Will initiatives to promote hydroelectricity consumption be effective? Evidence from univariate and panel LM unit root tests with structural breaks

Transcription

1 DEPARTMENT OF ECONOMICS ISSN DISCUSSION PAPER 47/13 Will initiatives to promote hydroelectricity consumption be effective? Evidence from univariate and panel LM unit root tests with structural breaks Hooi Hooi Lean and Russell Smyth Abstract This paper examines whether initiatives to promote hydroelectricity consumption are likely to be effective by applying univariate and panel Lagrange Multiplier (LM) unit root tests to hydroelectricity consumption in 55 countries over the period 1965 to We find that for the panel, as well as about four-fifths of individual countries, that hydroelectricity consumption is stationary. This result implies that shocks to hydroelectricity consumption in most countries will only result in temporary deviations from the long-run growth path. An important consequence of this finding is that initiatives designed to have permanent positive effects on hydroelectricity consumption, such as large-scale dam construction, are unlikely to be effective in increasing the share of hydroelectricity, relative to consumption of fossil fuels Hooi Hooi Lean and Russell Smyth All rights reserved. No part of this paper may be reproduced in any form, or stored in a retrieval system, without the prior written permission of the author. 1

2 Introduction Hydroelectricity is a form of hydropower used to produce electricity via the kinetic energy of moving water, which is converted to electrical energy by a water turbine driving a generator (Bakis, 2007). Hydropower has been described as the most technically mature and reliable source of renewable energy (Jia et al., 2012, p. 1357). It is the most important renewable energy source for the generation of electricity worldwide, providing 97 per cent of all electricity generated from renewable sources (Bakis, 2007). In 2008, hydropower generation was responsible for approximately 20 per cent of the world s electricity, and 40 per cent of the electricity in developing countries, which was second only to fossil fuels (Jia et al., 2012). In that same year, there were 16 countries that relied on hydropower for more than 90 per cent of their energy supply; 49 countries that relied on hydropower for more than 50 per cent of their energy supply and 57 countries that relied on hydropower for more than 40 per cent of their energy supply (Jia et al., 2012). The world s demand for electricity is expected to increase. The United States Energy Information Administration forecasts that world electricity consumption will almost double between 2007 and 2035, while the International Energy Agency predicts that by 2040 world energy consumption will be 30 per cent higher than in 2010 (Vandel, 2012). Fossil fuels are the major cause of climate change and global warming (Stern 2006). Given that burning fossil fuels is the major way of generating electricity, this raises serious concerns about how the projected growth in world electricity consumption will effect the environment under steady state assumptions. It is in this context that it is often argued that hydropower represents a clean alternative source of electricity to burning fossil fuels and that increasing the proportion of electricity generated from hydropower would be beneficial for the environment (see eg. Bakis, 2007; Jia et al., 2012; Vandal, 2012). There is plenty of scope to further develop hydropower use. As a 2

3 proportion of potential availability, current electricity generated from hydropower is 11 per cent in Africa, 25 per cent in Asia, 45 per cent in Oceania, 71 per cent in Europe, 65 per cent in North America and 40 per cent in South America (Jia et al., 2012). Perhaps the main benefit of increasing hydropower is that compared to burning coal, hydroelectricity produces virtually no air pollution (Bakis, 2007; Chamberland & Levesque, 1996; Childress, 2009; Rosa, 2001). Among the various sources of energy, hydropower has the highest energy payback and lowest greenhouse gas emission (Jia et al., 2012). The purpose of this study is to examine the integration properties of hydroelectricity consumption for 55 countries/subregions using univariate and panel unit root tests, with and without structural breaks. 1 There has been a surge in studies that examine the unit root properties of energy consumption, to the point that testing for a unit root in energy consumption has been described as a new branch of research in energy economics (Narayan et al., 2010, p. 1953). However, most studies have focused on aggregate energy consumption or consumption of specific fossil fuels and have used data from the United States. This study responds to calls for more studies that examine the unit root properties of disaggregated renewable energy consumption and consider countries other than the United States (Smyth, 2013). The use of a relatively large number of countries has the advantage that it is possible to use both a panel and allow for structural breaks, which has been recommended as providing the most reliable evidence of the order of integration of energy variables (Smyth, 2013). The issue of whether hydroelectricity consumption contains a unit root speaks directly to the efficacy of policies, such as construction of large-scale water storage infrastructure, promotion of small hydropower plants and investment in new hydro technologies such as hydrokinetics, designed to increase the proportion of electricity generated from hydropower. 1 The sample consists of 50 countries plus five residual series/sub-regions corresponding to five of the six regions considered (Other South and Central America, Other Europe and Eurasia, Other Middle East, Other Africa and Other Asia-Pacific). For simplicity, hereafter, we refer to countries. 3

4 Large-scale dams, in particular, often represent the single largest structure in terms of basic infrastructure in all countries and become major landmarks in the countries in which they are located (Bakis, 2007; Jia et al., 2012). Hence, each of these policies is designed to bring about permanent changes in the electricity mix and are, by their vary nature, long-term because they typically entail making large-scale investments. If hydroelectricity consumption contains a unit root, a shock to the long run growth path of hydroelectricity consumption will have permanent effects. This implies that policies involving large-scale investment, the purpose of which is to engineer positive shocks to hydroelectricity consumption, will result in continuing annual shocks and, as such, will be effective. On the other hand, if hydroelectricity consumption is stationary, following a policy-induced shock, hydroelectricity consumption will return to its long run growth path and policies designed to engineer permanent changes to consumption will be ineffective. Existing Literature Beginning with Narayan and Smyth (2007), a large literature has evolved which tests for a unit root in energy consumption. This literature has recently been comprehensively reviewed in Smyth (2013). A subset of this literature has examined the unit root properties of renewable energy or other alternative energy sources (Apergis & Tsoumas, 2011; Barros et al., 2012, 2013a, 2013b; Lean & Smyth, 2013a, 2013b). There are two features of this literature worth noting. The first is that with the exception of Lean and Smyth (2013a), who examine the unit root properties of aggregate renewable energy generation across 115 countries, each of these studies is focused on the United States. The second is that with the exception of Barros et al., (2013b), these studies do not examine the unit root properties of hydroelectricity consumption, despite its importance as a renewable energy source and the need to get a handle on the likely efficacy of policies to increase the proportion of hydropower in the electricity mix. Barros et al. (2013b) found that hydropower exhibits 4

5 nonstationarity, but mean reverting behaviour, in the United States. We address a gap in the literature by not only considering the unit root properties of hydroelectricity consumption in the United States, but other countries, and regions, as well. Methodology We apply the Lagrange Multiplier (LM) family of univariate and panel unit root tests with up to two structural breaks. The LM univariate unit root tests consist of the Schmidt and Phillips (1992) LM unit root test with no structural breaks and the Lee and Strazicich (2003) LM unit root test with one and two structural breaks. The LM unit root test with one and two breaks developed by Lee and Strazicich (2003) represent a methodological improvement over ADFtype endogenous one and two break unit root tests proposed by Zivot and Andrews (1992) and Lumsdaine and Papell (1997), which have the limitation that the critical values are derived while assuming no break under the null hypothesis. Nunes et al. (1997) showed that this assumption leads to size distortions in the presence of a unit root with structural breaks. As a result, when utilizing ADF-type endogenous break unit root tests, one might conclude that a time series is trend stationary, when in fact it is nonstationary with break(s), meaning that spurious rejections might occur. In contrast to the ADF-type endogenous break tests, the LM unit root test is unaffected by structural breaks under the null hypothesis (Lee & Strazicich, 2001). The panel LM unit root test with up to two structural breaks was proposed by Im et al. (2005). The panel LM unit root test proposed by Im et al. (2005) has the major advantage that, consistent with the univariate LM unit root tests, it allows for structural breaks under both the null and alternative hypotheses. Given that the univariate and panel LM unit root tests with up to two breaks are well known, we do not reproduce the methodology here, but refer the reader to the original articles or other recent applications of the LM unit root tests to energy consumption (see eg. Aslan & Kum, 2011; Aslan, 2011; Apergis & Payne, 2010; Apergis et al., 2010; Lean & Smyth, 2013a; Narayan et al., 2008, 2010). 5

6 In the univariate LM unit root case, we begin with the Schmidt and Phillips (1992) test with no breaks, which provides a benchmark for later results. As first noted by Perron (1989), a problem with the no-break test is that failure to incorporate a statistically significant structural break will reduce the power to reject the null hypothesis of a unit root. Hence, it is important to accommodate structural break(s). For the univariate unit root test with structural break(s), we employ both the LM unit root test with one and two breaks in the intercept (Model A, Model AA) (the so-called crash model ) and the LM unit root test with one and two breaks in the intercept and slope (Model C, Model CC) (the so-called crash-cum-growth model ). However, just as failure to accommodate structural breaks might result in loss of power to reject the unit root null, as Ben-David et al. (2003) note, allowing for a break does not necessarily produce more rejections of the unit root null hypothesis because the critical value increases in absolute value. Thus, in the analysis below, we employ a rule of thumb where in the event that the no-break case and one-break case give different results we prefer the onebreak case if the structural break is statistically significant and in the event that the one-break case and two-break case give different results we prefer the two-break case if the second structural break is statistically significant. This rule of thumb has been employed in previous studies that have compared no-break and one-break results (see eg. Lean & Smyth 2013c, 2013d). For the Lee and Strazicich (2003) LM tests, the break date is selected by searching for all possible break dates over the trimming region (0.15T, 0.85T), where T is the sample size. We present the results for the Schmidt and Phillips (1992) LM test for zero to four lags. For the Lee and Strazicich (2003) LM tests the lag length is selected using the general-to-specific approach suggested by Hall (1994) with a maximum of 4 lagged terms. Once the optimal lag 6

7 length at each combination of breakpoints is decided, we determined the breaks where the LM t-test statistic is at a minimum. We apply the panel LM unit root test with up to two breaks to regional panels. Taylor and Sarno (1998) suggested that rejection of the null hypothesis of a joint unit root using panel data tests might be due to as few as one of the series being stationary. To address this issue, we also apply the LM unit root test with no breaks and with one and two structural breaks to smaller panels excluding those countries for which Models A, AA, C and CC rejected the unit root null in hydroelectricity consumption. Data We have annual data on hydroelectricity consumption, measured in million tonnes oil equivalent (mtoe), for 55 countries for the period 1965 to The data source is the BP Statistical Review of World Energy. Table 1 shows the 55 countries, divided into six regions. Countries in Europe/Eurasia and the Asia Pacific constitute about 70 per cent of the sample. We computed the ratio of each region s consumption as a proportion of world consumption. Descriptive statistics are presented in Table 2. Europe/Eurasia has the highest mean consumption, while the Middle East has the lowest mean consumption over the sample period. The distribution for Africa and the Asia Pacific is significantly not normal at the 1 per cent level. 2 Table 3 presents descriptive statistics on total consumption of hydroelectricity for each of the 55 countries. Mean total consumption is highest in the United States and Canada and also exceeds 40 mtoe in Brazil and China. Mean total consumption is lowest in Denmark The standard deviation, which can be seen as a proxy for volatility in consumption, is highest 2 Based on plots of the regional consumption ratios, the consumption ratio is decreasing for developed regions like North America and Europe and Eurasia, while it is increasing for developing/emerging regions like South and Central America, Middle East, Africa and the Asia Pacific. 7

8 in China, followed by Brazil and Canada. The standard deviation is lowest in Denmark. There are just 12 significant series, based on the Jarque-Bera statistic. Results Univariate unit root test with no breaks We started through applying the Schmidt and Phillips (1992) LM unit root test to the data on hydroelectricity consumption. The results are reported in Table 4. There are two test statistics, Z(ρ) and Z(τ). There are 20 countries for which both statistics are significant for lags equal to zero to four (Mexico, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Republic of Ireland, Italy, Portugal, Spain, Sweden, Switzerland, United Kingdom, Algeria, Japan, Taiwan and Thailand). Thus, on the basis of the Schmidt and Phillips (1992) LM unit root test, we reject the unit root null for just over one-third of the sample countries. Univariate unit root test with one break Failure to reject the unit root null for the other approximately two-thirds of countries potentially reflects failure to allow for a structural break in hydroelectricity consumption. Tables 5 and 6 present the results for Lee and Strazicich s (2003) Models A and C, which allow for a structural break in the intercept and structural break in the intercept and slope respectively. Model A provides slightly more evidence than the Schmidt and Phillips (1992) test in favour of stationarity. Specifically, with Model A the unit root null is rejected for 26 countries (Mexico, Chile, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Republic of Ireland, Italy, Spain, Sweden, Switzerland, United Kingdom, Iran, Algeria, Australia, China, Japan, Malaysia, South Korea, Taiwan, Thailand and Other Asia Pacific). Thus, with Table A the unit root null is rejected for almost 50 per cent of countries. Model C provides more evidence in favour of stationarity again. For model C, the 8

9 unit root null is rejected for 41 countries (United States, Mexico, Chile, Ecuador, Belgium and Luxembourg, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Republic of Ireland, Italy, Norway, Poland, Portugal, Romania, Slovakia, Spain, Sweden, Switzerland, United Kingdom, Other Europe and Eurasia, Iran, Algeria, Egypt, South Africa, Australia, Bangladesh, China, India, Japan, Malaysia, New Zealand, Pakistan, Philippines, South Korea, Taiwan and Thailand), representing three quarters of the sample. Given that the results of the no-break case and Model A and Model C differ, it is important to determine which results are to be preferred. First, in terms of comparing Model A with Model C, Sen (2003a, 2003b) showed that Model C produced more reliable estimates on the basis of Monte Carlo simulations. Thus, we prefer Model C to Model A. Second, in choosing between Model C and the no-break case, we prefer the results from Model C if the break in the intercept and/or slope is significant. This was the case for each country for which Model C and the no-break case gave a different result. Thus, we prefer the results for Model C over the no-break case. Hence, after allowing for one break we conclude that the unit root null for hydroelectricity consumption can be rejected for three quarters of the countries. Univariate unit root test with two breaks This still leaves one quarter of the countries for which we do not reject the unit root null for hydroelectricity consumption. Failure to reject the unit root null for these countries could be a by-product of failure to allow for a second break in the series. Tables 7 and 8 present the results for Lee and Strazicich s (2003) Models AA and CC, which allow for two structural breaks in the intercept and two structural breaks in the intercept and slope respectively. Model AA rejects the unit root null for 30 countries (Mexico, Chile, Ecuador, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Republic of Ireland, Italy, Norway, Portugal, Slovakia, Spain, Sweden, Switzerland, Turkey, United Kingdom, Iran, Algeria, China, Japan, Malaysia, Philippines, South Korea, Taiwan and Thailand), which is 9

10 slightly more than Model A. Model CC rejects the unit root null for 44 countries (Canada, Mexico, Brazil, Chile, Ecuador, Venezuela, Austria, Belgium and Luxembourg, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Hungary, Republic of Ireland, Italy, Norway, Poland, Portugal, Romania, Slovakia, Spain, Sweden, Switzerland, Turkey, United Kingdom, Other Europe and Eurasia, Iran, Algeria, Egypt, South Africa, Australia, Bangladesh, China, Japan, Malaysia, New Zealand, Pakistan, Philippines, South Korea, Taiwan, Thailand and Vietnam). This represents 80 per cent of the sample and is a small increase in the number of rejections on Model A. We next reach a conclusion on a preferred set of results based on the no-break case as well as Models A, C, AA and CC. First, in choosing between Models AA and CC, there is no Monte Carlo evidence, of which we are aware, as in the one-break case. However, Model CC is more general than Model AA in that it allows for breaks in the slope. On this basis, we prefer Model CC over Model AA. This has generally been the approach taken in the literature (see eg. Lean & Smyth, 2013b). Second, in choosing between Model CC and a no break/model C hybrid, we prefer Model CC if the second break in the intercept and/or slope is significant. The results of this exercise are summarized in Table 9. The first column summarizes the results comparing Model C with the no break case. As discussed above, for each country for which Model C differs from the no-break case, the break in the intercept or slope is significant. Thus, the no-break/model C hybrid corresponds to Model C with the unit root null rejected for 41 of 55 countries. The second column summarizes the results comparing Model CC with the no-break/model C hybrid. The results are similar to Model CC, with the exception of Austria. Model CC suggests that the unit root null is rejected for Austria. However, as neither the second break in the intercept nor slope is significant, the results for Model C for Austria are preferred. Thus, overall we conclude that on the basis of a Model 10

11 CC/Model C/no break hybrid that the unit root null is rejected for 43 countries, representing 78 per cent of the total. We now turn to the question of why hydroelectricity consumption exhibits a unit root in some countries, but not others. Narayan et al. (2008) hypothesised that energy consumption is more likely to be non-stationary in countries in which energy consumption is more volatile, reflected in a higher standard deviation. The rationale is that in those countries in which energy consumption is more volatile, the level of persistence following a shock will be higher. There is, at best, mixed support for this hypothesis. Of the 12 countries for which the standard deviation in hydroelectricity consumption is greater than three, hydroelectricity consumption is non-stationary in five countries (United States, Argentina, Colombia, Other South and Central America and India). However, hydroelectricity consumption is stationary in each of the three countries with the highest standard deviation (China, Brazil and Canada). An alternative hypothesis, which has been proposed, is that energy consumption is more likely to be non-stationary in countries that are large consumers (Narayan et al., 2008; Barros et al, 2011; Maslyuk & Smyth, 2009). The rationale is that in such countries, shocks will generate larger deviations from the long-run growth path. There is, however, little support for this hypothesis. Of the 11 countries for which average hydroelectricity consumption exceeds 10 mtoe, hydroelectricity consumption is non-stationary in just two countries (India and the United States). Thus, overall, the order of integration of hydroelectricity consumption seems largely unrelated to either the size of consumption or volatility in consumption. This is similar to the conclusion reached by Lean and Smyth (2013a) for renewable energy generation for 115 countries, which is the closest, large-scale multi-country study to this one in terms of the type of disaggregated energy series considered. The one qualification to this conclusion is that, consistent with the 11

12 result for hydropower in Barros et al. (2013b), we conclude that hydroelectricity consumption in the United States is non-stationary. The United States has the highest mean consumption of hydroelectricity and a sizeable standard deviation, which is in excess of eight. As one would expect, given that we apply four models (Models A, AA, C, CC) to hydroelectricity consumption across 55 countries, there is considerable variation in the location of structural breaks. One set of breaks at the beginning of the 1980s, the late 1980s and early 1990s coincide with spikes in oil prices, which were catalysts in the high-income countries to find renewable energy alternatives to fossil fuels. A second set of breaks coincide with the ratification of international treaties, such as the Kyoto Protocol, which provided the stimulus for providing renewable alternatives to fossil fuels, or forums devoted to the advancement of hydropower. A third set of breaks, at the end of the sample period, correspond with international agreements recognising the importance of hydropower. Resulting from the United Nations Symposium on Hydropower and Sustainable Development in 2004, 44 countries signed the Beijing Declaration on Hydropower and Sustainable Development which confirmed the strategic importance of hydropower to sustainable development. Some of these agreements provided particular impetus for the promotion of hydropower in specific regions of the world. For instance, in 2008 a number of international organisations published the World Declaration on Hydropower (Africa), stressing the importance of dam and hydropower development to the continent. As a consequence, international organisations that signed, agreed to assist African countries in joint efforts to promote hydropower development. A fourth set of breaks are linked to policies in specific countries to promote renewable energy as well as the construction of dams in specific countries. As an example of the former, the 12

13 breaks in the United States in 1999 and 2000 coincided with various innovations at the state level, such as the introduction of renewable portfolio standards, that required utilities to generate, or purchase, minimum levels of renewable energy (Lean & Smyth, 2013b). Examples of the latter include the breaks in the 2000s in China, which are associated with the completion of Three Gorges in 2009, Xiluodu in 2009 and Longtan in 2001; the break in 1985 in Brazil, which is associated with the completion of Tucurui in 1984 and the breaks in 1987 and 1989 in Venezuela, which are associated with the construction of Guri in Panel unit root test with up to two breaks We finish by applying the LM panel unit root with up to two breaks, proposed by Im et al. (2005). We first applied the LM unit root test with up to two breaks to a world panel of the 55 countries, as well as regional panels. The unit root null is rejected for the world panel in each case. The unit root null is also rejected for each regional panel, at least in the no break or one break case. In cases, such as South and Central America, in which the unit root null is rejected with no breaks, but not one or two breaks, the likely explanation is that adding one or more breaks when the unit root null is rejected weakens the power of the panel test to reject the unit root null. We also apply the panel unit root test to the world panel, excluding those countries which alternative models (Model A, AA, C, CC, CC/C hybrid) suggest are stationary. The results of this exercise suggest that stationary individual series are not driving the results for the world panel. Excluding stationary individual series, we conclude that the panel unit root null is still rejected, at least in the no-break or one break case. Conclusion Hydroelectricity is already responsible for one fifth of the world s electricity consumption. This, however, still trails fossil fuel combustion as a source of the world s electricity supply 13

14 by a relatively large margin. Concern about the adverse effects of fossil fuel combustion on global warming has led to calls to increase the proportion of electricity generated from hydropower. International resolutions, such as the United Nations sponsored Beijing Declaration on Hydropower and Sustainable Development, see increasing the share of hydroelectricity in the world electricity mix as an important platform of ensuring global sustainable development. This study has applied the LM unit root family of tests to the time series of hydroelectricity consumption in 55 countries to examine the likely success of initiatives and policies designed to increase the share of electricity coming from hydropower. The results suggest that the unit root null is rejected for roughly four fifths of the sample, as well for world and regional panels. Overall, this suggests that shocks to hydroelectricity consumption in most countries will only result in temporary deviations from the long-run growth path. An important consequence of this finding is that initiatives/policies designed to have permanent positive effects on hydroelectricity generation through shocking the long-run growth path, such as large-scale investment in dams and hydrokinetics, are unlikely to be effective in generating continuing annual shocks. This rather pessimistic conclusion is similar to the assessment of the potential for renewable energy in the United States by Apergis and Tsoumas (2011), but differs from the more optimistic conclusions in Barros et al. (2012) and Lean and Smyth (2013a, 2013b). The upside is that we conclude that hydroelectricity consumption is nonstationary in some relatively large consumers of hydroelectricity, which are pushing policies to further increase the share of hydroelectricity in electricity consumption (eg. Argentina, India and the United States). The assessment of the potential for long-term permanent investments in hydroelectricity to reap benefits in these countries is much more rosy. 14

15 References Apergis N, Loomis D and Payne J.E. (2010) Are shocks to natural gas consumption temporary or permanent? Evidence from a panel of US states, Energy Policy, 38, Apergis N and Payne J.E. (2010) Structural breaks and petroleum consumption in US states: are shocks transitory or permanent? Energy Policy, 38, Apergis, N. and Tsoumas, C. (2011) Integration properties of disaggregated solar, geothermal and biomass energy consumption in the US, Energy Policy, 39, Aslan A. (2011) Does natural gas consumption follow a nonlinear path over time? Evidence from 50 states. Renewable and Sustainable Energy Reviews, 15, Aslan A and Kum H. (2011) The stationarity of energy consumption for Turkish disaggregate data by employing linear and nonlinear unit root tests, Energy, 36, Bakis, R. (2007) The current status and future opportunities for hydroelectricity, Energy Sources Part B: Economics, Planning and Policy, 2, Barros C.P., Gil-Alana L.A., Payne J.E. (2011) An analysis of oil production by OPEC countries : persistence, breaks and outliers, Energy Policy, 39, Barros C.P., Gil-Alana L.A., Payne J.E. (2012) Evidence of long memory behaviour in US renewable energy consumption, Energy Policy, 41, Barros, C.P., Gil-Alana, L.A., Payne, J.E. (2013a) Evidence of long-memory behavior in US nuclear electricity net generation, Energy Systems, 4, Barros, C.P., Gil-Alana, L.A., Payne, J.E. (2013b) U.S. disaggregated renewable energy consumption: persistence and long memory behavior, Energy Economics, in press. Ben-David, D., Lumsdaine, R. and Papell, D. (2003) Unit root, postwar slowdowns and long-run growth: evidence from two structural breaks, Empirical Economics, 28, Chamberland, A. and Levesque, S. (1996) Hydroelectricity, an option to reduce greenhouse gas emissions from thermal power plants, Energy Conservation Management, 37, Childress, V.W. (2009) Energy perspective: Is hydroelectricity green? The Technology Teacher, December/January, 4-9. Hall, A. (1994) Testing for a unit root in time series with pre-test data based model selection, Journal of Business and Economic Statistics, 12, Im, K. S., Lee, J. and Tieslau, M. (2005) Panel LM unit root tests with level shifts, Oxford Bulletin of Economics and Statistics, 67, Jia, J., Punys, P. and Ma, J. (2012) Hydropower. In Chen, W-Y., Seiner, J., Suzuki, T. and Lackner, M. (Eds.) Handbook of Climate Change Mitigation. Berlin: Springer. Lean, H.H. and Smyth, R. (2013a) Will policies to promote renewable electricity generation be effective? Evidence from panel stationarity and unit root tests for 115 countries, Renewable and Sustainable Energy Reviews, 22, Lean, H.H. and Smyth, R. (2013b) Are fluctuations in US production of renewable energy permanent or transitory? Applied Energy, 101,

16 Lean, H.H. and Smyth, R. (2013c) Disaggregated energy demand by fuel type and economic growth in Malaysia, Department of Economics Monash University Discussion Paper No. 43/13. Lean, H.H. and Smyth, R. (2013d) Are shocks to disaggregated energy consumption in Malaysia permanent or temporary? Evidence from LM unit root tests with structural breaks, Department of Economics Monash University Discussion Paper No. 7/13. Lee, J. and Strazicich, M. C. (2001) Break point estimation and spurious rejections with endogenous unit root tests, Oxford Bulletin of Economics and Statistics, 63, Lee, J. and Strazicich, M. C. (2003) Minimum Lagrange multiplier unit root test with two structural breaks, Review of Economics and Statistics, 85, Lumsdaine, R. and Papell, D. (1997) Multiple trend breaks and the unit root hypothesis, Review of Economics and Statistics, 79, Maslyuk S. and Smyth R. (2008) Non-linear unit root properties of crude oil production. Energy Economics, 31, Narayan PK, Narayan S and Popp S. (2010) Energy consumption at the state level: the unit root null hypothesis from Australia, Applied Energy, 87, Narayan, P. K., Narayan, S. and Smyth, R. (2008) Are oil shocks permanent or temporary? Panel data evidence from crude oil and NGL production in 60 countries, Energy Economics, 30, Narayan P.K. and Smyth R. (2007) Are shocks to energy consumption permanent or temporary? Evidence from 182 countries, Energy Policy, 35, Nunes, L., Newbold, P. and Kaun, C. (1997) Testing for unit roots with structural breaks: evidence on the great crash and the unit root hypothesis reconsidered, Oxford Bulletin of Economics and Statistics, 59, Perron, P. (1989) The great crash, the oil price shock and the unit root hypothesis, Econometrica, 57, Rosa, L.P. (2001) Rio CHG Working Group Report, International Workshop on Hydro Reservoirs and Greenhouse Gas Emissions, COPPE/UFRJ, January. Schmidt, P. and Phillips, P. C. B. (1992) LM tests for a unit root in the presence of deterministic trends, Oxford Bulletin of Economics and Statistics, 54, Sen, A. (2003a) On unit-root tests when the alternative is a trend-break stationary process, Journal of Business and Economics Statistics, 21, Sen, A. (2003b) Some aspects of the unit root testing methodology with application to real per capita GDP, Manuscript, Xavier University, Cincinnati, OH. Smyth, R. (2013) Are fluctuations in energy variables permanent or transitory? A survey of the literature on the integration properties of energy consumption and production, Applied Energy, 104, Stern, N. (2006) Stern Review on the Economics of Climate Change. London: UK Treasury. Taylor, M. and Sarno, L. (1998) The behavior of real exchange rates during the post-bretton Woods period, Journal of International Economics, 46,

17 Vandel, T. (2012) Hydroelectricity: green and renewable, Policy Options, February. Zivot, E. and Andrews, D. (1992) Further evidence of the great crash, the oil-price shock and the unitroot hypothesis, Journal of Business and Economic Statistics, 10,

18 Table 1: Countries in the sample by region Country/Region Number North America: 3 US, Canada, Mexico South and Central America: 8 Argentina, Brazil, Chile, Colombia, Ecuador, Peru, Venezuela, Other South and Central America Europe and Eurasia: 23 Austria, Belgium and Luxembourg, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Republic of Ireland, Italy, Norway, Poland, Portugal, Romania, Slovakia, Spain, Sweden, Switzerland, Turkey, United Kingdom, Other Europe and Eurasia Middle East: 2 Iran, Other Middle East Africa: 4 Algeria, Egypt, South Africa, Other Africa Asia Pacific: 15 Australia, Bangladesh, China, India, Indonesia, Japan, Malaysia, New Zealand, Pakistan, Philippines, South Korea, Taiwan, Thailand, Vietnam, Other Asia Pacific Total countries 55 18

19 Table 2: Descriptive statistics on the regional consumption ratios Countries Mean Std. Dev. Skewness Kurtosis Jarque-Bera North America South and Central America * Europe and Eurasia Middle East Africa *** Asia Pacific *** * ( ** ) *** denote statistical significance at the 10%, 5% and 1% levels respectively. 19

20 Table 3: Descriptive statistics for hydroelectricity consumption in each country Mean Std. Dev. Skewness Kurtosis Jarque-Bera US Canada * Mexico Argentina Brazil Chile Colombia Ecuador Peru Venezuela Other South and Central America * Austria Belgium and Luxembourg Bulgaria *** Czech Republic Denmark *** Finland France Germany Greece ** Hungary Republic of Ireland Italy Norway Poland Portugal Romania Slovakia * Spain Sweden Switzerland Turkey United Kingdom Other Europe and Eurasia ** Iran *** Other Middle East * Algeria Egypt South Africa Other Africa Australia ** Bangladesh China *** India Indonesia Japan Malaysia New Zealand Pakistan Philippines South Korea Taiwan Thailand Vietnam ** Other Asia Pacific * ( ** ) *** denote statistical significance at the 10%, 5% and 1% levels respectively. 20

21 Table 4: Results of LM unit root test with no break (Schmidt and Phillips, 1992) Lags Countries Z(ρ) Z(τ) Z(ρ) Z(τ) Z(ρ) Z(τ) Z(ρ) Z(τ) Z(ρ) Z(τ) US Canada Mexico ** ** * ** * * * ** ** ** Argentina Brazil Chile Colombia Ecuador Peru Venezuela Other South and Central America Austria Belgium and Luxembourg Bulgaria ** *** *** *** ** *** ** *** ** *** Czech Republic ** ** ** ** ** ** ** ** ** ** Denmark ** ** ** ** * ** * ** * * Finland *** *** *** *** *** *** *** *** *** *** France ** ** ** ** ** ** ** ** ** ** Germany *** *** *** *** *** *** *** *** *** *** Greece ** ** ** ** ** ** ** ** ** ** Hungary *** *** *** *** *** *** *** *** *** *** Republic of Ireland *** *** *** *** *** *** *** *** *** *** Italy *** *** *** *** *** *** *** *** ** ** Norway Poland Portugal *** *** *** *** *** *** *** *** *** *** Romania Slovakia Spain *** *** *** *** *** *** *** *** *** *** Sweden *** *** ** *** ** *** *** *** *** *** Switzerland *** *** *** *** *** *** *** *** *** *** Turkey

22 United Kingdom *** *** *** *** *** *** *** *** *** *** Other Europe and Eurasia Iran * Other Middle East Algeria *** *** ** ** ** *** *** *** *** *** Egypt South Africa Other Africa Australia Bangladesh China * * * * ** ** ** ** India Indonesia Japan *** *** *** *** *** *** *** *** *** *** Malaysia New Zealand Pakistan Philippines South Korea Taiwan *** *** *** *** *** *** *** *** *** *** Thailand *** *** *** *** *** *** *** *** *** *** Vietnam Other Asia Pacific Note: * ( ** ) *** denote statistical significance at the 10%, 5% and 1% levels respectively. 22

23 Table 5: Results of LM test with one break in the intercept (Model A) Series TB K S t-1 B t US ( ) * ( ) Canada ( ) *** ( ) Mexico ** ( ) * ( ) Argentina ( ) ( ) Brazil ( ) *** ( ) Chile ** ( ) ** ( ) Colombia ( ) ** ( ) Ecuador ( ) *** (3.2891) Peru ( ) *** ( ) Venezuela ( ) * (1.4415) Other South and Central America ( ) (0.8559) Austria ( ) (1.2752) Belgium and Luxembourg ( ) * (1.5247) Bulgaria *** ( ) * ( ) Czech Republic ** ( ) * (1.3693) Denmark ** ( ) *** (3.7718) Finland *** ( ) *** (3.0140) France *** ( ) *** (4.1868) Germany *** ( ) ** ( ) Greece ** ( ) ** ( ) Hungary *** ( ) ( ) Republic of Ireland * ( ) * ( ) 23

24 Italy *** ( ) (1.2225) Norway ( ) ** ( ) Poland ( ) *** (2.5522) Portugal ( ) (0.2525) Romania ( ) ** ( ) Slovakia ( ) * (1.3113) Spain *** ( ) *** ( ) Sweden *** ( ) *** ( ) Switzerland *** ( ) ** (1.8156) Turkey ( ) ** (2.2128) United Kingdom *** ( ) ( ) Other Europe and Eurasia ( ) ** (2.0226) Iran ** ( ) ** ( ) Other Middle East ( ) *** (5.3447) Algeria *** ( ) ** (2.1929) Egypt ( ) ** (2.0525) South Africa ( ) * (1.4252) Other Africa ( ) ** ( ) Australia * ( ) * ( ) Bangladesh ( ) (0.9240) China *** ( ) * (1.3059) India ( ) ** (2.1594) Indonesia ( ) * (1.6054) Japan *** ( ) (1.2106) Malaysia * *** 24

25 New Zealand Pakistan Philippines South Korea Taiwan Thailand Vietnam Other Asia Pacific ( ) (4.8132) ** ( ) ( ) * ( ) ( ) ( ) (0.9656) ** *** ( ) ( ) *** ( ) ( ) *** * ( ) ( ) ( ) ( ) * ** ( ) ( ) Notes: Critical values for the LM test at 10%, 5% and 1% significance levels = , and respectively. Critical values for the dummy variable denoting the break date follows the standard asymptotic distribution. * ( ** ) *** denote statistical significance at the 10%, 5% and 1% levels respectively. TB is the break date; K is the lag length; S t-1 is the LM test statistic; B t is the coefficient on the break in the intercept. 25

26 Table 6: Results of LM test with one break in the intercept and slope (Model C) Series TB K S t-1 B t D t US * ( ) (0.5965) *** ( ) Canada ( ) (1.2479) *** ( ) Mexico *** ( ) (0.8806) *** ( ) Argentina ( ) (0.3999) *** (2.6003) Brazil ( ) (1.1471) *** ( ) Chile *** ( ) ** ( ) ** (1.8775) Colombia ( ) (0.0792) *** ( ) Ecuador * ( ) *** (2.7283) ** (1.7449) Peru ( ) (0.0486) *** ( ) Venezuela ( ) (0.7737) *** ( ) Other South and Central America ( ) (1.0029) * ( ) Austria ( ) ( ) ** (2.0342) Belgium and Luxembourg *** ( ) ( ) *** (4.3266) Bulgaria * ( ) ** ( ) (0.1423) Czech Republic * ( ) (0.4686) ** (2.1534) Denmark *** ( ) *** ( ) *** (5.3557) Finland *** ( ) *** (3.1448) ** ( ) France *** ( ) *** ( ) *** (3.9207) Germany ** ( ) (0.8700) (1.2494) Greece ** ( ) ( ) *** ( ) Hungary *** ( ) *** (8.3368) *** ( ) Republic of Ireland *** ( ) (0.2318) *** (4.3604) Italy ** **

27 Norway Poland Portugal Romania Slovakia Spain Sweden Switzerland Turkey United Kingdom Other Europe and Eurasia Iran Other Middle East Algeria Egypt South Africa Other Africa Australia Bangladesh China India Indonesia Japan Malaysia ( ) (1.7108) ( ) *** *** ** ( ) ( ) (2.1480) *** *** ( ) ( ) ( ) ** *** *** ( ) (3.5654) ( ) *** ** ( ) ( ) ( ) *** *** ( ) (0.6872) (4.8808) *** *** ( ) ( ) (0.9227) *** ( ) ( ) (1.0095) *** ( ) ( ) ( ) *** ( ) (0.1988) ( ) *** ( ) (0.8638) ( ) ** *** *** ( ) ( ) ( ) ** *** ( ) ( ) ( ) *** ( ) (1.0644) ( ) *** *** ( ) (1.0728) ( ) ** *** ( ) (0.0949) ( ) *** *** ( ) (3.3647) (0.3934) *** ( ) (1.0802) ( ) ** ** ** ( ) ( ) ( ) *** *** ( ) ( ) (3.8459) ** *** ( ) ( ) (3.1373) * ** *** ( ) (1.8498) ( ) ** ( ) (0.8601) (2.0041) *** ** *** ( ) (1.7798) ( ) * *** ** ( ) (4.6100) ( ) 27