Electricity consumption and gross domestic product in an electricity community : Evidence from bound testing cointegration and Granger-causality tests

This study probes the nexus between electricity consumption and gross domestic product (GDP) for the electricity community of Togo and Benin using ARDL bounds testing approach of cointegration. Long-run equilibrium has been established among these variables for Benin. The study further establishes long and short-run Granger causality running from GDP to electricity consumption for Benin and short-run Granger causality running from GDP to electricity consumption for Togo. The results of the cointegration test and the causality reflect better the Benin and Togo economies that are less dependent on electricity. The absence of causality running from electricity consumption to GDP implies that the very low electricity consumption in both countries do not allow them to take advantage of the benefits that electricity energy brings in terms of adoption of new technology as well as technical efficiency.


INTRODUCTION
Benin and Togo stand in Sub-Saharan Africa as a successful example of electricity energy cooperation.Indeed, since 1973, Benin and Togo cooperate with regard to the production of the electric energy through the electric Community of Benin (CEB).The Electric Community of Benin (CEB) is an international public establishment for electric energy development in Benin and Togo.It has been created by the international agreement of July 1968 instituting the electricity code between the Republic of Benin and the Republic of Togo.From this agreement, the CEB has the monopole of the production, the transport and the importations/ exportations of the electric energy within the two countries.The regulation of the electricity sector as well as the development of the sector is managed by the CEB.
In spite this example of integration, the electricity sector of both countries is characterized by outages and shortage.The total capacity installed by the CEB in 2006 is 1309 Gwh of which 664 Gwh and 645 Gwh respectively for Benin and Togo.
The total electricity capacity has risen around 200% from 1990 to 2006 for both countries (CEB, 2008).However, this increasing in electricity capacity proves to fail to satisfy the demand in these countries.On consumption side, electricity consumption per capita is about 47.82 Kwh per year for Benin and about 89.37 Kwh for Togo during this last two decades, those rates are among the lowest in sub-Saharan Africa countries1 .Studies on the causal relationship between electricity consumption and economic growth occupy a substantial portion of economic literature.These studies have undergone extensively after the oil embargos in 1970s, and recently in developing countries since access to modern energy like electricity is found to be important for poverty reduction.The central issue has been whether economic growth stimulates consumption of energy or if the energy consumption itself is a stimulus for economic growth via the indirect channel of effective aggregate demand.Blanchard and Gali (2007) and Brown and Yücel (2002) among others provide a survey of the theory and evidence on the macroeconomic impact of energy prices.The relationship between energy consumption and economic growth was also explored.As the level of electricity consumption can be taken to signal, in addition to economic growth, the level of socioeconomic development of a country, more recent studies have focused on examining the causality relationship between electricity consumption and economic growth (Belloumi, 2009;Ghosh, 2009;Akinlo, 2009;Narayan and Singh, 2007;Wolde-Rufael, 2006, 2009;Ferguson et al., 2000).However, the debate on the nature of the relationship is far from being settled.On Benin and Togo, few studies try to probe the nexus between electricity production/ consumption and economic growth.Wolde-Rufael (2006, 2009) examined the cointegration and Granger causality using Pesaran et al. (2001) procedure for Benin and Togo.Wolde-Rufael (2006) failed to establish a cointegration relationship between GDP (Gross domestic product) and electricity consumption for Benin but found a causal relation running from electricity to growth.Wolde-Rufael (2009) by using a multivariable analysis demonstrates a causal relation from electricity to growth for Benin and a bidirectional relation for Togo.The present study tempts to account for this fact and examines the causal relation between the consumption of electricity and the GDP in both countries.The rationality of this study stands on the fact that despite the strong increasing in the consumption of electricity in both countries during this last decade, economic growth is recorded in a downtrend, demonstrating that the debate on the causality relation between electricity and growth is not yet closed for both countries.The difference between this study and Wolde-Rufael (2006, 2009) is that it uses Narayan (2005) critical values table for the ARDL bound testing cointegration rather than Pesaran and al. (2001) used by Wolde-Rufael (2006, 2009).Narayan (2005) argued that since Pesaran et al. (2001) critical values table are based on a large sample sizes, they cannot be used for small sample sizes, like the one used by Wolde-Rufael (2006, 2009) and in this study.To meet its goal, this study uses the autoregressive distributed lag (ARDL) bounds testing approach of cointegration developed by Pesaran et al. (2001).ARDL approach of cointegration has become popular in energy market analysis (Narayan and Smyth, 2005;Ghosh, 2009;Odhiambo, 2009) and is intensively used in other disciplines like macroeconomics, applied finance, education economics, tourism, etc.Some of the articles in these areas include those of Katircioglu (2009), Muchapondwa and Pimhidzai (2009) and Narayan (2005).
For the rest of the study, a brief literature review on electricity consumption and economic growth is provided; the methodology and the results obtained are presented.And lastly, the concluding remark and the policy suggestions.

BRIEF LITERATURE REVIEW
The role of electricity consumption in economic growth has produced diverse results across time and countries.Some empirical studies have identified a causal relation running from electricity consumption to economic growth (Akinlo, 2009;Wolde-Rufael, 2006, 2009), while few others reported the opposite (Wolde-Rufael, 2006;Jumbe, 2004).Few others have provided evidence of bidirectional causality between the electricity consumption and economic growth (Mozumder and Marathe, 2007;Wolde-Rufael, 2006, 2009).Yet, a handful of studies have reported neutral causal relation between electricity consumption and economic growth (Mozumber and Marathe, 2007;Jumbe, 2004;Asafu-Adjaye, 2000).The findings from the studies vary not only across countries, but depend also on methodologies within the same country (Akinlo, 2009;Wolde-Rufael, 2006, 2009;Soytas and Sari, 2003).Table 1 provides a short synthesis of studies probe the nexus electricity consumption and growth.

Data description
Annual data on real GDP (in constant local currency) and electricity consumption have been collected from the World Development Indicator (2011) produced by the World Bank and the Electricity Community of Benin (CEB), for the time span 1973-1974 to 2008-2009.Electric power consumption measures the production of power plants and combined heat and power plants less transmission, distribution, and transformation losses and own use by heat and power plants.Table 2 gives the summary statistics of each of the variables used in the analysis.All the variables are in logarithm transformation.

Cointegration
ARDL bounds testing approach has been employed to examine both short run and long run elasticities among the variables.An ARDL model is a general dynamic specification, which uses the lags of the dependent variable and the lagged and contemporaneous values of the independent variables, through which the short-run effects can be directly estimated, and the long-run equilibrium relationship can be indirectly estimated (Ghosh, 2009).Unlike other single-equation estimation frameworks, it offers explicit tests for identifying a unique cointegration vector rather than assuming it.However, the ARDL approach is only valid when there is a unique cointegration vector.ARDL technique involves estimating the following unrestricted error-correction model: Where, ∆ is the first difference operator.
The existence of a long run relationship is independent of the variables order of integration provided that none of the variables are integrated of order 2.
The test statistics is the standard F-Statistic ‫ܨ(‬ ).However, the Ftest has a non-standard distribution which depends upon (i) whether variables included in the ARDL model are I(0) or I(1); (ii) number of regressors; (iii) whether the ARDL model contains an intercept end/or a trend; and (iv) the sample size.Two sets of critical F values have been provided by Pesaran and Shin(1999) and Pesaran et al. (2001) for large samples and by Narayan (2005) for sample size ranging from 30 to 80, where one set assuming that all variables in ARDL model are I(1) and another assuming that all variables are I(0) in nature.It is important to note that the critical based on large sample size deviates significantly from that of small sample size.Narayan (2004aNarayan ( , 2004b) ) compares the critical values (CV) generated with 31 observations and the critical values reported by Pesaran et al. (2001) and finds that the upper bound CV at the 5% significance level for 31 observations with 4 regressors is 4.13 while the corresponding CV for 1000 observations is 3.49, which is 15.5% lower than the CV for 31 observations.We then extract appropriate CVs from Narayan (2005).
The critical value has lower bound ‫ܨ(‬ ) and upper bound ‫ܨ(‬ ).If ‫ܨ‬ < ‫ܨ‬ no cointegration relation exists and when ‫ܨ‬ > ‫ܨ‬ a cointegration relation exists.However, when ‫ܨ‬ < ‫ܨ‬ < ‫ܨ‬ , inference remains inconclusive under such circumstance, a knowledge of the order of integration of the underlying variables is needed to proceed further.

Granger causality
If we do not find evidence for cointegration among the variables, then the specification of the Granger causality test will be a vector autoregression (VAR) in first difference form.However, if we find evidence for cointegration then we need to augment the Grangertype causality test model with a one period lagged error correction term.This is an important step because Engle and Granger (1987) caution that if the series are integrated of order one, in the presence of cointegration VAR estimation in first differences will be misleading.Granger-causality test is a convenient approach for detecting causal relationship between two or more variables.In our case, test for Granger causality can be done through following equations:

‫ݐ1ݑ‬
(3) Where ߚ's are parameters to be estimated, ‫ݑ‬ ௧ 's are the serially uncorrelated error terms, and ‫ܶܥܧ‬ ௧ିଵ is the error-correction term (ECT).The F-statistics on the lagged explanatory variables of the ECM indicates the significance of the short-run causal effects.The t-statistics on the coefficients of the lagged error correction term indicates the significance of the long-run causal effect.The lag length p is based on Schwarz-Bayessian (SBS) and/or Akaike Information Criteria (AIC).Letting ‫ܯ‬ ଵ = (ߚ ଵଶଵୀ⋯ୀ ߚ ଵଶ), and ‫ܯ‬ ଶ = (ߚ ଶଵଵୀ⋯ୀ ߚ ଶଵ ) the causality test is carried out by generating F statistics to establish whether the null hypotheses can be accepted or rejected.For equation (3) this amounts to ‫ܪ‬ : ‫ܯ‬ ଵ = 0 and for equation (4) it is ‫ܪ‬ : ‫ܯ‬ ଶ = 0.

Cointegration tests
The cointegration test under the bonds framework involves the comparison of the F-statistics against the critical values, which are generated for specific sample sizes.Using equation ( 1) and ( 2), the calculated F- statistics are reported in Table 3.When electricity consumption is the dependent variable for Benin, the calculated F-statistics ‫ܨ‬ ாா ‫43.5=)ܲܦܩ|ܥܧܮܧ(‬ is higher than the upper bound critical value of 5.080 at the 10% level.However if Benin's GDP is the dependent variable over the same period , the calculated F-statistics ‫ܨ‬ ீ ‫88.3=)ܥܧܮܧ|ܲܦܩ(‬ is lower than the lower bound critical value at 10% level (4.290).This suggests that the null hypothesis of no cointegration cannot be supported for Benin when electricity is the dependent variable.
When electricity consumption is the dependent variable for Togo, the calculated F-statistics ‫ܨ‬ ாா ‫50.4=)ܲܦܩ|ܥܧܮܧ(‬ is lower than the lower bound critical value of 4.290 at the 10% level.Likewise if Togo's GDP is the dependent variable over the same period , the calculated F-statistics ‫ܨ‬ ீ ‫49.0=)ܥܧܮܧ|ܲܦܩ(‬ is lower than the lower bound critical value at 10% level (4.290).This suggests that the null hypothesis of no cointegration between growth and electricity consumption cannot be rejected for Togo.
Once a long-term relationship has been established for Benin, in the next stage, a further two-step procedure is carried out.In the first step, the optimal order of lags in the model are selected based on Schwarz-Bayessian information criteria; we ensured that residuals do not suffer from serial correlation, the LM test for serial correlation is used in this regard and in the second step, the selected model (equation 3) is estimated through ordinary least-square technique.
The existence of a long-run relationship among electricity consumption and GDP suggests that there must be Granger causality at least in one direction.Table 4 reveals the results of the short and long run Granger causality within ECM framework.In the short-run there is no significant variable, which imply that GDP doesn't cause electricity consumption in the short-run.Turning to the long-run causality result, the coefficient of the lagged error-correction term is statistically significant at 1% level with correct sign implying that the series is non-explosive and long-run equilibrium is attainable.The speed of adjustment is 88.05%.The long-run elasticity of GDP is 1.80; in other words, an increase in GDP of 1 percentage point in the long-run increases electricity consumption by 1.80%.

Granger causality
To complete the obtained results, Granger causality test were also carried out.While the bounds test for cointegration does not depend on pre-testing the order of integration, all variables need to be integrated of order one in order to apply the Granger causality test.To determine the order of integration, the work applies the Augmented Dickey-Fuller (ADF) unit root tests.All the series are I(1) we can then proceed to the Granger causality test.For Benin since there is a long-run relation between the variable we use equation ( 3), for Togo, we do not find any evidence for cointegration among the variables then the specification of the Granger causality test will be a vector autoregression (VAR) in first difference form (Narayan and Singh, 2007;Engle and Granger, 1987); that is equations ( 3) and ( 4) without the error-correction term.
Beginning on Benin, the existence of a cointegration relationship between electricity consumption and GDP suggest that there must be Granger causality in at least one direction, but it does not indicate the direction of temporal causality between the variables.We examine both short-run and long-run Granger causality in this section.The short-run causal effect is obtained by the Ftest of the lagged GDP variable, while the t-statistics on the coefficient of the lagged error-correction term in equation ( 3) indicates the significance of the long-run causal effect.In the short-run, GDP is significant at 10%, this implies that GDP Granger cause electricity  Turning to Togo, we estimated Eq(3) and Eq(4) without the error-correction term.We use the LR test to determine the appropriate lag.In Eq(3) the lagged GDP variable is significant at 5%, this imply that in the shortrun, GDP Granger causes electricity consumption in Togo.
The common fact between these two countries is that for both countries GDP explained the electricity consumption supporting the argument that it is the economic expansion that determines the electricity consumption.For Benin this relation is weak in the short run while strong in the long-run while for Togo.This difference could be explained by the significative difference in the electricity consumption per capita in both countries.In Benin it's equal to 47.82 Kwh and about 89.37 Kwh in Togo.This suggests that the economic expansion is likely to have more training effect on electricity consumption in Togo than in Benin.

Concluding remarks and policy suggestion
In this paper, we use Pesaran et al. (2001) cointegration technique and Granger causality tests to investigate the long-run and causal relationship between real GDP and electricity consumption for an electricity community: CEB (Benin and Togo).The study detected long run cointegration relationship between electricity consumption and GDP for Benin, and causality for both countries.The results of the cointegration test and the causality reflect better the Benin and Togo economies that are less dependent on electricity.The economies of these two countries are heavily dominated by the agriculture sector contributing more than 33% to the total GDP.The result suggests that a permanent rise in GDP may cause a permanent growth in electricity consumption.This shows that growth in this line is more beneficial to people in urban area since, so far, electrification is a matter of urban area in both countries (in both countries about 52% of the urban population are electrified when less than 2% are in rural area; where the majority of the agricultural production in undertaken).These results suggest clearly that Benin and Togo to engage in pro-poor growth strategies so that the fruit of growth can be reoriented to electrify rural zone that contribute to more than 33% to the GDP.

Table 1 .
Synthesis of some studies of GDP -electricity consumption nexus.

Table 2 .
Summary of the variables.

Table 4 .
Results of Granger causality tests.