole sample) and seven similar groups of countries, the study first applied the second generational panel unit root test due to cross-sectional dependency. To check for the existence of a long-run relationship between EC and economic growth, Hatemi-J’s (2018) test for asymmetric co-integration was employed for this purpose. To determine whether the impact of EC on economic growth is growth-enhancing or growth contracting both in the short run and in the long run, we use the non-linear ARDL model. Using these methods, some interesting results are found.This paper finds a linear relationship between EC and economic growth in BRICS and SAARC countries while an asymmetric relationship in the rest of the countries, including the whole sample of 85. Furthermore, among those groups of countries where the relationship is asymmetric, we have only found a long-run asymmetry. This confirms the possible impact of changes in EC on economic growth in the short run but not in the long run.The asymmetric relationship between the subject variables means that the impact of EC on economic growth varies on the phase/state of the economic cycle. In this context, our results suggest that positive shocks on energy use have a positive and significant effect on economic growth only in ECO and the Next Eleven in the periods of economic expansion. This means that in ECO and the Next Eleven groups of countries, a positive shock to energy use is growth-enhancing while in the rest of the group of countries it is growth-contacting. Moreover, the study results suggest that economic recovery is the case in the Arab League, Asia-Pacific region, Group of Seven, and the whole sample. However, a negative shock to EC in the Group of Seven, Asia-Pacific region, Big Four in Western Europe, ECO, and the whole sample worsens the economic contraction. In the short run, the results are mostly insignificant.Based on the findings, the relationship between economic growth and EC is asymmetric, and this paper suggests adjusting the energy supply and energy demand in a group of countries where a positive shock to energy use is growth-enhancing. Similarly, EC could be encouraged by expanding industrialisation, urbanisation, and transport in countries where a positive shock to EC leads to economic recovery. However, in a group of countries where a positive shock to EC worsens economic growth, EC should be discouraged while encouraging the use of close substitutes of energy, lowering the use of private vehicles, promoting solar and water energy, reducing an excess use of luxurious goods. Although we have tried to carry out this research carefully, there may be some methodological pitfalls. Thus, further research may be carried out to solve these shortcomings. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Quaestiones Geographicae de Gruyter

Energy Consumption and Economic Growth Linkage: Global Evidence from Symmetric and Asymmetric Simulations

Download PDF
16 pages

Loading next page...
 
/lp/de-gruyter/energy-consumption-and-economic-growth-linkage-global-evidence-from-G4HL376H6d

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
de Gruyter
Copyright
© 2022 Wajid Ali et al., published by Sciendo
ISSN
2081-6383
eISSN
2081-6383
DOI
10.2478/quageo-2022-0012
Publisher site
See Article on Publisher Site

Abstract

IntroductionA plethora of work has been done across geography and it is high time we examined the relationship between economic growth and energy consumption (EC). Although comprehensive studies on the phenomena have been done using linear regression, they failed to account for the data’s hidden cointegrating properties, which lead to spurious, unreliable and myopic outcomes. This empirical irregularity has culminated in the misalignment of energy policies with growth and development objectives almost undesirable. Past studies investigating energy use and economic growth through conventional econometric methods had their own limitations. This paper applies the recent non-linear autoregressive distributed lag (ARDL) procedure of Pesaran (2014) to illuminate and clarify an essential concept that has remained unclear for a long time. This paper examines the asymmetric relationship between energy use and economic growth using panel data for all the countries where data is available as a whole sample and in Brazil, Russia, India, China, and South Africa (BRICS), the Next Eleven, Big Four in Western Europe, Asia-Pacific region, Group of Seven, and South Asian Association for Regional Cooperation (SAARC) region as a sub-sample analysis. The previous studies, in particular, are country-specific, which utilised the non-linear ARDL methodology and/or non-linear co-integration in time-series (Shin et al. 2014; Bayramoglu, Yildirim 2017; Ndoricimpa 2017; Tugcu, Topcu 2018). We specifically use the Pooled Mean Group (PMG) estimator of the dynamic heterogeneous panel (Pesaran et al. 1999) to account for heterogeneity related to country-specific effects of energy use on economic growth but in a non-linear way. Short-run and long-run coefficients of PMG estimates respectively vary across cross-sections in the short run and are constrained to be the same in the long run. Though Huang et al. (2008) believe that EC and economic growth are related only in the long run, Ouedraogo (2013) maintains that it may lead to a false conclusion if we test only the long-run relationship between the two. Therefore, the examination of the dynamic relationship between the two is a must.We considered precisely 85 countries in this study. However, they are not represented geographically. Therefore, we believe a separate analysis for BRICS, the Next Eleven, Big Four in Western Europe, Asia-Pacific region, Group of Seven and SAARC region as a sub-sample analysis will produce outcomes most inclined to the development objectives of each region. To the best of our knowledge, studies in this field used linear panel data methods to examine the subject relationship. Some panel data studies are conducted on a similar subject in African countries while utilising non-linear threshold models, non-linear Granger causality tests, and panel smooth transition vector error correction models (Huang et al. 2008; Omay et al. 2014). Nonetheless, these studies did not account for asymmetries by decomposing energy use and economic growth in their positive or negative components. Moreover, the framework of threshold models does not consider asymmetries and/or shocks; therefore, the econometric method we choose for this study deals with all the limitations and can produce more meaningful results that can rejuvenate policies towards the realisation of growth and development objectives.To test for co-integration between the positive and negative energy use components and economic growth, we used the newly developed hidden panel co-integration test of Hatemi-J (2018). This test can be considered an equivalent to Shin et al.’s (2014) hidden co-integration test on time series. There are several advantages of using Hatemi-J’s (2018) test. To name a few, it can test for hidden co-integration in a panel data framework and separate the variables in the equations for the positive and negative components to test for the presence of co-integration.There is a conflicting set of findings on nature, direction, magnitude, and the drivers of economic growth in the context of energy use and economic growth nexus in the literature. These varying results can be attributed to various sets of specifications/explanatory variables and estimation techniques. The examination of modelling and estimation techniques applied in these studies clearly indicates that much of the literature ignores the asymmetries in interest variables. Economic series adapt to different regimes such as financial crises and unexpected changes in economic policy. This analysis has severe implications for the individual economies in general and the group of economies in particular.Review of literatureEC and economic growth have been a topical issue for the past three decades holding to the increasing energy demand and economic expansion in many developed and developing economies. However, the relationship between both variables has yielded mixed and controversial results. There is still no consensus on which drives which. As a result, four known hypotheses (growth hypothesis, conservation hypothesis, feedback hypothesis, and neutrality hypothesis) have been tested in previous studies. The growth hypothesis posits that EC drives economic growth, while the conservation hypothesis assumes the opposite. The feedback hypothesis argues for bidirectional causality between both variables, unlike the neutrality hypothesis, suggesting no causality between both. Numerous studies (Bayat et al. 2017; Fotourehchi 2017; Hasanov et al. 2017; Lechthaler 2017; Obradovic, Lojanica 2017; Atems, Hotaling 2018; Kumari, Sharma 2018) discovered that EC drives economic growth which is in line with the growth hypothesis mentioned earlier. These studies emphasised the vital role of EC in manufacturing and other production processes, which led to the creation of goods for local consumption and exportation, and hence economic growth. However, the studies mentioned above did not bother much about reverse causality. Studies (like Burakov, Freidin 2017; Kyophilavong et al. 2017; Tatlı 2017; Chen, Fang 2018; Kirikkaleli et al. 2018; Nyasha et al. 2018; Rahman, Velayutham 2020) were able to show that economic growth is capable of increasing EC. This is in line with the conservation hypothesis. According to these studies, economic growth necessitates more energy demand for home appliances, factory machines, waste management, etc. Now, it is also possible that the direction of causality is mutual. That is, EC drives economic growth, and vice versa. This was the major argument of Lin and Wang (2019) for China, Saad and Taleb (2018) for the EU countries, Bazarcheh Shabestari (2018) for Sweden, Zafar et al. (2019) for Asia countries, and Hasan et al. (2017) for Bangladesh. Moreover, Bah and Azam (2017) have a case for the neutrality hypothesis in South Africa.Besides, most of the existing studies have explored the relationship between both variables within a linear framework (see, for instance, Gozgor 2018; Gozgor et al. 2018; Mensah et al. 2019; Wang et al. 2019; Zafar et al. 2019; Ali et al. 2020; Le et al. 2020). There are a few studies that examined a similar relationship, but within a non-linear framework via various techniques, such as non-linear models (Baz et al. 2019; Awodumi, Adewuyi 2020; Ghosh, Kanjilal 2020; Munir et al. 2020), a panel threshold model (Chen et al. 2020; Wang, Wang 2020) and quantile regression (Shahbaz et al. 2018; Zhou et al. 2019).Munir et al. (2020) explored the direction of non-linear causality between economic growth and ASEAN countries’ EC from 1980 to 2016. They discovered a unidirectional causality from economic growth to EC in Malaysia, Thailand, and Indonesia; unidirectional causality from EC to economic growth in Singapore, and feedback causality between the Philippines’ variables. Ivanovski et al. (2020) investigated the impact of renewable and non-renewable energy on OECD (Organisation of Economic Co-operation and Development) countries’ economic growth from 1990 to 2015 via a nonparametric method. Their findings reveal that the latter has no adverse effect on economic growth. Also, both energy sources advance economic growth in non-OECD countries. The authors reiterated the need for the transition to renewables to consolidate growth and uphold environmental sustainability. Awodumi and Adewuyi (2020) examined the effect of EC on economic growth and emissions for oil-exporting countries in Africa from 1980 to 2015. Their findings suggest that EC exacts an asymmetric impact on both economic growth and CO2 emissions. However, the response of both variables to an increase in EC was mixed. Chen et al. (2020) applied a threshold model to examine the link between EC and economic growth for 103 countries from 1995 to 2015. They reported that EC had no significant/positive effect on growth in developed/OECD countries, respectively. However, the consumption of energy above/below a certain threshold will have a significant positive/negative effect on developing countries’ growth.Ali et al. (2020) examined the impact of EC and urbanisation on Nigeria’s economic growth after considering structural breaks. The FMOLS (fully modified ordinary least square) and DOLS (dynamic ordinary least square) results confirmed that EC added to economic growth, while urbanisation inhibited growth in Nigeria. Further findings revealed a one-way causality from urbanisation to economic growth and EC. This contradicts Nathaniel and Bekun’s (2020) earlier study which discovered no direction of causality; supporting the neutrality hypothesis, for Nigeria, Wang and Wang (2020) investigated the non-linear effect of EC on the OECD countries’ economic growth. They discovered that EC has a positive effect on economic growth. Nonetheless, this positive impact changes at different threshold levels.In summary, the relationship between EC and economic growth remains inconclusive with discrepancies in findings emanating from regional differences, peculiarities about countries, estimation techniques, and choice of variables. The majority of existing studies assumed a linear relationship between both, but recent studies have focused on nonlinearity because most economic variables do not exhibit a linear relationship. This study adds to this argument by examining the relationship between both variables in a non-linear framework for different blocs, regions, and organisations.Materials and methodsDataTwo different variables (real GDP per capita and EC) are used in this study. The time period, consistent with data availability, spans 1971 to 2014. This study has been carried out for BRICS, Next Eleven countries, the Economic Cooperation Organization (ECO), the Big Four in Western Europe, Asia-Pacific region, Group of Seven, Arab League, SAARC region, and World sample. EC is measured in kg of oil equivalent per capita, while real GDP per capita, our proxy for economic growth, is measured in constant 2010 USD. The data for both variables were obtained from the World Development Indicator (WDI 2020). These two variables are widely used in the energy-economic growth literature (see Ali et al. 2020; Meo et al. 2021; Nathaniel 2020). In all, 836 observations were used for conducting the relevant analysis. See Table A1 in Appendix for the list of countries adopted in this study.MethodsThe globalisation of the world’s economy has promoted interconnectivity and the transfer of shocks from one economy to another. Therefore, interdependence among countries is possible. The study proceeds with Pesaran’s (2004) cross-sectional dependence (CD) test. The CD test is of utmost importance to ensure unbiased estimates and estimator efficiency. We further applied the Im-Pesaran-Shin (CIPS) test to ascertain the stationarity properties of the variables. The CIPS test is robust amidst CD. Once variables are stationary, a co-integration test is required to ascertain the possibility of a long-run relationship. In this study, the Westerlund (2007) test investigates the presence of co-integration among the variables. Just like the CIPS test, the Westerlund (2007) test is also robust amidst CD and performs better than the traditional first-generation co-integration tests, including the Kao and Chiang (1999) and Pedroni (2000) co-integration tests.Unlike most previous studies, we examine the asymmetric impact of EC on economic growth (GR) via the non-linear autoregressive distributive lag (NARDL) model. The primary model for this study is given as:(1)GRt=φ0+φ1(ECt)+μt{GR_t} = {\varphi _0} + {\varphi _1}\left( {{EC_t}} \right) + {\mu _t}The NARDL model of Shin et al. (2014) can estimate short and long-run dynamics even when the sample size is small (Pesaran et al. 2001; Narayan, Narayan 2005; Meo et al. 2020). The non-linear form of the model is given as:(2)GRt=φ0+φ1(ECt+)+φ2(ECt−)G{R_t} = {\varphi _0} + {\varphi _1}\left( {EC_t^ + } \right) + {\varphi _2}\left( {EC_t^ - } \right)where ϕ0, ϕ1 and ϕ2 are the long-run parameters. The asymmetric effects of EC are incorporated by positive change ECt+EC_t^ + and negative changes ECt−EC_t^ - respectively, whereas ECt+EC_t^ + and ECt−EC_t^ - are the partial sums of positive and negative changes in EC, respectively. Now, to obtain the short-run effect, Eq. (1) can be re-specified as:(3)ΔGRt=∑k=1mθ1kΔGRt−k+∑k=1mθ2kΔECt−k+∂1GRt−1+∂1ECt−1+μt\Delta G{R_t} = \sum\nolimits_{k = 1}^m {{\theta _{1k}}} \Delta G{R_{t - k}} + \sum\nolimits_{k = 1}^m {{\theta _{2k}}} \Delta E{C_{t - k}} + {\partial _1}G{R_{t - 1}} + {\partial _1}E{C_{t - 1}} + {\mu _t}Eq. (3) is the specification that encompasses both long-run and short-run coefficients. Δ is the short-run coefficient while ∂1 and ∂2 are the long-run coefficients. We specify Eq. (4) to derive the asymmetric cointegrating.(4)γt=β+xt++β−xt−+μt{\gamma _t} = {\beta ^ + }x_t^ + + {\beta ^ - }x_t^ - + {\mu _t}where xt are the long term parameters of kx1 vector of regressors decomposed as (5)xt=xt++xt−{x_t} = x_t^ + + x_t^ - where xt+x_t^ + and xt−x_t^ - are the explanatory variables, decomposed into a partial sum of positive and negative changes. Eqs. (6) and (7) are the partial sums of positive and negative changes in EC.(6)EC+=∑i=1tΔECi+=∑i=1tmin(ΔECi,0)E{C^ + } = \sum\nolimits_{i = 1}^t {\Delta EC_i^ + = } \sum\nolimits_{i = 1}^t {\min \left( {\Delta E{C_i},0} \right)} (7)EC−=∑i=1tΔECi−=∑i=1tmin(ΔECi,0)E{C^ - } = \sum\nolimits_{i = 1}^t {\Delta EC_i^ - = } \sum\nolimits_{i = 1}^t {\min \left( {\Delta E{C_i},0} \right)} Replacing EC in Eq. (3) by EC+ and EC− gives us the asymmetric ARDL model:(8)ΔGRt=∂0+∑k=1mϕkΔGRt−k+∑k=1mϕkΔECt−k++∑k=1mϕkΔECt−k−+∂1GRt−1+∂2ECt−1++∂3ECt−1−+μt\matrix{ {\Delta G{R_t} = {\partial _0} + \sum\nolimits_{k = 1}^m {{\phi _k}\Delta G{R_{t - k}} + } \sum\nolimits_{k = 1}^m {{\phi _k}\Delta EC_{t - k}^ + + } } \cr {\sum\nolimits_{k = 1}^m {{\phi _k}\Delta EC_{t - k}^ - + {\partial _1}G{R_{t - 1}}} + {\partial _2}EC_{t - 1}^ + + {\partial _3}EC_{t - 1}^ - + {\mu _t}} \cr } From Eq. (8), the long-run impact of both positive and negative changes in EC are −∂2 / ∂1 and −∂3 / ∂1. One huge advantage of the NARDL technique is in its ability to accommodate variables with various levels of stationarity whether I (0) or I (1) or I (0) and I (1)Results and discussionsThe descriptive statistics of the regressors used in the NARDL model of this paper have been presented in Table 1 with mean, standard deviation, minimum, maximum, and coefficient of variation.Table 1Descriptive statistics.VariableObsMeanStd. dev.MinMaxCoef of variationWhole sampleGDPG3,4311.7154.945−64.9953.972.88EU3,4312,130.8862,369.31699.7118,178.1410.011BRICSGDPG1146.8234613.996938−7.388113.638330.585EU114842.0321559.7145279.3802,236.7300.664Next ElevenGDPG4182.494.55−30.1822.121.82EU418958.1930.1599.705,248.50.97ECOGDPG1141.406.05−30.1822.144.32EU1141,064.1680.6302.542,889.100.63Big Four in Western EuropeGDPG1521.661.78−5.465.871.07EU1524,053.8341.33,172.84,658.30.048Asia-Pacific regionGDPG5703.473.47−14.3513.641EU570944.231,122.899.714,083.81.189Group of SevenGDPG2661.662.10−7.946.331.26EU2664,770.71,970.92,203.88,455.50.41Arab LeagueGDPG4181.1485.10−25.6111.834.44EU4181,811.71,903.7235.186,871.81.05SAARC regionGDPG1903.012.35−7.3990.78EU190334.7119.699.71636.50.35BRICS – Brazil, Russia, India, China, and South Africa; ECO – Economic Cooperation Organization; SAARC – South Asian Association for Regional Cooperation; GDPG – GDP growth ; Obs – Observations.Source: own elaboration.In Table 1, we report the summary statistics that contain the mean (averages), observations, standard deviation, the minimum (Min) values as well as their maximums (Max). In our joint sample analysis, the coefficient of variation measures the heterogeneity of regressors. With a large coefficient of variation in the whole sample, we established the heterogeneity of regressors with GDP per capita having the more severe sample variances than energy use.Across the sample period, the averages across varying samples were considered within the study. In the whole model, the averages for energy use are large compared to GDP per capita averages. Apart from the fact that this statistical revelation points to ample energy use among the sample frame, the structural properties of data permit drawing credible inferences in a broad regional specific categorisation but not without controlling for the heterogeneity of regressors (Lewbel 2012). Across the BRICS, Next Eleven, Big Four in Western Europe, Asia-Pacific region, Group of Seven, and SAARC region economic classifications, the heterogeneity of regressors seems relatively weak and the averages of energy use seem predominate their large variances arrangement. Nonetheless, the Arab League shows heterogeneity well above the whole sample, and this implies its heterogeneity properties are greater than those observed in the entire sample. With the data extensive and fraught with the heterogeneity of regressors across samples and a pointer to nonlinearities among regressors, we proceed to test for common factor restrictions leading to the verification of the presence of cross-sectional dependence.The regional EC pattern and their growth-inducing strategies are almost identical with notable variances (Ahmed, Wahid 2011). These homogenous growth stabilisation approaches, induced by varying EC, ensure that economic indices are fraught with interdependence among cross-sections. It then becomes imperative to test for CD and consider controlling such disturbances to reach conclusions that are non-spurious (Pesaran 2015). We tried the null of no CD using the Pesaran (2004) CD test. In Table 2, Pesaran (2004) confirms the existence of CD with probability values <1%. The presence of CD calls for estimation procedures that control such disturbances. We proceed to estimate the second-generation panel stationarity test of CIPS of Pesaran (2004) because of its high-level precisions and capacity to control CD in panel data econometrics. In consonance with Phillips and Sul (2003), we followed this path to avert spurious outcomes when CD is not controlled.Table 2CD test results (Pesaran’s test 2004).VariablesCD-testP-valueAverage joint TMean ÏMean abs (Ï)GDP growth per capita21.4170.00037.950.060.18Energy use102.570.00037.950.270.54Notes: Under the null hypothesis of cross-sectional independence, CD ~ N (0, 1) P-values close to zero indicate data are correlated across panel groups.CD – cross-sectional dependence.Source: own elaboration.Beyond the descriptive revelation of heteroskedasticity and nonlinear properties among the regressors, it becomes apt and imperative to adopt the leading stationarity test to set a clear line of thought on the stationarity characteristics of the data set to reach non-spurious, policy and theory consistent empirical outcomes. Using the cross-sectionally augmented IPS test of Pesaran (2007) that considers CD when estimating its series, we found impressive stationarity outcomes. Table 3 reports the second generation panel stationarity test from the CIPS test of Pesaran (2007). We found regressors to be stationary at first difference across all the series I (1). These empirical outcomes establish an apparent need to uncover the covariance characteristics of the data. We proceed to estimate the Durbin-Hausman co-integration (Westerlund 2007) test to establish long-run cointegrating relations because of its relative sensitivity to CD.Table 3Unit root test results in the presence of cross-section dependence (Pesaran’s test 2007).VariablesLevelFirst differenceCIPS (−1)CIPS (−2)CIPS (−3)Critical value (5%)CIPS (−1)CIPS (−2)CIPS (−3)Critical value (5%)No intercept no trendGDP growth−1.231−3.761−3.774−1.52−5.963−5.935−5.934−1.52Energy use−1.134−1.267−1.268−1.52−5.097−4.940−4.883−1.52Only interceptGDP growth−1.107−4.107−4.107−2.08−5.977−5.977−5.977−2.08Energy use−1.606−1.606−1.606−2.08−5.272−5.272−5.272−2.08Linear trendGDP growth−2.48−4.470−4.453−2.56−6.206−6.161−6.181−2.65Energy use−1.836−1.936−1.893−2.56−5.537−5.461−5.410−2.65Notes: H0 (homogeneous non-stationary): bi = 0 for all i.Source: own elaboration.We prefer the Westerlund (2007) co-integration test to the Pedroni (1999) test due to its relative sensitivity to CD and slope homogeneity of regressors. Westerlund (2007) produces consistent estimates under the mild assumption. Since it disregards lag information about integrating series orders, it is widely applicable in a broader context. By permitting spatial correlation of CD, the Westerlund (2007) test controls for unobserved heterogeneity of regressors that characterises panel data estimation. To reach meaningful outcomes in the presence of cross-section dependence issues, the Westerlund (2007) test computes P-values robustly based on bootstrap iterations. Through the error correction component, cointegrating properties are revealed in the individual sample and across samples. Westerlund (2008) went further by computing the nullity and otherwise of the cross-section unit using the weighted averages and associative t-stats.Further evidence is sought into the cointegrating properties when the Westerlund (2007) tests establish the nullity of co-integration across panels by pooling the estimates in a comprehensive panel framework. The result presented in Table 4 shows that variables in different samples do not tend towards their long-run equilibrating position even when CD exists. Our result is robust at a 1% level of significance. We proceed to estimate the asymmetric co-integration in tandem with Hatemi-J’s (2020) in Table 5.Table 4Panel co-integration test (Westerlund’s test 2007).StatisticsModel with interceptModel with trendValueZ-valueP-valueValueZ-valueP-valueBRICSGt−4.369−4.9970.000−4.010−3.5040.000Ga−23.489−5. 2010.000−24.688−3.2290.001Pt−6.332−3.8470.000−5.953−2.6230.004Pa−22.190−7.0070.000−22.953−3.9610.000Next ElevenGt−1.7280.1830.573−2.421−0.2220.412Ga−4.9981.3070.904−7.7462.0810.981Pt−0.5954.2311.000−4.9152.3910.992Pa−0.2422.9800.999−2.9423.2410.999ECOGt−0.2432.9590.999−2.0330.7100.761Ga−1.5991.7640.961−3.2512.2310.987Pt0.7703.2971.000−1.7242.2130.987Pa0.5871.8790.970−2.3061.8720.969Big Four in Western EuropeGt−2.295−1.1520.125−2.696−0.8120.208Ga−12.127−1.8320.034−11.7060.0900.536Pt0.6973.6131.000−3.0921.2950.902Pa0.8282.2790.989−3.7191.7010.956Asia-Pacific regionGt−2.365−2.5300.006−2.421−0.2610.397Ga−12.430−3.7620.000−13.242−0.7010.242Pt−9.287−3.7020.000−8.0280.1760.570Pa−11.345−6.2070.000−11.655−1.7200.043Group of SevenGt−2.633−2.5200.006−2.831−1.5130.065Ga−13.333−3.0090.001−12.795−0.3050.380Pt−2.2931.5460.939−3.8711.9640.975Pa−3.1470.6460.741−4.8771.7500.960Arab LeagueGt−1.6290.5480.708−1.7772.4050.992Ga−100.936−57.1450.000−67.851−27.2350.000Pt−2.2232.5930.995−1.6306.1471.000Pa−1.9011.7400.959−1.8063.8551.000SAARC regionGt−2.532−1.8770.030−2.579−0.5840.280Ga−14.997−3.2260.001−13.943−0.6350.263Pt−5.671−2.4480.007−4.5310.2200.587Pa−15.506−5.6790.000−14.606−2.0690.019World sampleGt−2.715−1.8070.035−2.3310.0770.530Ga−9.223−0.6620.254−9.1360.7330.768Pt−2.751−0.2450.403−2.0541.8350.967Pa−2.6940.5990.726−4.0681.3740.915BRICS – Brazil, Russia, India, China, and South Africa; ECO – Economic Cooperation Organization; SAARC – South Asian Association for Regional Cooperation.Source: own elaboration.Table 5Asymmetric co-integration results of Hatemi-J’s test (2020).VariablesKao test resultsIm et al. (2003) unit root test results on residualsDecision regarding residualsArab LeagueY+, X+−2.4916 (0.0003)I (0), StationaryY+, X−−2.7552 (0.000)I (0), StationaryY−, X−−2.7677 (0.000)I (0), StationaryY−, X+−2.6207 (0.000)I (0), StationaryAsia-Pacific regionY+, X+−2.6571 (0.000)I (0), StationaryY+, X−−2.9797 (0.000)I (0), StationaryY−, X−−2.3041 (0.003)I (0), StationaryY−, X+−2.4447 (0.0001)I (0), StationaryBig Four in Western EuropeY+, X+−4.0673 (0.000)I (0), StationaryY+, X−−3.2235 (0.000)I (0), StationaryY−, X−−2.5821 (0.012)I (0), StationaryY−, X+−2.6298 (0.010)I (0), StationaryECOY+, X+−3.5681 (0.0004)I (0), StationaryY+, X−−2.5875 (0.025)I (0), StationaryY−, X−−2.8706 (0.013)I (0), StationaryY−, X+−2.7547 (0.013)I (0), StationaryGroup of SevenY+, X+−3.2938 (0.000)I (0), StationaryY+, X−−3.3661 (0.000)I (0), StationaryY−, X−−2.5256 (0.002)I (0), StationaryY−, X+−2.5917 (0.001)I (0), StationaryNext ElevenY+, X+−1.9333 (0.081)I (0), StationaryY+, X−−1.7222 (0.301)I (0), Stationary at 10%Y−, X−−2.0643 (0.021)I (0), StationaryY−, X+−2.0375 (0.026)I (0), StationaryWorld sampleY+, X+−2.3252 (0.000)I (0), StationaryY+, X−−2.6464 (0.000)I (0), StationaryY−, X−−2.6051 (0.000)I (0), StationaryY−, X+−2.4566 (0.000)I (0), StationaryECO – Economic Cooperation Organization.Source: own elaboration.The results of no co-integration between energy use and economic growth across the various samples led to the verification of hidden co-integration to reveal the series’s negative and positive cointegrating properties of components. Hatemi-J’s (2018) test for asymmetric co-integration was employed for this purpose. It revealed (2020) the hidden panel cointegrating properties. The initial analysis presents false outcomes of no co-integration by hidden co-integration, but further examination reveals co-integration between negative and positive components. This hidden composition leading to false cointegrating outcomes could result from unobserved factors in the non-cointegrating relations causing negative and components series to vary differently from idiosyncratic influences. Establishing a clear line of thought on the heterogeneous impact of energy use and economic growth asymmetric relationship informs about the need for this approach. Heterogenous energy pricing, the various institutional setup that moderates policy and content of reforms towards change, trade and export strategies, and so on, influence the output elasticity of the model in a non-linear and different manner. Thus, establishing the asymmetric properties of the model’s parameter will produce results that are most inclined to each region’s development objectives. Hatemi (2020) assumes the first difference integration test stationary of the series before decomposing regressors into the positive and negative components. It is also important to note that the residual from the series has to be stationary to establish regressors’ collinearity. The result in Table 5 reveals nonlinearities in the co-integration relationship between energy use and economic growth for all samples with residuals stationary at the level.Table 6 presents the linear ARDL result for the model. We relied on the Mean Group estimation procedure because of the high precision and relevance in establishing a clear line of thought on the energy use-growth relations. The convergence term is negative and statistically significant at a 1% level. The convergence terms give the speed of adjustment from long-run tendencies to short-run equilibrating positions. We have found long-run and short-run behavioural movements of the data set in the BRICS and SAARC regions. The Wald test has revealed the asymmetric properties of the model and affirmed no short-run and long-run asymmetric relations. In the search for hidden asymmetric properties of the energy use induced growth model, we proceed to estimate the non-linear ARDL model in consonance with Shin et al. (2014).Table 6Linear ARDL results.ΔYMG (BRICS)PMG (SAARC)ECT−0.008* (0.043)−0.0064577 (0.389)Long-run effect−0.829* (0.000)−1.040631* (0.000)Short run effect0.054* (0.001)0.0496974* (0.006)_cons4.914* (0.000)2.895211* (0.000)Hausman testH0 = PMGH1 = MG7.51** (0.0234)0.21 (0.6462)Wald test (long-run asymmetry)1.75 (0.1855)1.09 (0.2971)Wald test (short-run asymmetry)1.26 (0.2616)0.42 (0.5146)Log likelihood−366.9499Notes: The P-values are given [* at 1%, ** at 5%].ARDL – autoregressive distributed lag; BRICS – Brazil, Russia, India, China, and South Africa; PMG – Pooled Mean Group; SAARC – South Asian Association for Regional Cooperation; ECT – error correction term.Source: own elaboration.To assign numerical weights to the asymmetric relationship between energy use and economic growth within a global analytic framework, we report the non-linear ARDL result in Table 7. Beyond linear relations in the energy use-growth dynamic associations, we rely on the PMG to establish the regressors’ independent capacities to predict variations in the response variables. Following Shin et al. (2014), we unravel the energy use-growth dynamic relations’ asymmetric properties in an idiosyncratic capacity.Table 7Nonlinear panel ARDL.VariablesArab LeagueAsia-Pacific regionBig Four in Western EuropeECOΔY+ΔY−ΔY+ΔY−ΔY+ΔY−ΔY+ΔY−ECT−0.585* (0.000)−0.409* (0.000)−0.520* (0.000)−0.455* (0.000)−0.415* (0.000)−0.577* (0.000)−0.260** (0.033)−0.598 (0.000)eut+eu_t^ + −0.0080* (0.000)0.0047* (0.001)−0.0114 (0.000)0.014* (0.000)−0.005** (0.030)−0.002* (0.992)0.069 (0.009)−0.025 (0.012)eut−eu_t^ - 0.003** (0.011)0.0054* (0.002)0.0242* (0.000)−0.028* (0.000)0.025* (0.000)−0.011* (0.000)0.220* (0.001)−0.065 (0.000)Δeut+\Delta eu_t^ + 0.0011 (0.267)−0.0006 (0.974)−0.046** (0.052)0.045** (0.045)0.001 (0.914)−0.007 (0.393)0.036 (0.061)−0.021 (0.078)Δeut−\Delta eu_t^ - 0.952 (0.267)0.051 (0.242)0.006 (0.874)0.001 (0.974)0.0019 (0.595)0.028 (0.929)0.018 (0.490)0.002 (0.894)Δeut−1+\Delta eu_{t - 1}^ + 0.005 (0.703)−0.006 (0.686)−0.018 (0.146)0.018 (0.084)0.004 (0.318)−0.002 (0.456)0.023 (0.166)−0.020 (0.217)Δeut−1−\Delta eu_{t - 1}^ - −0.103 (0.290)0.105 (0.269)0.060 (0.619)−0.038 (0.689)−0.002 (0.579)0.004 (0.238)−0.012 (0.445)0.013 (0.424)Constant2.328* (0.005)−6.688 (0.229)−0.296 (0.862)−3.550** (0.013)−1.984 (0.173)−1.281 (0.368)3.920 (0.725)−2.563 (0.783)Hausman (χ2)H0 = PMGH1 = MG3.17 (0.2054)1.55 (0.461)3.56 (0.613)2.32 (0.192)Wald test (long-run asymmetry)3.320** (0.0367)5.69* (0.002)5.61** (0.020)7.76* (0.005)Wald test (short-run asymmetry)0.031 (0.142)0.29 (0.370)0.91 (0.213)0.12 (0.730)Log likelihood−1176.886−1072.3−193.7512−258.049Notes: P values are given in [], where eu is energy use; *1%; **5%; ARDL – autoregressive distributed lag; ECO – Economic Cooperation Organization; ECT – error correction term.VariablesGroup of SevenNext ElevenWorld sampleΔY+ΔY−ΔY+ΔY−ΔY+ΔY−ECT−0.427* (0.000)−0.5535* (0.000)−0.458* (0.000)−0.413* (0.000)−0.494* (0.000)−0.446* (0.000)eut+eu_t^ + −0.006* (0.000)0.021** (0.036)0.034* (0.007)−0.036* (0.000)−0.001** (0.013)0.001* (0.003)eut−eu_t^ - 0.021* (0.000)−0.013* (0.000)0.440* (0.000)0.066* (0.000)0.015* (0.000)−0.011* (0.000)Δeut+\Delta eu_t^ + −0.061 (0.466)−0.049 (0.710)−1.114 (0.264)1.167 (0.242)−0.004 (0.577)0.005 (0.544)Δeut−\Delta eu_t^ - 0.026 (0.243)−0.002 (0.290)1.337** (0.026)−1.217** (0.036)−0.005 (0.480)0.005 (0.463)Δeut−1+\Delta eu_{t - 1}^ + 0.011 (0.992)−0.009 (0.627)0.945 (0.257)−0.955 (0.259)−0.010 (0.397)0.011 (0.380)Δeut−1−\Delta eu_{t - 1}^ - −0.001 (0.626)0.009 (0.396)−0.339 (0.640)0.436 (0.556)−0.003 (0.794)0.004 (0.747)Constant−1.813 (0.075)−1.394 (0.145)−296.64* (0.006)−71.781 (0.121)1.881744 (0.236)−5.050* (0.001)Hausman (χ2)H0 = PMGH1 = MG2.17 (0.250)2.55 (0.601)4.17 (0.542)Wald test (long-run asymmetry)4.05** (0.0443)7.76* (0.005)5.76** (0.051)Wald test (short-run asymmetry)0.07 (0.7917)0.12 (0.730)0.83 (0.305)Log likelihood−284.11111.42−3620.302P values are given in (), where eu is energy use; *1%; **5%.Source: own elaboration.Table 6 rejected the nullity of covariance among economic growth and positive and negative components of energy use. With the Wald test outcomes not statistically significant at any significance threshold, we failed to reject the null of asymmetry in the short run while confirming the presence of asymmetry in the long run across the Arab League, Asia-Pacific region, Big Four in Western Europe, ECO, Group of Seven, Next Eleven and World sample. Across the heterogeneous sample leading to a global framework, the error correction components are appropriately signed (negative) and statistically significant at a 1% level of significance. By intuitions, the convergence from long-run behaviours is gradually adjusted to the short-run equilibrating positions. The sign and magnitude of the convergence terms established asymmetric properties in the energy use-growth relationship in a broad-based sample. For clarity, positive changes in our response variable ΔY+ denote growth expansion while negative changes ΔY− denote growth contraction. Furthermore, positive values of the parameters estimate eu+ that are associated with the positive value of the response variables ΔY+ imply growth expansion and negative values of parameters eu− associated with positive response variables ΔY+ imply economic contractions. Nonetheless, a positive value of parameter estimates eu+ with the negative response variable ΔY− implies economic recovery and negative values of parameter eu− estimated in a negative response variable ΔY− imply advance in economic contractions.The result from Table 7 was analysed based on regional sample categorisation, and the interpretations were presented logically and sequentially. With economic growth expected to react differently to heterogenous positive and negative energy shocks and produce non-linear results that are most inclined to each region’s development objective, our result follows a stratified and sample-specific pattern to reach conclusions regionally generalisable. The asymmetric short-run relations are observed for the Arab League, Asia-Pacific region, and Next Eleven. The short-run asymmetric properties in these samples depict energy shocks that could be a result of their extensive energy dependence. There is no observed asymmetric relationship between energy use and growth in the Big Four in Western Europe, ECO, Group of Seven, and World sample with just two of their coefficients significant at established thresholds. Short-run asymmetric relations are a testament to variations in cross-sectional domains that defines a country or regional groupings. The variations in the asymmetric properties across the sample categorisation could result from the heterogeneous level of reliance on energy use by different samples observed in this study.To establish linearity in reverse causative structure in the energy use-growth relation across varying samples, we reported the Dumitrescu and Hurlin (2012) panel Granger test in Table 8. The Dumitrescu and Hurlin (2012) panel Granger test accounts for the endogeneity of the regressor by computing the probability values of the numerical coefficient matrix. Using a large bootstrap iterative process, we accounted for cross-sections of the country and regional dimensions of the data set. We confirmed stationarity at the first difference before engaging the Panel Causality test in consonance with contemporaneous Granger causality test literature. We rejected the null hypothesis of Real GDP per capita does not Granger-cause energy use in the Big Four countries in Western Europe, ECO, Group of Seven and SAARC regions. By rejecting the nullity of the no Granger cause relations in the real GDP-energy use, we established uni-directional causality from real GDP to energy use in the Big Four countries in Western Europe, ECO, Group of Seven, and SAARC regions. In other related but distinct findings, we also rejected the null of Energy use does not Granger-cause real GDP per capita in the Group of Seven and SAARC region. This implies that a one-way causality exists from energy use to real GDP per capita in the Group of Seven and SAARC region. In summary, a bilateral causality exists between energy use and real GDP per capita in the Group of Seven.Table 8The result of Dumitrescu and Hurlin’s (2012) panel causality test.Null hypothesisW-statZ-barP-valueLagArab LeagueReal GDP per capita does not Granger-cause energy use3.502375.134793.E−071Energy use does not Granger-cause real GDP per capita1.888561.738470.08211Asia-Pacific regionReal GDP per capita does not Granger-cause energy use2.580860.748830.45402Energy use does not Granger-cause real GDP per capita5.586115.829656.E−092Big Four in Western EuropeReal GDP per capita does not Granger-cause energy use3.369032.927180.0034*1Energy use does not Granger-cause real GDP per capita1.344340.357670.72061BRICSReal GDP per capita does not Granger-cause energy use0.83674−0.248120.80401Energy use does not Granger-cause real GDP per capita2.777341.884700.05951ECOReal GDP per capita does not Granger-cause energy use16.213616.65190.0000*1Energy use does not Granger-cause real GDP per capita2.497381.577010.11481Group of SevenReal GDP per capita does not Granger-cause energy use10.430615.72750.0000*1Energy use does not Granger-cause real GDP per capita2.988223.232980.0012*1Next ElevenReal GDP per capita does not Granger-cause energy use4.696907.648712.E−141Energy use does not Granger-cause real GDP per capita5.205408.718880.0000*1SAARCReal GDP per capita does not Granger-cause energy use1.02665−0.050860.95941Energy use does not Granger-cause real GDP per capita9.7771912.36510.0000*1World sampleReal GDP per capita does not Granger-cause energy use5.7498426.76880.0000*1Energy use does not Granger-cause real GDP per capita2.237456.709962.E−111Notes: We computed the probability values using large iterative bootstrap identifications; W-Stat statistic presents the averages from cross-sections of N standard individual Wald statistics of Granger non-causality tests. The Z-bar statistic is a standardised statistic.BRICS – Brazil, Russia, India, China, and South Africa; ECO – Economic Cooperation Organization; SAARC – South Asian Association for Regional Cooperation.Source: own elaboration.Conclusion and policy suggestionsThis paper investigates the following questions. Is the relationship between economic growth and EC asymmetric? What if we investigate the same connection across a different group of countries? Suppose the relationship between the variables of interest is asymmetric, does it prevail both in the short run and in the long run? Does the impact of EC on economic growth enhance or contract growth across a different group of countries and in the whole sample? These and other relevant questions are answered while examining the impact of EC on economic growth in a non-linear panel ARDL framework.Using data from 1977 to 2014 for 85 countries (as a whole sample) and seven similar groups of countries, the study first applied the second generational panel unit root test due to cross-sectional dependency. To check for the existence of a long-run relationship between EC and economic growth, Hatemi-J’s (2018) test for asymmetric co-integration was employed for this purpose. To determine whether the impact of EC on economic growth is growth-enhancing or growth contracting both in the short run and in the long run, we use the non-linear ARDL model. Using these methods, some interesting results are found.This paper finds a linear relationship between EC and economic growth in BRICS and SAARC countries while an asymmetric relationship in the rest of the countries, including the whole sample of 85. Furthermore, among those groups of countries where the relationship is asymmetric, we have only found a long-run asymmetry. This confirms the possible impact of changes in EC on economic growth in the short run but not in the long run.The asymmetric relationship between the subject variables means that the impact of EC on economic growth varies on the phase/state of the economic cycle. In this context, our results suggest that positive shocks on energy use have a positive and significant effect on economic growth only in ECO and the Next Eleven in the periods of economic expansion. This means that in ECO and the Next Eleven groups of countries, a positive shock to energy use is growth-enhancing while in the rest of the group of countries it is growth-contacting. Moreover, the study results suggest that economic recovery is the case in the Arab League, Asia-Pacific region, Group of Seven, and the whole sample. However, a negative shock to EC in the Group of Seven, Asia-Pacific region, Big Four in Western Europe, ECO, and the whole sample worsens the economic contraction. In the short run, the results are mostly insignificant.Based on the findings, the relationship between economic growth and EC is asymmetric, and this paper suggests adjusting the energy supply and energy demand in a group of countries where a positive shock to energy use is growth-enhancing. Similarly, EC could be encouraged by expanding industrialisation, urbanisation, and transport in countries where a positive shock to EC leads to economic recovery. However, in a group of countries where a positive shock to EC worsens economic growth, EC should be discouraged while encouraging the use of close substitutes of energy, lowering the use of private vehicles, promoting solar and water energy, reducing an excess use of luxurious goods. Although we have tried to carry out this research carefully, there may be some methodological pitfalls. Thus, further research may be carried out to solve these shortcomings.

Journal

Quaestiones Geographicae – de Gruyter

Published: Jun 1, 2022

Keywords: energy consumption; economic growth; NARDL; co-integration; C33; O44; O50; Q43w

There are no references for this article.