Energy consumption and economic growth: assessing the evidence from Greece

Energy Economics 24 Ž2002. 319᎐336

Energy consumption and economic growth: assessing the evidence from Greece George Hondroyiannis a,U,1, Sarantis Lolos b, Evangelia Papapetrou c a

Bank of Greece, Economic Research Department and Harokopio Uni¨ ersity, Athens, Greece b Panteion Uni¨ ersity, Athens, Greece c Uni¨ ersity of Athens and Bank of Greece, Economic Research Department, Athens, Greece

Abstract This paper attempts to shed light into the empirical relationship between energy consumption and economic growth, for Greece Ž1960᎐1996. employing the vector error-correction model estimation. The vector specification includes energy consumption, real GDP and price developments, the latter taken to represent a measure of economic efficiency. The empirical evidence suggests that there is a long-run relationship between the three variables, supporting the endogeneity of energy consumption and real output. These findings have important policy implications, since the adoption of suitable structural policies aiming at improving economic efficiency can induce energy conservation without impeding economic growth. 䊚 2002 Elsevier Science B.V. All rights reserved. JEL classifications: Q43; C52 Keywords: Energy consumption; Economic growth; Granger temporal causality

1. Introduction The energy crises in the 1970s and the unprecedented high levels of energy prices, especially oil, which had a detrimental effect on growth, called for the implementation of energy conservation processes. Indeed, most of the industrialized countries managed to have gradually curtailed energy requirements. 1

The views expressed in this paper are those of the author and not those of the Bank of Greece. Corresponding author. Bank of Greece, Economic Research Department, El Venizelou 21, 102 50 Athens, Greece. Tel.: q3010-320-2429; fax: q3010-323-3025. E-mail address: [email protected] ŽG. Hondroyiannis.. U

0140-9883r02r$ - see front matter 䊚 2002 Elsevier Science B.V. All rights reserved. PII: S 0 1 4 0 - 9 8 8 3 Ž 0 2 . 0 0 0 0 6 - 3

320

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

Since the end of the 1970s, there has been fairly extensive empirical research interest on the temporal causality between energy consumption and economic growth or employment, with neither conclusive results nor persuasive explanations. The research effort, very much facilitated by the ‘newly’ developed statistical techniques Že.g. Sim’s test., aimed at investigating whether economic growth takes precedence over energy consumption or if energy consumption can boost economic growth. The advance of econometric techniques in recent years stimulated further empirical research on the energy consumption᎐economic growth debate, still with elusive results. As regards to policy issues, the question is whether the adoption of energy saving processes is a stimulus to growth or the opposite, the relevant literature referring to the possibility or feasibility for the adoption of energy saving processes. In this paper, we take a fresh look at the empirical evidence on the relationship between economic growth and energy consumption with a view of offering suggestions about how the issue may be addressed in future. We feel that the question of energy growth is more usefully analyzed if it is placed in a broader perspective, rather than within the partial consideration of energy conservation issues. The case of Greece, a medium size country, serves as an example in our empirical investigation and the conclusions drawn could be useful for the analysis of other mediumsized economies. The paper evaluates empirically the dynamic interaction between energy consumption, real output and the price level and tests for the endogeneity of these three variables included in the vector, utilizing the vector error-correction models ŽVECM. technique. The relevance of endogenous energy consumption is investigated and its dynamic response to structural shocks is estimated. Furthermore, the empirical analysis distinguishes between three categories of energy consumption: total; residential; and industrial energy consumption, all treated independently. The empirical results showed that total energy consumption is an endogenous variable also affecting economic growth. In addition, economic efficiency, as reflected in price developments, is a determining factor of both energy consumption and income behavior. The same applies to industrial energy consumption, while energy consumption for residential uses behaves independently of price and income developments. As regards to policy implications, policies aiming at improving economic efficiency via structural reforms Žimplementation of adjustment policies, enforcement of endogenous growth mechanisms. boost economic growth and result in lower energy requirements. At the same time, these policies induce energy conservation, positively affecting economic growth. Hence, structural policies are a stimulus to both economic growth and energy saving. This being the case, energy conservation does not impede economic growth. The paper proceeds as follows. Section 2 briefly reviews previous empirical work on the relationship between energy requirements and growth, also considering the developments in Greece. Section 3 deals with methodological issues and the data used in the empirical analysis, while in Section 4 the empirical evidence is presented. Finally, in Section 5, the conclusions of the analysis are summarized and policy implications are discussed.

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

321

2. Energy consumption and economic growth 2.1. Empirical e¨ idence The bulk of the literature has so far offered conflicting and inconsistent results concerning the causal relationship between energy consumption and economic growth. This might be attributed first to the different institutional, structural frameworks and the policies followed of the countries under consideration and second to methodological differences. Granger’s and Sim’s tests, that have been used extensively by many researches to detect causality, have undergone major criticism as the time series affected the sensitivity of both tests. Also, most studies have assumed that the time-series data are stationary and, therefore, have used inappropriate estimation techniques. The pioneer work in this area was by Kraft and Kraft Ž1978., who found a unidirectional causality running only from gross national product ŽGNP. to energy consumption for the period 1947᎐1974 for the United States. The results implied that energy conservation policies could be implemented without affecting GNP growth. However, Akarca and Long Ž1980. failed to obtain similar results when they shortened the data sample of Kraft and Kraft Ž1978. implying that the time period chosen might have influenced the results. Yu and Hwang Ž1984. updated the US data for the period 1947᎐1979 and found no casual relationship between energy consumption and GNP. Yu and Choi Ž1985., using data from five countries, confirmed the absence of causality between GNP and total energy consumption for the US, UK and Poland but causal relationship from GNP to energy consumption was detected for South Korea and the opposite for Philippines. Erol and Yu Ž1987. found mixed results using a sample of six industrialized countries. Although the empirical evidence seems to support the neutrality hypothesis for the post-war US data, the evidence on other countries is mixed. Hwang and Gum Ž1992. found bi-directional causality in Taiwan. Recently, Masih and Masih Ž1996, 1997. in a multivariate framework examined the relationship between total energy consumption and real income of Asian Economies such as: India; Pakistan; Malaysia; Singapore; Indonesia; Philippines; Korea; and Taiwan. Energy consumption was found to be neutral with respect to income for Malaysia, Singapore and Philippines, unidirectional causality existed from energy consumption to GNP for India, exactly the reverse for Indonesia and mutual causality was present for Pakistan. For Taiwan, their results are in line with Hwang and Gum Ž1992.. Recently, Yu and Jin Ž1992. used the Engle᎐Granger testing procedure to study whether energy consumption and income are cointegrated and to determine the presence of a long-run relationship between these two variables for the United States over the period 1974.1᎐1990.04. They found evidence against a long-run relationship among the variables. Stern Ž1993. used a VAR model of four variables and performed standard causality tests. Although there was no evidence that gross energy use Granger caused GDP, a measure of final energy use adjusted for changing fuel composition was found to Granger cause GDP. However, these

322

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

studies suffered from various methodological shortcomings that might have resulted in spurious results. Recent advances in multivariate co-integration approaches, such as the Johansen᎐Juselius maximum likelihood approach Ž1990, 1992., have allowed to test for the presence of all long-run relationships that the Engle᎐Granger tests might have failed to unveil. Furthermore, besides short-run causality Žthrough the differenced variables. an additional channel of Granger causality emerges from any long-run co-movements that the variables may share, that was ignored by the standard causality tests ŽGranger, 1988.. Empirical studies for the Greek economy have investigated energy demand and its association to output and prices. However, these studies have not addressed the problem of the possible interdependence of output and energy consumption and its policy implications. In particular, Samouilidis and Mitropoulos Ž1984., investigating the relationship between economic growth and energy demand over the 1960s and 1970s, observed falling income and price elasticities of energy demand, while Mitropoulos et al. Ž1982. using Kalman filter techniques supported the view that elasticities behave as a cluster against energy demand. Donatos and Mergos Ž1989. estimated energy demand and price equations for the period 1963᎐1984 and came to the conclusion that energy demand is rather inelastic with respect to prices. Similar results for energy demand in Greece are reached by Christodoulakis and Kalyvitis Ž1997. and by Zonzilos and Lolos Ž1996. analyzing the recent time period.2 2.2. De¨ elopments in Greece The pattern of development of energy consumption in Greece is closely related to that of output ŽTable 1.. In particular, over the 1960s and until the first energy crisis in 1973, the Greek economy experienced exceptionally high rates of economic growth Ž7.7% on average. as a result of the industrialization process, with the industrial sector gaining an increasing share in GDP. At the same time, the average rate of increase of total energy consumption Ž12.3%. and especially industrial energy uses Ž14.3%. exceeded considerably that of output. In the rest of the 1970s and until the second energy shock in 1979 the pace of increase of GDP lowered, so did energy consumption. Over the 1980s and the early 1990s, the economic activity in Greece registered low average growth rates Ž1.6%. and industrial output declined, a fact reflected in the energy consumption pattern. Finally, since the mid 1970s, a rise of energy consumption in transport and residential uses has been observed, faster than that of total energy consumption, mostly related to a general improvement of the standard of living in Greece Žincrease of private cars, extended use of electric appliances, etc...3 Greece registers a relatively high degree of energy requirements in production, making the economy particularly vulnerable to energy-price shocks. Despite the 2 The demand for energy in Greek manufacturing is analyzed extensively. See the recent papers by Christopoulos Ž2000. and Calogirou et al. Ž1997.. 3 On this point, see Zonzilos and Lolos Ž1996..

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

323

Table 1 Energy consumption, economic growth and inflation in Greece Žannual growth rates. 1960᎐1973 Real GDP Energy consumptiona ᎐Total ᎐Industrial ᎐Residential ᎐Transport etc. Consumer price index

1974᎐1979

1980᎐1996

7.7

3.7

1.6

12.3 14.3 10.9 11.7 3.3

3.9 4.7 1.2 4.5 14.2

2.2 y0.1 3.2 3.1 16.7

Source: Energy Balances in OECD Countries; International Energy Agency Žvarious issues.. Energy consumption in Mteos.

a

oil-price shocks in the 1970s, energy intensity in Greece, measured by the ratio of energy requirements to total output, increased sharply, owing to the absence of an effective energy saving policy. However, energy intensity slowed down somewhat in later years. This contrasts with developments in most other countries where energy consumption gradually declined in relation to output.4 With regard to energy forms in Greece, the absence of natural gas and the small share of solid fuel Žcoal. 5 allow for an especially high contribution of oil Žabout 75%. and electricity Žjust under 20%. in energy consumption. Also, the pattern of energy consumption of the industrial and the commercial᎐residential sectors has been shifted gradually towards electricity facilitated by the state-regulated prices. Furthermore, Greece has always been highly dependent upon energy imports, mainly oil, which accounted for 79% of total energy requirements in 1996. As a result, prices in Greece closely follow energy prices. In particular, over the second half of the 1970s, the GDP deflator increased with average annual rates of 19% and by 15% during the 1980s, while the Consumer Price Index rose by 16% and 17%, the price of electricity by 18% and 17% and the price of oil by 23% and 15% over the corresponding periods. Besides, empirical evidence strongly suggests that price developments in Greece have been closely associated with the macroeconomic and microeconomic policy stance.6 The persistence of macroeconomic imbalances and the inefficient functioning of the economy over the 1980s allowed for undesirable high inflation rates. On the contrary, the successful stabilization and liberalization effort initiated in the early 1990s managed to reduce economic imbalances, to improve efficiency and substantially decrease the inflation rate Ž1992: 15.9%; 1998: 3.7%.. The policy mix included: Ža. the adoption of the Greek Convergence Program aiming at achieving

4

For instance, total primary energy requirements over GDP in Greece increased from 0.51 in 1973 to 0.58 by the end of the 1980s, while in OECD countries it decreased from 0.54 to 0.41 respectively. 5 Coal is mainly consumed by the cement industry. 6 For a discussion on these issues see Lolos Ž1998..

324

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

the Maastricht criteria; and Žb. the implementation of a structural adjustment program, the Community Support Framework, co-financed by the EU.7

3. Methodological issues and data The empirical analysis tests for the endogeneity of energy consumption and economic growth. Prices are included in the specification for two reasons. Firstly, because they play a crucial role in affecting energy consumption, since the Greek economy is highly dependent on energy and energy imports.8 Secondly, and most importantly, as a proxy for the degree of the efficient functioning of the economy. In other words, we assign a broader interpretation to the role of prices, so far not explicitly recognized in the related body of literature. An improvement in economic efficiency through structural measures is reflected to price developments and to economic growth. At the same time the implementation of energy conservation policies cannot be disassociated from improvements in economic efficiency and technological progress. Also, the analysis distinguishes between different categories of energy consumption: total; industrial; and residential consumption, in order to investigate for any difference in behavior of the energy᎐income relationship among various sectors of economic activity. In addition we examine the responses of the levels of these variables to their rates of change Žchange in energy consumption, changes in output and inflation., in order to capture their short-run dynamics of the variables. Testing for the existence of a statistical relationship among the three variables is carried out in three steps. The first step is to verify the order of integration of the variables since the causality tests are valid if the variables have the same order of integration. Standard tests for the presence of a unit root based on the work of: Dickey and Fuller Ž1979, 1981., ŽADF.; Perron Ž1988., Phillips Ž1987. and Phillips and Perron Ž1988., ŽPP.;9 Kwiatkowski et al. Ž1992., ŽKPSS.;10 and Perron Ž1989., ŽPB.11 are

7 For the effects of the Community Support Frameworks on the Greek economy see Lolos et al. Ž1995., Lolos and Theodoulides Ž2001., also Christodoulakis and Kalyvitis Ž1998.. 8 See Masih and Masih Ž1997., Hoa Ž1992. and Dunkerly Ž1982.. 9 This version of the test is an extension of the Dickey᎐Fuller test, which makes a semi-parametric correction for autocorrelation and is more robust in the case of weakly autocorrelated and heteroskedastic regression residuals. According to Choi Ž1992. the Phillips᎐Perron tests ŽPP. extension appear to be more powerful than the ADF tests for aggregate data. For more details see Perron Ž1990.. 10 The KPSS procedure assumes the univariate series can be decomposed into the sum of a deterministic trend, random walk and stationary IŽ0. disturbance and is based on a Lagrange Multiplier score testing principle. This test reverses the null and the alternative hypothesis. A finding favorable to a unit root in this case requires strong evidence against the null hypothesis of stationarity. 11 When there are structural breaks, the previous unit root tests are biased toward the non-rejection of a unit root. Perron Ž1989. developed a formal procedure to test for unit roots in the presence of a structural break at a time period.

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

325

used to investigate the degree of integration of the variables used in the empirical analysis.12 The second step involves testing for cointegration using the Johansen maximum likelihood approach ŽJohansen, 1988; Johansen and Juselius, 1990, 1992.. The Johansen᎐Juselius estimation method is based on the error-correction representation of the VAR model with Gaussian errors.13 Evidence of cointegration rules out the possibility of the estimated relationship being ‘spurious’. So long as the four variables have common trend, causality, in the Granger sense and not in the structural sense, must exist in at least one direction. Although cointegration implies the presence of Granger causality it does not necessarily identify the direction of causality between variables. This temporal Granger causality can be captured through the vector error-correction model derived from the long-run cointegrating vectors ŽGranger, 1986, 1988.. Thus, the third step involves utilization of the vector error-correction modeling and testing for exogeneity of variables. Engle and Granger Ž1987. show that in the presence of cointegration, there always exists a corresponding error-correction representation which implies that changes in the dependent variable are a function of the level of disequilibrium in the cointegrating relationship, captured by the error-correction term ŽECT., as well as changes in other explanatory variables. Thus, through ECT, the VECM modeling establishes an additional way to examine the Granger causality ignored initially from the Granger᎐Sims tests. The Wald test applied to the joint significance of the sum of the lags of each explanatory variable and the t-test of the lagged error-correction term will imply statistically the Granger exogeneity or endogeneity of the dependent variable. The non-significance of ECT is referred as a long-run non-causality which is equivalent to that the variable is weakly exogenous with respect to long-run parameters. The absence of short-run causality ŽGranger causality in the strict sense. is established from the non-significance of the sums of the lags of each explanatory variable. Finally, the non-significance of all the explanatory variables including the ECT term in the VECM indicates the econometric strong exogeneity of the dependent variable, that is the absence of Granger causality. 12

The combined use of the three tests employed to investigate the degree of integration of the series may result in four possible outcomes: Ža. rejection by the ADF and PP statistics and non-rejection by the KPSS test offers strong evidence of stationarity. Žb. Non-rejection by both ADF and PP and rejection by the KPSS is a strong indication of IŽ1.. Žc. Non-rejection by all tests suggests that the data are not sufficiently informative on the long-run characteristics of the series. Žd. Rejection by all tests indicates that the series is neither an IŽ1. nor an IŽ0. process. The gain from testing both stationarity and non-stationarity employing the three tests and comparing their outcomes is that the inference is not uncertain. Specifically, the inference is either stationarity, non-stationarity, or we do not know, which is preferable to being uncertain. 13 This approach has several advantages over the Engle and Granger Ž1987. technique employed in most recent empirical studies: Ža. the Johansen and Juselius method tests for all the number of cointegrating vectors between the variables, all tests being based on the trace statistic test and the maximum eigenvalue test. Žb. It treats all variables as endogenous, thus avoiding an arbitrary choice of dependent variable. Žc. It provides a unified framework for estimating and testing cointegrating relations within the framework of a vector error-correction model.

326

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

The empirical analysis has been carried out using annual data for the period 1960᎐1996 for Greece. The energy consumption variables, namely total energy ŽENT., residential ŽENH. and industry ŽENI. are in Mtoe, real output is the real GDP ŽRGDP. at 1988 market prices, the price variable ŽPRICE. is the Consumer Price Index. All variables are expressed in logarithms ŽLENT, LENH, LENI, LRGDP and LPRICE.. Data for the energy variables are obtained from various issues of the Energy Balances of OECD published by the OECD. The Consumer Price Index is taken from the Bulletin of Conjectural Indicators of the Bank of Greece, while real GDP is obtained from the National Accounts of Greece published by the National Statistical Service of Greece.

4. Empirical results Table 2 presents the results of the PB, PP and KPSS tests for the variables used in the analysis in levels and in first differences Žtotal, residential and industrial energy consumption, real output and prices..14 The ADF statistic suggests that all variables are integrated of order one, IŽ1., whereas for the first differences the results indicate that all variables except price are integrated of order zero, IŽ0.. Therefore, the hypothesis that the time series except price contain an autoregressive unit root is accepted in all cases. Although, employing the Phillips᎐Perron test gives different lag profiles for the various time series and sometimes lowered the level of significance, the main conclusion is qualitatively the same as reported above by the Dickey᎐Fuller tests. In particular, the Phillips᎐Perron test based on the 5 and 1% critical values supports the hypothesis that all series except price contain a unit root. Thus, both tests are in favor of the unit root hypothesis in all time series except for prices. The KPSS statistics test for lag-truncation parameters one and four Žl s 1 and l s 4.15 since it is unknown how many lagged residuals should be used to construct a consistent estimator of the residual variance. The KPSS test rejects the null hypothesis of level and trend stationarity for both lag truncation parameters. The KPSS statistics does not reject the IŽ0. hypothesis for the first differences of the most of the series at different levels of significance. Therefore, the combined results of all tests suggest that all the series except price appear to be IŽ1. processes. Contrary, the ADF, PP and KPSS statistics suggest that the price level variable is integrated of order 2, IŽ2., whereas the first differences Žinflation. is integrated of order 1, IŽ1.. However, since for all the variables under consideration, there is a structural shock in 1973 due to the oil crisis we employed the PB test to perform a

14

Table 2 does not present the results of ADF tests. However, the results are available upon request. The KPSS statistics are known to be sensitive to the choice of truncation parameter one and tend to decline monotonically as one increases. 15

Table 2 Tests of unit roots hypothesis Perron Žwith structural break. Variable

0.30 1.00 0.57 y0.48 y2.03 y6.89a y5.25a y5.76a y7.60a y4.53a

␶␶

y0.04 0.89 y0.59 y0.38 y2.45 y6.74a y5.14a y5.93a y7.77a y4.41a

k

0 0 0 0 0 0 0 0 0 1

␶␮

y2.12 y3.26 y1.46 y5.20 2.49 y3.81b U y3.46b y5.18a y4.29a y1.91

KPSS ␶␣

ls1

y1.58 y1.97 y1.65 y1.78 y3.00 y4.87a y4.41a y5.21a y5.81a y1.74

ls4

␩␮

␩␶

␩␮

␩␶

1.496a 1.696a 1.876a 1.738a 1.677a 1.018a 0.175 1.017a 1.164a 0.964a

0.458a 0.341a 0426a 0.454a 0.467a 0.119 0.072 0317a 0.118 0.095

0.683a 0.760a 0.813a 0.779a 0.754a 0.614a 0.154 0.501a 0.654a 0.597a

0.214b 0.169b 0.198b 0.212b 0.216b 0.148b 0.068 0.186b 0.126 0.109

Notes: The relevant tests are derived from the OLS estimation of the following autoregression for the variable involved: k

⌬ x t s ␦ 0 q ␦1Ž time . t y ␦ 2 x ty1 q

Ý ␾ i ⌬ x ty1 q u t␶␮

Ž1.

is1

327

is the t-statistic for testing the significance of ␦ 2 when a time trend is not included in Eq. Ž1. and ␶␶ is the t-statistic for testing the significance of ␦ 2 when a time trend is included in Eq. Ž1.. The calculated statistics are those reported in Dickey and Fuller Ž1981.. The critical values at 5 and 1% for N s 50 are y2.93 and y3.58 for ␶␮ and y3.5 and y4.15 for ␶␶ , respectively. The lag length structure of ⌽ i of the dependent variable x t is determined using a recursive procedure in the light of a Lagrange multiplier ŽLM. autocorrelation test Žfor orders up to 4. which is asymptotically distributed as ␹ 2 distribution and the value of t-statistic of the coefficient associated with the last lag in the estimated autoregression. The critical values for the Phillips᎐Perron unit root tests are obtained from Dickey and Fuller Ž1981.. ␶␣ denotes the t-statistic for testing the significance of ␦ 2 in Eq. Ž1. with an exogenous break in 1973. The calculated statistics are those reported in Perron Ž1989.. Specifically, the critical values at 5 and 1% for ␭ s 0.4 are y3.72 and y4.34, respectively. ␩␮ and ␩␶ are the KPSS statistics for testing the null hypothesis that the series are IŽ0. when the residuals are computed from a regression equation with only an intercept and intercept and time trend, respectively. The critical values for ␩␮ and ␩␶ at 5% are 0.463 and 0.146, and at 1% are 0.739 and 0.216, respectively ŽKwiatkowski et al., 1992, table 1.. a Indicates significance at the 1% level. b Indicates significance at the 5% level.

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

LENT LENI LENH LRGDP LPRICE ⌬LENT ⌬LENI ⌬LENH ⌬LRGDP ⌬LPRICE

␶␮

Phillips᎐Perron

328

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

unit root test with a trend break in 1973. The empirical results suggest that all the variables, including price, are integrated of order 1 IŽ1.. Table 3 summarizes the results of cointegration analysis between the three categories of energy consumption, real output and prices. To test for cointegration we use the Johansen maximum likelihood approach employing both the maximum eigenvalue and trace statistic. For the determination of the lag length of the VAR three versions of system were initially estimated: a four, a three and a two lag version. Then, an Akaike Information Criterion ŽAIC., a Schwarz Bayesian Criterion ŽSBC. and a likelihood ratio test ŽSims’ test. were used to test that all three specifications are statistically equivalent. All tests reject the null hypothesis that all the specifications are equivalent. In particular, the tests suggest that VAR s 3 should be used in the estimation procedure of cointegration to avoid over-para-

Table 3 Johansen and Juselius cointegration test energy consumption, real output and prices, 1960᎐1996 Null

Alternative

Eigenvalue

Critical values 95%

90%

Total energy consumption: VAR s 3, variables: LENT, LRGDP, LPRICE Maximum eigen¨ alues rs0 rs1 32.68a r -s 1 rs2 6.91 Trace statistic rs0 r)1 44.82a r -s 1 r s) 2 12.15

21.12 14.88

19.02 12.98

31.54 17.86

28.78 15.75

Residential energy consumption: VAR s 3, variables: LENH, LRGDP, LPRICE Maximum eigen¨ alues rs0 rs1 r -s 1 rs2 Trace statistic rs0 r)1 r -s 1 r s) 2

35.74a 9.21

21.12 14.88

19.02 12.98

50.48a 14.74

31.54 17.86

28.78 15.75

Industry energy consumption: VAR s 3, variables: LENI, LRGDP, LPRICE Maximum eigen¨ alues rs0 rs1 r -s 1 rs2 Trace statistic rs0 r)1 r -s 1 r s) 2

28.62a 11.12

21.12 14.88

19.02 12.98

43.12a 14.50

31.54 17.86

28.78 15.75

r: Indicates the number of cointegrating relationships. Maximum eigenvalue and trace test statistics are compared with the critical values from Johansen and Juselius Ž1990.. a Indicates rejection of the null hypothesis at 95% critical value.

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

329

Table 4 Long-run test for the hypothesis that each variable does not enter any cointegrating vector Variables

LR test of restrictions for energy consumption Total

LENT LENH LENI LRGDP LPRICE

Residential

Industrial

a

14.94

13.28a a

13.77 24.23a

7.88a 17.44a 12.30a

a

11.51 11.47a

The reported statistics are distributed as ␹ 2 distribution with degrees of freedom the number of cointegrating vectors. a Indicates rejection of the null hypothesis at 1% level of significance.

meterization of the estimated models. Finally, a log-likelihood ratio test is used to test for the deletion of a dummy variable Žfor the 1973 energy crisis. and the trend from the VAR model. All tests reject the null hypothesis of the deletion of the dummy variable and trend from the VAR model. The estimation procedure assumes unrestricted intercepts and unrestricted trends in the VAR estimation. The two test statistics give similar results, providing evidence to reject the null of zero cointegrating vectors in favor of one cointegrating vector at 5% level. On the basis of the results, the long-run relationship among the variables finds statistical support in Greece over the period under examination. This implies that energy consumption has a long-run equilibrium relationship with income and prices. Such a finding suggests that a long-run neutrality hypothesis is rejected for Greece over the period under consideration. Then, we investigate whether all variables Ženergy consumption, output and prices. enter statistically significant in the cointegrating vectors. Table 4 reports the likelihood ratio tests as described in Johansen Ž1992. and Johansen and Juselius

Table 5 Stationarity hypothesis testing Variables

LR test of restrictions for energy consumption Total

LENT LENH LENI LRGDP LPRICE

Residential

Industrial

a

24.67

24.04a a

24.84 17.23a

a

17.27 13.68a

20.38a 21.57a 22.04a

The reported statistics are distributed as ␹ 2 distribution with degrees of freedom the number of restrictions. a Indicates rejection of the null hypothesis at 1% level of significance.

330

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

Ž1992.. The results suggest that all variables for the three categories of energy consumption enter statistically significantly into the cointegrating vector. Table 5 reports the statistics testing for the stationarity hypothesis of all variables. In particular, we investigate whether all the variables except one, which takes the value one, are equal to zero. The tests which follow a ␹ 2 distribution with 2 degrees of freedom reject the stationarity hypothesis for all variables at 1% level of significance. These findings are in accordance with the rejection of unit root hypothesis, from all the tests, implying that all variables employed in the analysis are not stationary. Having verified that the variables are cointegrated, vector error-correction models can be applied. The lagged residuals from the cointegrating regression with the appropriate number of lags are included in the Granger causality test structure. The lag lengths structure depends on the restricted error-correction models. The restricted error-correction models pass a series of diagnostic tests including serial correlation based on the inspection of the autocorrelation functions of the residuals as well as the reported Lagrange multiplier. Table 6 reports the findings for the endogeneity of the variables, for all three categories of energy consumption, based on error-correction equations. The errorcorrection term measures the proportion by which the long-term imbalance in the dependent variable is corrected in each short-run period. The size and the statistical significance of the error-correction term measures the extent to which each dependent variable has the tendency to return to its long-run equilibrium. Estimates of the parameters show that the error-correction term measuring the long-run disequilibrium is significant only in the equation of total energy consumption, in the income equation Žfor all categories of energy consumption. and in the price level equation Žonly for industrial energy uses.. This implies that total energy consumption, real income and the price level Žfor industrial energy consumption. have the tendency to restore equilibrium and take the burden of the shock to the system. The t-tests for the error-correction terms, at different levels of significance, indicate that residential energy consumption, industrial energy consumption and the price level of total and residential energy consumption variables are weakly exogenous. In the short-run dynamics ŽGranger causality in the strict sense. the Wald tests indicate that there is a relationship in the income equation. In particular, the results suggest that in the short-run changes in income are affected by changes in total energy consumption and inflation. Finally, the significance levels associated with the Wald tests of joint significance of the sum of the lags of the explanatory variable and the error-correction term, provide more information on the causal relationships among the variables, as a further channel of Granger causality is exposed. For all equations, except for the price of total energy consumption and residential energy consumption the results imply the Granger endogeneity of all variables. Finally, the empirical results reject the hypothesis of strong exogeneity of all variables, except the price and the residential energy consumption, supporting the proposition that there is a relationship among energy consumption, income and prices in Greece.

Equations

Test of restrictions Short-run dynamics Non-causality ⌬ LENT

⌬LENT ⌬LRGDP ⌬ LPRICE ⌬ LENH ⌬LRGDP ⌬ LPRICE ⌬LENI ⌬LRGDP ⌬LPRICE

⌬ LENH

⌬ LENI

11.03 a 1.17

⌬ LRGDP

⌬ LPRICE

3.89

14.04 38.34 a

2.22 0.09 2.37 0.26

7.28 1.75 3.57 1.23

1.23

1.84 33.85 a 2.21 0.54

Weak exogeneity, ECT coefficient Zs0

0.61 b 0.67 a 0.02 0.03 y0.02 a 0.005 0.001 0.18 a y0.30 a

Tests for Granger non-causality, Žjoint short-run dynamics and ECT . ⌬ LENT and ECT

⌬ LENH and ECT

⌬ LENI and ECT

Tests for strong exogeneity

⌬ LGDP and ECT 8.24 b

28.34 a 1.72

2.25 2.19 29.90 a 0.58

7.31 1.76 b

9.24 14.58 a

31.46 a

⌬ LPRICE and ECT 20.78 a 71.25 a 3.43 67.96 a 10.51 b 33.67 a

27.53 a 75.49 a 11.0 4.28 74.08 a 7.55 13.91 b 34.74 a 32.23 a

The lagged ECT is derived by normalizing the cointegrating vector on energy consumption. The statistics reported are distributed as chi-square distribution with degrees of freedom the number of restrictions. In the short-run dynamics asterisks indicate rejection of the H o that there is short-run non-causal relationship between the two variables. The coefficients of the lagged ECTs are negative. Asterisks indicate rejection of the null hypothesis that the estimated coefficient is equal to zero Žweak exogeneity.. Finally, in the tests for Granger non-causality and strong exogeneity, asterisks denote rejection of the null hypothesis of Granger non-causality and strong exogeneity respectively. a Indicates significance at the 1% levels. b Indicates significance at the 5% levels.

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

Table 6 Summary of tests for weak and strong exogeneity of variables based on vector error-correction models

331

332

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

From the empirical analysis certain conclusions can be drawn: Total energy consumption in Greece should not be considered exogenous to inflation and to the growth process. The same applies to industrial energy consumption. Also, energy consumption seems to play an important role in determining economic growth. It appears the all forms of energy consumption and prices are responsible for output variations.16 On the contrary, residential energy consumption is an exogenous variable not affected by changes in output and inflation. Thus residential energy consumption should be treated as a necessity with a perfectly inelastic demand curve.17

䢇

䢇

䢇

Our findings suggest that both economic growth and energy saving are an integral part of the same process. Our results for the case of Greece are in line with Hwang and Gum Ž1992. and Masih and Masih Ž1996. who found bi-directional causality between income and energy consumption in Taiwan and Pakistan. The bi-directional relationship suggested by our results seems to be consistent with developments in Greece. The high growth rates in the 1970s, based on the expansion of energy intensive industrial sectors, was accompanied by high energy consumption rates, while the sluggish output developments in the 1980s and early 1990s decelerated energy consumption rates. In the absence of an active energy conservation policy, lower energy consumption in the 1980s is greatly associated with the retardation of industrial activity which counterbalanced the increase of the energy intensive service sector Žresidential and transport..

5. Summary, conclusions and policy implications In this paper, the interrelationship between economic growth, energy consumption and the price level is analyzed. Various categories of energy consumption changes and economic developments, such as income and prices, are explained in a temporal Granger causal framework, employing data for Greece. This is accomplished by examining the dynamic relationship between energy consumption changes and economic conditions in a multivariate cointegrated system. The evidence of cointegration between the three variables suggests they are bound together by common trends or that a long-run equilibrium condition exists. This implies that although the variables may exhibit occasional short-term or transitory deviations from their long-run equilibrium, eventually forces will prevail that will drive them together. In addition, the evidence of cointegration rules out the possibility that the estimated relationships are ‘spurious’, implying that Granger causality must exist 16

Similar results are obtained by Samouilidis and Mitropoulos Ž1984.. Similar result is reached by Zonzilos and Lolos Ž1996. investigating the consumption of energy by households over the period 1970᎐1993. 17

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

333

among these variables either uni-directional or bi-directional. This suggests that although in the short-term some determinants may not be related to energy consumption changes, in the long run these variables are causally interrelated. The estimated cointegrating vector suggests that there is a causal relationship between energy consumption and GNP. This important finding of a long-run relationship is of major importance for policy designers. Furthermore, using vector error-correction model estimation, the direction of Granger causality, within-sample Granger exogeneity or endogeneity for each variable is detected, also distinguishing between short-term and long-term Granger causality. Our results indicate that total energy consumption changes should be considered as endogenous variables to inflation and to economic growth. The same applies to the consumption of energy in industrial uses. However, residential energy consumption is a weakly exogenous variable. In the short-run dynamics ŽGranger causality in the strict sense. the empirical results indicate that there is not a relationship among energy consumption, real output and prices, with the exception of a short-run relationship between total output and total energy consumption. In particular, the results suggest that in the short-run real output is affected by changes in total energy consumption and inflation. Furthermore, the use of strong exogeneity reveals the Granger endogeneity of energy consumption Žwith the exception of residential energy consumption. and economic growth. Our analysis indicates that in the case of Greece, total and industrial energy consumption are closely related to economic developments. On the contrary, residential energy consumption is exogenous to economic developments. In short, the results support the proposition of the endogeneity of energy consumption and economic growth and that these two variables are interrelated in Greece. In the long run, improvements in economic efficiency, stemming from productivity increases via the strengthening of endogenous growth mechanisms and the curtailment of economic imbalances, would promote economic growth, and consequently affect favorably energy consumption. At the same time, improvements in economic efficiency would activate energy conservation mechanisms, which in their turn would affect the growth process positively. As a result, efficiency induced energy saving should not pose impediments to long-run growth, since energy conservation and economic growth are closely interlinked, both being an integral part of the same process. This argument has important policy implications, calling for the adoption of a blend of structural reform and adjustment policies, aiming at increasing economic efficiency. This was the case of Greece over the 1990s. Thus, the implementation of structural policies through the Community Support Framework, together with the adoption of stabilization and liberalization policies Žin line with the Greek Convergence Programme aiming at achieving the Maastricht criteria ., has lead to substantial inflation reductions, has given an accelerating impetus to economic growth and lowered the, albeit high, energy requirements. On the contrary, the adoption of

334

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

stop᎐go policies in the 1980s led to a sluggish development of economic activity, high inflation rates and relatively high energy requirements.18 A similar rationale applies to the industrial sector, which is greatly responsible for the achievement of economic growth due of its strong forward and backward linkages with the other sectors of the economy. In the case of Greece, the successful implementation of structural industrial policies along with macroeconomic restructuring policies in the 1990s, resulted in accelerating economic growth and industrial production rates and to a stabilization of energy consumption by industry. However, the failure of industrial policies in the 1980s and the prevalence of economic instability did not mobilize economic forces adequately. As regards to residential energy consumption, it follows an autonomous path which is independent of economic developments. It is mainly determined by long-run trends in wealth, by international trends and fashions, by consumption tastes and habits, which are not directly associated with fluctuations of economic activity. Finally, the results of our analysis have implications relating to the EU convergence process. The implementation by the Greek government of micro᎐macro structural policies, aiming at achieving convergence towards the other EU member states through the acceleration of GDP growth, gives rise to efficient energy consumption also reducing the high dependence of the economy to energy imports.

References Akarca, A.T., Long, T.V., 1980. On the relationship between energy and GNP: a re-examination. J. Energy Development 5, 326᎐331. Choi, I., 1992. Effects of data aggregation on the power of tests for a unit root. Econ. Lett. 40, 397᎐401. Calogirou, Y.D., Mourelatos, G.A., Thompson, H., 1997. Industrial energy substitution during the 1980s in the Greek economy. Energy Economics 19, 476᎐491. Christodoulakis, N., Kalyvitis, S., 1997. The demand for energy in Greece: assessing the effects of the community support framework 1994᎐99. Energy Economics 19, 393᎐416. Christodoulakis, N., Kalyvitis, S., 1998. A four-sector macroeconomic model for Greece and the evaluation of the community support framework 1994᎐99. Econ. Modelling 15, 520᎐572. Christopoulos, D.K., 2000. The demand for energy in Greek manufacturing. Energy Economics 5, 569᎐586. Dickey, D.A., Fuller, W.A., 1979. Distributions of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 74, 427᎐431. Dickey, D.A., Fuller, W.A., 1981. The likelihood ratio statistics for autoregressive time series with a unit root. Econometrica 49, 1057᎐1072. Donatos, G., Mergos, G., 1989. Energy demand in Greece. Energy Economics 11, 147᎐152. Dunkerly, J., 1982. Estimation of energy demand: the developing countries. Energy J. 3, 79᎐99.

18 The annual growth rate of GDP increased from 1.5% in the 1980s to around 3% in the second half of the 1990s, while inflation dropped from 16% in the 1980s and early 1990s to 3.8% in 1998. Also, the rate of growth of energy consumption decreased from 2.7% in the 1980s to around 1.6% in the 1990s.

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

335

Engle, R.F., Granger, C.W.J., 1987. Cointegration and error-correction: representation, estimation and testing. Econometrica 55, 251᎐276. Erol, U., Yu, E.S.H., 1987. On the causal relationship between energy and income for industrializing countries. J. Energy Dev. 13, 113᎐122. Granger, C.W.J., 1986. Developments in the study of cointegrated economics variables. Oxf. Bull. Economics Statistics 48, 213᎐228. Granger, C.W.J., 1988. Developments in a concept of causality. J. Econometrics 39, 199᎐211. Hwang, D.B.K., Gum, B., 1992. The causal relationship between energy and GNP: the case of Taiwan. J. Energy Development 16, 219᎐226. Hoa, T.V., 1992. A multi-equation model of energy consumption in Thailand. Int. J. Energy Res. 17, 29᎐33. Johansen, S., 1988. Statistical and hypothesis testing of cointegration vectors. J. Econ. Dynamics Control 12, 231᎐254. Johansen, S., 1992. Cointegration in partial systems and the efficiency of single equation analysis. J. Econometrics 52, 389᎐402. Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegrationᎏwith applications to the demand for money. Oxf. Bull. Economics 52, 169᎐210. Johansen, S., Juselius, K., 1992. Testing structural hypotheses in a multivariate cointegration analysis at the purchasing power parity and the uncovered interest parity for the UK. J. Econometrics 53, 211᎐244. Kraft, J., Kraft, A., 1978. On the relationship between energy and GNP. J. Energy Development 3, 401᎐403. Kwiatkowski, D., Phillips, P.C.B., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationarity against the alternative of a unit root. J. Econometrics 54, 159᎐178. Lolos, S. 1998. Success and failure of economic policies: the experience of Greece and Portugal. Comparative Econ. Stud. 40, 1, 72᎐102. Lolos, S., Theodoulides, A., 2001. The role of E.U. structural funds in the development of lagging behind European regions. Economic Financial Modelling 8, 1, 28᎐46. Lolos, S., Suwa, A., Zonzilos, N., Bourguignon, F., 1995. Evaluating the CSF with an extended computable general equilibrium model: the case of Greece Ž1988᎐1995.. J. Policy Modeling 17, 177᎐197. Masih, A.M.M., Masih, R., 1996. Energy consumption, real income and temporal causality: results from a multi-country study based on cointegration and error-correction modelling techniques. Energy Economics 18, 165᎐183. Masih, A.M.M., Masih, R., 1997. On the temporal relationship between energy consumption, real income and prices: some new evidence from Asian-energy dependent NICs based on a multivatiate cointegrationrvector error-correction approach. J. Policy Modeling 19, 417᎐440. Mitropoulos, C., Samouilidis, J., Protonotarios, E., 1982. Forecasting Energy Demand in the Greek Economy, A Kalman Filter Approach, Working Paper. National Technical University of Athens. Žin Greek. Perron, P., 1988. Trends and random walks in macroeconomic time: series further evidence from a new approach. J. Econ. Dynamics Control 12, 297᎐332. Perron, P., 1989. The great crash, the oil price shock and the unit root hypothesis. Econometrica 57, 1361᎐1401. Perron, P., 1990. Tests of joint hypotheses in time series regression with a unit root. In: Rhodes, G.F., Fomby, T.B. ŽEds.., Advances in Econometrics: Co-Integration, Spurious Regression and Unit Roots, 8, pp. 10᎐20. New York. Phillips, P.C.B., 1987. Time series regression with a unit root. Econometrica 55, 277᎐346. Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrica 75, 335᎐346. Samouilidis, J., Mitropoulos, C., 1984. Energy and economic growth in industrializing countries. Energy Economics 6, 191᎐201. Stern, D.I., 1993. Energy and growth in the USA: a multivariate approach. Energy Economics 15, 137᎐150.

336

G. Hondroyiannis et al. r Energy Economics 24 (2002) 319᎐336

Yu, E.S.H., Choi, J.Y., 1985. The causal relationship between energy and GNP: an international comparison. J. Energy Development 10, 249᎐272. Yu, E.S.H., Hwang, B.K., 1984. The relationship between energy and GNP: further results. Energy Economics 6, 168᎐190. Yu, E.S.H., Jin, J.C., 1992. Cointegration tests of energy consumption, income and employment. Resources Energy 14, 259᎐266. Zonzilos, N., Lolos, S., 1996. Household Energy Demand in Greece, Discussion Paper. National Technical University of Athens. Žin Greek..