key: cord-0982001-imuzdvti authors: Lai, T.M.; To, W.M.; Lo, W.C.; Choy, Y.S.; Lam, K.H. title: The causal relationship between electricity consumption and economic growth in a Gaming and Tourism Center: The case of Macao SAR, the People’s Republic of China date: 2010-12-31 journal: Energy (Oxf) DOI: 10.1016/j.energy.2010.11.036 sha: 2221e6593a50a94cc5c2f592b9f970c92d2cc2e5 doc_id: 982001 cord_uid: imuzdvti A number of Asian cities decided to establish gaming and resort facilities in order to capitalize on the growing number of gamblers and their family members in Asia. In doing so, they expect to sustain economic growth but, on the other hand, will consume a considerable amount of energy. Nevertheless, the causal relationship between economic growth and electricity consumption in this type of service-oriented territories has never been investigated. Using the historical data obtained from the Government of Macao SAR, we found that electricity consumption and economic growth in terms of gross domestic product are co-integrated for the period of 1999 Quarter 1–2008 Quarter 4. Moreover, vector error correction (VEC) models indicated a lack of short-run relationships but showed that there was a long-run equilibrium relationship between electricity consumption and gross domestic product. The accuracy of VEC models was assessed by using the mean squared error and the mean absolute error. The error analysis shows that VEC models reproduced time series of gross domestic product and electricity consumption in difference form accurately. Macao, a special administrative region in the People's Republic of China, has experienced a phenomenal economic growth since the liberalization of the gaming industry in 2002. According to the statistics provided by the Gaming Inspection and Coordination Bureau of Macao, gross revenue from Macao's casinos amounted to US$13.6 billion in 2008, representing an increase of 31% on year-toyear basis [1] . This figure exceeded the gross income of US$6.12 billion from Las Vegas Strip's casinos as reported by the Nevada Gaming Commission by a very wide margin. The direct taxes from gaming amounted to US$4.9 billion in 2008, representing 77% of the total income of the Government of Macao SAR. Besides, the annualized gross domestic product (GDP) per capita in Macao was estimated as US$39,036 at the end of 2008, higher than that of Hong Kong and Singapore. As the majority of Macao GDP (62.25%) depends on the gaming and the associated hospitality services provided to visitors, the change in GDP is a good proxy for economic growth in Macao e the world gaming center. In fact, many Asian cities including Singapore and Penghu of Taiwan have decided or proposed to jump on the bandwagon. However, this type of service industry, like the manufacturing sector, consumes a substantial amount of electricity to operate and brighten its infrastructure such as casinos, hotels, resorts, and meeting, exhibition and convention venues day and night. To cope with the increasing demand of electricity consumption, public policy makers in those cities need to have a better understanding on the relationships between economic activities, electricity consumption, and economic growth. The purposes of this paper are, therefore, (i) to describe the relationship between economic activities and electricity consumption, and (ii) to investigate the long-run relationship and short-run causality relationship between electricity consumption and economic growth in terms of GDP in Macao. The paper is organized as follows. In the next section, we review previous studies on modeling of electricity consumption and literature on causality studies of energy consumption and economic growth. In Section 3, the methodology adopted in the study is presented. Section 4 describes the data employed and reports the findings. Concluding remarks and policy implications are given in Section 5. Numerous studies have been conducted to examine the relationship between electricity consumption and various economic indicators in the past fifty years. In the 1950s, Houthakker [2] analyzed electricity demand on domestic two-part tariffs for 42 provincial towns in the U.K. He found that average annual electricity consumption per consumer was a function of the average income per household, marginal price of elasticity and marginal price of competing forms of energy, such as gas. Houthakker also reported that the monthly electricity consumption of families had a strong seasonal variation, depending on average temperature and average hours of daylight per day for each month. Foss [3] studied the utilization of capital equipment and suggested that electricity consumption was an indicator of capital usage, especially in an industrial country such as the USA. Foss's idea was adopted by Jorgensen and Griliches [4] and Heathfield [5] to measure capital usage using electricity consumption data. Mount et al. [6] analyzed both the short-run and long-run demand for electricity for three classes of consumers, namely residential, commercial and industrial. They demonstrated that long-run electricity demand was generally price elastic and became increasingly elastic as prices rose. In contrast, demand was general inelastic with respect to income, especially for residential and industrial classes that approached zero as income increased. Population exhibited approximately unit elasticity for all classes implying the common practice of estimating demand models on a per capita basis to be reasonable. In the past decade, researchers [7e12] studied the modeling of electricity consumption in different countries or cities. When annual data were used, researchers [8, 10] found that gross national product (GNP) or its derivative such as the number of tourists in a tourist center was the major determinant of electricity consumption. When monthly data were used, researchers found that population (POP), temperature (TEMP) and other economic or industrial factors are the major determinants of electricity consumption [7, 9, 11, 12] . Moreover, Pao [11] reported that economic indicators such as gross domestic product (GDP) and consumer price index (CPI) have a very weak instantaneous effect on Taiwan's electricity consumption. Lai et al. [12] used multiple regression, artificial neural work (ANN), and wavelet-ANN to model electricity consumption in Macao for the period of January 2000eDecember 2006. The results show that the total monthly electricity consumption depends on economic factors such as the number of visitors and their hotel-room occupancy rate, demographic and climatic conditions in Macao. Most of the more recent studies show electricity consumption to be strongly associated with the business activities of a country or city. In pioneering the study of causal relationship between energy consumption and economic growth, Kraft and Kraft [13] used annual data on gross energy inputs and GNP between 1947 and 1974 in the USA and utilized the test for unidirectional causality as outlined by Sims [14] . They found the evidence of a unidirectional causality running from GNP to energy consumption. Yu and Hwang [15] reexamined the causality between energy consumption and GNP over the time period of 1947e1979 using annual data. However, they did not find any evidence to support any causality between energy consumption and GNP over the entire sample period. When the time period was changed to 1973e1981 and quarterly data were used, Yu and Hwang [15] reported that unidirectional causality ran from GNP to energy consumption. A year later, Yu and Choi [16] studied the causal relationship between energy consumption and GNP using Sims and Granger causality tests for five countries. They reported that there was (i) no causal relationship between energy consumption and GNP in the USA, UK and Poland; (ii) unidirectional causality from GNP to energy consumption in South Korea; and (iii) unidirectional causality from energy consumption to GNP in the Philippines. Nachane et al. [17] used EngleeGranger co-integration approach [18] to test the energyeGDP relationship over the time period 1950e51 to 1984e85 for 16 countries. They reported that there was long-run relationship between energy consumption and GDP for 11 developing countries and 5 developed countries. Since then, many researchers [19e34] adopted EngleeGranger co-integration approach or its modified version to study the causal relationship between energy consumption or electricity consumption and economic growth such as GDP or GNP. Table 1 summarizes the empirical findings of the causality tests between energy consumption and economic growth over the past three decades. The results from more recent studies [23e34] showed that there was in general a relationship between energy consumption and economic growth. More specifically, Chontanawat et al. [32] suggested that causality from energy to GDP was found to be more prevalent in the Organization for Economic Cooperation and Development (OECD) countries compared to the developing non-OECD countries. However, when it comes to whether the increase in energy consumption is a result of or a cause of economic growth, there is no conclusive agreement on this issue, possibly due to other exogenous factors that may affect energy consumption and GDP simultaneously or sequentially (but have been ignored in most prior studies). Therefore, our study was aimed at studying the relationship between electricity consumption and GDP with the number of tourists and population as important exogenous factors in the world's gaming center e Macao SAR using Johansen's methodology [35] . In his 1969 classic paper on causality, Granger [36] stated that a causal model with two variables is expressed as: where X t and Y t are stationary time series. He mentioned that when b 0 ¼ c 0 ¼ 0, Eqs. (1) and (2) are the expression for a simple causal model. Granger further noticed that "Whether or not a model involving some group of economic variables can be a simple causal model depends on what one considers to be the speed with which information flows through the economy and also on the sampling period of the data used. It might be true that when quarterly data are used (let alone the annual data), for example, a simple causal model is not sufficient to explain the relationships between the variables, while for monthly data a simple causal model would be all that is required. Thus, some nonsimple causal models may be constructed not because of the basic properties of the economy being studied but because of the data being used. It has been shown elsewhere that a simple causal mechanism can appear to be a feedback mechanism if the sampling period for the data is so long that details of causality cannot be picked out." (Granger, 1969 , p. 427) Table 1 shows that there were relative few studies on the causal relationship between energy consumption and economic growth employed quarterly data and most of previous studies were all based on annual data. As the increase in GDP and electricity consumption is more pronounced only after Macao was returned to mainland China in 1999, our study decided to use quarterly data because of the number of data points available and its better ability to capture short-term changes. United States 1946e2000 annual data Co-integration and the Granger causality tests Exergy / GDP a Note: ), /, )/ and [] mean "forward causality", "backward causality", "bidirectional causality" and "no relationship", respectively. a Exergy (the amount of energy available for useful work) was used in Warr and Ayres's study [34] . Standard Granger causality tests have to be conducted on stationary time series. Engle and Granger [18] stated that economic variables themselves may not be stationary and can have infinite variance, however, they can achieve temporarily stationary after differencing d times (where d is 1, 2, . and so on) and the series can be designated as I(d). When two economic variables X t and Y t are both I(d), it is generally true that the linear combination Z t ¼ X t À aY t will also be I(d) [or I(d À b), b > 0. When the latter case occurs, a very special constraint operates on the long-run components of the series. And X t and Y t are called co-integrated. Following this line, we first test the unit roots of X t and Y t to confirm the stationary properties of each variable. This is achieved by using the Augmented DickeyeFuller [37] test and the PhillipsePerron [38] test. For the time series X t , the Augmented DickeyeFuller (ADF) relationship is expressed as: where D is the difference operator, p is the auto-regressive lag length that has to be large enough to eliminate possible serial correlation in b i and a and r are the coefficients of interest. For time series with strong seasonality, such as quarterly GDP, electricity consumption and the number of tourists, we apply seasonal adjustment before taking the logarithmic transformation as suggested by Pao [33] . When these variables are found to be non-stationary, we take the first-difference and then apply the ADF and PhillipsePerron tests again on the differenced data and so on. To test for co-integration, we employ the Johansen's vector auto-regressive methodology [35] . Intuitively, the Johansen's VAR procedure is a multivariate version of the univariate DickeyeFuller test. Considering a structural form VAR of order p, we have where k is the number of lags, Z t ,Z tÀ1 ,.,Z tÀk are (2  1) vectors containing the logarithmic of seasonally adjusted GDPeELEC pairs, terms, 3 t is an (2  1) vector of disturbance terms. We can rewrite this structural VAR in error form (Verbeek [39] , p. 326) as: .k À 1. The P matrix represents the adjustment to disequilibrium following an exogenous disturbance. If P has a reduced rank r < k where r and k denote the rank of P and the number of variables constituting the long-run relationship respectively, then there exist two k  r matrices a and b, each with rank r, such that P ¼ ab 0 and b 0 y t is stationary. r is called the co-integration rank and each column of b is a co-integrating vector. Johansen's VAR method estimates the P matrix from an unrestricted VAR and tests whether one can reject the restrictions implied by the reduced rank of P. There are two common approaches to test the reduced rank of P, namely the maximum eigenvalue test and the trace test. For the maximum eigenvalue test, the likelihood ratio is determined from: where T is the maximum time in the time series t. The null hypothesis of r co-integrating vectors is tested against the alternative hypothesis of r þ 1 or more co-integrating vectors. Since the co-integration tests are sensitive to the choice of lag length, we use the Schwartz Information Criteria to determine the optimal lag lengths. On the other hands, the likelihood ratio statistic for the trace test is: where b l i are the estimated 2 À r smallest values and r ¼ 0 and 1. The null hypothesis is that there are at most r co-integrating vectors. The alternative is that there are r or more co-integrating vectors. Again, we use the Schwartz Information Criteria to determine the optimal lag lengths. If two variables are co-integrated, there is causality between these two variables in at least one direction [40] . If co-integration does not exist between variables, standard VAR in difference form is applied [36] . If co-integration exists between variables, Engle and Granger [18] propose that the VEC model can be used to test Granger causality for at least one direction (also see Oxley and Greasley [41] ). The general VECM form is presented as follows. where X t and Y t represent the seasonally adjusted GDP and ELEC in logarithmic form, (DX t , DY t ) are the differences in variables, and b i & g j and b i & c j capture the short-term relationships respectively. The m t , v t are the serially uncorrelated error terms. The ECT tÀ1 is derived from the long-run co-integration relationship. The optimum lag lengths p, q, r and s are determined based on Schwartz Information Criteria. This specification allows us to test for both short-run and longrun causality. For example, electricity consumption does not Granger cause the growth in GDP in the short-run if and only if all the coefficients g j are equal to zero in Eq. (8) . On the other hand, the growth in GDP does not Granger cause electricity consumption if and only if all the coefficients b i are equal to zero in Eq. (9) . The presence of long-run causality can be established by examining the significance of the coefficient of the error correction term, ECT tÀ1 in Eqs. (8) and (9) using a t-test. Finally, we conduct a joint test of ECT tÀ1 and the respective interactive terms in Eqs. (8) and (9) using joint F-statistics. However, if the data are I(1) but not co-integrated, VAR can be applied to the data generated by taking the first-order difference of the variables as shown below [36, 41] : 3.5. Accuracy of VEC models VEC models do not only describe the causal relationships between endogenous variables, they also present the vector error correction form of dependent variables, i.e. gross domestic product and electricity consumption when the variables are co-integrated in the study, as a function of their past values and exogenous variables. The VEC equations can be used to regenerate time series of dependent variables. As suggested by Lai et al. [12] , we adopted the mean squared error (MSE) and the mean absolute error (MAE) to evaluate the accuracy of VEC models. The formulation of MSE and MAE is given as follows: where e is the error defined as the difference between the actual value and the predicted value using the VEC models. In this study, we focused on determining the causal relationship between gross domestic product (GDP) and electricity consumption (ELEC), and determined to what extent the number of tourists (TOUR) and population (POP) affect the relationship in the world gaming center. As GDP data was only available on a quarterly basis, we obtained the values of ELEC and TOUR based on quarterly data while POP was taken as the number of residents at the end of each quarter. The values of GDP, ELEC, TOUR and POP for the period of 1999 Quarter 1e2008 Quarter 4 were collected from the Principle Statistical Indicators of Macao published by Macao's Statistics and Census Service. ELEC was expressed in terms of kilowatt-hour (kWh) while economic output was characterized by GDP in 2002 prices, using GDP deflators. These variables were transformed to logarithmic values, namely LGDP, LELEC, LTOUR and LPOP, and logarithmic seasonally adjusted values, namely LGDPSA, LELECSA and LTOURSA, before tests were performed as suggested by Pao [33] for quarterly data. in Asia. Since the aim of the study was to explore the causal relationship between electricity consumption and economic growth, LGDPSA and LELECSA were set as endogenous variables and LTOURSA and LPOP as exogenous variables in causal analysis. As data magnitude might have an effect on the analysis of causal relationship between variables, time series data were processed by logarithmic transformation and tests were repeated using different combinations of endogenous and exogenous variables, namely LGDPSA, LELECSA, LTOURSA and LPOP. The computer software employed in the study was Eviews 6.0. Table 2 presents the results of the augmented DickeyeFuller (ADF) and PhilipsePerron (PP) unit-root tests for the two endogenous and two exogenous variables used in the analysis. In each test, we included a constant in the auto-regressive model. According to the ADF results shown in Table 2 , the null hypotheses could not be rejected even at the 5 percent level for LGDPSA, LELECSA, LTOURSA, and LPOP in levels but the null hypotheses were rejected at the 1 or 5 percent level for LGDPSA, LELECSA, LPOP and LTOURSA in first and second differences. Therefore, the ADF statistics show that all investigated variables including LGDPSA, LELECSA, LPOP and LTOURSA were integrated of order one, I(1), which are normal for typical economic parameters. Similar conclusions were obtained from the PhillipsePerron (PP) unit-root tests. The null hypotheses were not rejected at the 5 percent level for LGDPSA, LELECSA, LPOP and LTOURSA in levels but were rejected at 1 or 5 percent level for LGDPSA, LELECSA, LPOP and LTOURSA in first and second differences, which indicated an integration order of I(1). The result of this PP unit-root test for all parameters (i.e. I(1)) are the same to that of ADF unit-root test (i.e. I(1)). As all variables LGDPSA, LELECSA, LPOP and LTOURSA were nonstationary in levels based on ADF and PP unit-root tests, co-integration test was used to check whether a linear combination of two or more non-stationary series is stationary. The co-integration analysis is typically applied to verify if there exists a long-run relationship between the variables. The results for the Johansen maximum likelihood tests on the unrestricted models are shown in Table 3 . Table 3 Table 3 , H 0 : r ¼ 0, r 1 and 2 were accepted at the 5 percent critical level) but LELECSA & LGDPSA and LELECSA & LGDPSA & LTOURSA had one or more than one co-integration equations (in Table 3 , H 0 : r ¼ 0 or r ¼ 0 & r 1 were rejected at the 5 percent critical level). The optimum lag lengths for different combinations of endogenous and exogenous variables, as shown in Table 3 , were determined using the minimum Schwartz Information Criterion (SIC) through unconstrained VAR estimation. After having verified the co-integration properties of the variables, vector error correction models were applied to the firstdifference variables to determine the long-run and short-run relationships between variables respectively [43e45]. Based on the previous results of Johansen co-integration tests, two endogenous and exogenous combinations (i.e. LELECSA & LGDPSA and LELECSA & LGDPSA & LTOURSA) were selected for vector error correction (VEC) modeling. However, the combinations with the endogenous variable LPOP did not show co-integration and the major purpose of this analysis is to determine the causality between LELECSA and LGDPSA, the analyses for the combinations with LPOP would not further be conducted. The appropriate lag numbers of different endogenous and exogenous combinations were determined using the minimum Schwartz Information Criterion (SIC). Since the appropriate lag number of these VEC models was 1, we tried different combinations of lag numbers from 1 to 3 in VEC models for comparison. Three major criteria were adopted to compare various VEC models. They were the goodness of fit, stability and causality. First, the goodness of fit of a VEC model describes how well the regenerated data fit a set of observations as it measures the variability between the actual data and the estimated data using VEC equations. The adjusted coefficient of determination (adjusted R 2 ) was used to indicate the goodness of fit of the model investigated. The higher adjusted R 2 , the better goodness of fit. Moreover, the stability of VEC models is a statistical property of continuance to estimate the stable distributions provided by a linear combination of variables in models. For example, structural changes and/or discontinues in the data typically may cause unstable in estimated models. In VEC models, we applied unit-root tests for determining model stability. Furthermore, t-tests and Wald tests were applied to test whether the long-run relationships and short-run causalities exist in VEC models. In Wald tests, chi-square c 2 statistics and their probabilities were obtained to determine the short-run causalities in VEC models, respectively. When the probability of chi-square statistics was below 5 percent, the null hypothesis was rejected based on the 5 percent significance level, implying a VEC model having a short-run causal relationship between dependent and independent variables. Tables 4 and 5 Notes: For ADF tests, the optimal lag length was determined using Schwartz Information Criteria. The critical values of the ADF statistic at the 0.01, and 0.05 levels are À4.334, and À3.603 respectively (see MacKinnon [42] ). * and ** denote the rejection of the null hypothesis at the 1 percent and 5 percent significance value, respectively and number in parentheses are the corresponding p-values. a D and D 2 denote first-difference and second difference of variables. LGDPSA / LELECSA in the long-run). Moreover, the joint F test for the sum of lagged terms of each explanatory variable (i.e. DLGDPSA and DLELECSA) and ECT tÀ1 term in DLGDPSA and DLELECSA equations evaluated whether short-run adjustment to re-establish long-run equilibrium, given a shock to the system. The results in Table 6 show that the joint significance of DLGDPSA and ECT tÀ1 term in DLELECSA equation existed but that of DLELECSA and ECT tÀ1 term in DLGDPSA equation did not exist. This implies that electricity consumption and gross domestic product interact in the short-term to restore long-term equilibrium after a change in gross domestic product. The situation is mainly caused by change in the number of tourists (and their spending behaviors) affecting electricity consumption and gross domestic product at the same time. And after a longer period, the responses in electricity consumption and gross domestic product become stable and restore the equilibrium between electricity consumption and gross domestic product. Overall, the DLGDPSA & DLELECSA VEC models indicated a longrun causality and joint short/long-run causality from gross domestic product to electricity consumption (DLGDPSA / DLE-LECSA in the long-run), but no short-run causality existed. They represent that short-run and long-run relationships are different. The DLGDPSA equation indicated that there was no short-run, longrun and joint short/long-run causality. The DLELECSA equation, however, indicated that there was a long-run (and joint short/longrun) causality but no short-run causality. These results indicated that although no short-run causality from gross domestic product to electricity existed, short-run interaction can lead to a long-run causality. Moreover, the long-run causality from gross domestic product to electricity existed significantly. The VEC model shown in Table 7 The results of VEC shown in Tables 4e6 matched with the results of co-integration tests shown in Table 3 . Based on these analyses, we conclude that there was a unidirectional long-run relationship from gross domestic product to electricity consumption. Eqs. (14) and (15) were used to regenerate the values of DLGDPSA and DLELECSA for the period of 1999 Quarter 3e2008 Quarter 4. The regenerated data were then compared to the actual data and the MSE and MAE were determined using Eq. (12) and (13). The values of MSE and MAE were 0.0038 and 0.0433 for DLGDPSA and 0.0008 and 0.0224 for DLELECSA respectively, as shown in Table 7 . One-sample t-tests show that the values of squared error and the values of absolute error of DLDGPSA and DLELECSA were not significantly different from zero at the 95 percent level of confidence. The results illustrate that VEC models reproduced time series of DLGDPSA and DLELECSA accurately. Co-integration analyses show that there was a co-integrated relationship between seasonally adjusted electricity consumption and economic growth (LELECSA and LGDPSA) in Macao over the period of 1999 Quarter 1e2008 Quarter 4. Vector error correction models indicated that there was a long-run unidirectional causal relationship from gross domestic product to electricity consumption. Moreover, given a change in gross domestic product, gross domestic product and electricity consumption interacted in shortterm and then re-established the long-run equilibrium. Co-integration test showed that seasonally adjusted gross domestic product, electricity and tourists' amount in natural log scale (LGDPSA, LELECSA and LTOURSA) were co-integrated. Although in previous section, we adopted the VEC model with DLGDPSA and DLELECSA only, change in the number of tourists (DTOURSA) will have a peculiar effect on DLGDPSA than on DLELECSA. It can be understood because the gaming and tourism center in Asia Macao SAR depends heavily on the tourism industry (but in fact more on the high rollers among those tourists). The number of tourists (more specifically, the number of high rollers) affects significantly and positively on gross domestic product and in turn affects electricity consumption. A detailed observation of the DLELECSA VEC equation reveals that DLELECSA depends heavily and significantly on the lag 1 of DLGDPSA. This implies that in a service-oriented territory, infrastructures such as casinos, hotels, resorts, convention centers, food-serving outlets, etc. have to be built (leading to a sudden increase in GDPSA) and operated to attract tourists, consequently leading to a positive change in the city's or national electricity consumption. The results provide useful information to policy makers in other Asian cities or countries. First, a city can enjoy the economic growth brought by the development of the gaming and tourism industry but policy makers have to find ways to produce and utilize their electricity more environmentally friendly and efficiently. It is because the depletion of energy resources and the emission of carbon dioxide e a major greenhouse gas are recognized as international problems. Nevertheless, it is possible that operators of entertainment venues can adopt a wide range of energy saving practices, such as adopting variable speed drives for fans, pumps and lifts, introducing heat exchangers and heat pumps in airconditioning and refrigerating systems, and adopting energy-efficient and yet cost effective and environmental-friendly commercial applicants [46] , etc. Second, whether the gaming industry will bring negative impacts on social and cultural systems is debatable [47] . Macao's economic development in recent years is eyeopening. However, the number of young problem gamblers has increased over time. Gaming statistical information of Macao. Macao Gaming Inspection and Coordination Bureau Some calculations on electricity consumption in Great Britain The utilization of capital equipment: postwar compared with prewar The explanation of productivity change The measurement of capital usage using electricity consumption data for the UK Electricity demand in the US: an econometric analysis Modeling of electrical energy consumption in Delhi Economic variables and electricity consumption in North Cyprus Neural networks in forecasting electrical energy consumption: univariate and multivariate approaches Forecasting total and industrial sector electricity demand based on genetic algorithm approach: Turkey case study Comparing linear and nonlinear forecasts for Taiwan's electricity consumption Modeling of electricity consumption in the Asian gaming and tourism center e Macao SAR, People's Republic of China On the relationship between energy and GNP The relationship between energy and GNP: further results The causal relationship between energy and GNP: an international comparison Co-integration and causality testing of the energyeGDP relationship: a cross-country study Co-integration and error correction: representation, estimation, and testing An investigation of co-integration and causality between energy consumption and economic growth Energy consumption, real income and temporal causality: results from a multi-country study based on cointegration and error-correction modeling techniques An investigation of co-integration and causality between energy consumption and economic activity in Taiwan Cointegration, error correction and the relationship between GDP and energy: the case of South Korea and Singapore A note on the causal relationship between energy and GDP in Taiwan Energy consumption and economic growth assessing the evidence from Greece Electricity consumption and economic growth in India Electricity consumption and economic growth in China Energy consumption and economic growth in Korea: testing the causality relation Electricity consumption and economic growth: a time series experience for 17 African countries The causal relationship between electricity consumption and economic growth in the ASEAN countries Electricity consumption and economic growth in China: cointegration and co-feature analysis Energy consumption and GDP revisited: a panel analysis of developed and developing countries Does energy consumption cause economic growth?: Evidence from a systematic study of over 100 countries Forecast of electricity consumption and economic growth in Taiwan by state space modeling Evidence of causality between the quantity and quality of energy consumption and economic growth Statistical analysis of cointegration vectors Investigating causal relations by econometric models and crossspectral methods Distribution of the estimators for autoregressive time series with a unit root Testing for a unit root in time series regression A guide to modern econometrics Some recent developments in a concept of causality Vector autoregression, cointegration and causality: testing for causes of the British industrial revolution Numerical distribution functions for unit root and cointegration tests Statistical inference in vector autoregressions with possibly integrated process Monte Carlo evidence on cointegration and causation Critical values for an F-test for cointegration in the multivariate model Characterization of commercial clothes dryers based on energy-efficiency analysis Changes in residents' gambling attitudes and perceived impacts at the fifth anniversary of Macao's gaming deregulation The authors gratefully acknowledge support of this work by the Macao Polytechnic Institute through Grant P014/DECP/2009. We also thank anonymous reviewers for many valuable comments that improve the paper greatly.