Electricity intensity across Chinese provinces: New evidence on convergence and threshold effects

Energy Economics 36 (2013) 268–276

Contents lists available at SciVerse ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneco

Electricity intensity across Chinese provinces: New evidence on convergence and threshold effects M.J. Herrerias a,⁎, G. Liu b a b

University of Nottingham, School of Contemporary Chinese Studies, United Kingdom Brunel University, United Kingdom

a r t i c l e

i n f o

Article history: Received 17 March 2012 Received in revised form 25 August 2012 Accepted 26 August 2012 Available online 1 September 2012 JEL classification: Q43 Q47 Q50 O11 R11 Keywords: Energy intensity Convergence China provinces Unit roots

a b s t r a c t Energy intensity has gone through different stages across Chinese regions. In this paper, we investigate the stochastic electricity-intensity convergence across the Chinese provinces. Unlike previous work, this paper highlights the relevance of the level of technology of each province and takes into account the economic geography through the examination of club convergence. We perform several unit root tests that introduce structural breaks, nonlinearities and time variation, with the aim to capture the economic transformation of the Chinese economy. Results indicate that the majority of the Chinese regions have converged according to the unit-root tests in time-series analysis, indicating that technological differences diminish over time. However, this convergence pattern occurs within groups of regions, according with club convergence test. Indeed, we find a dominant club and others smaller clubs that few regions belong. However, it is observed that there are regions that still diverge. These findings support our argument that special policy attention is required for those regions displaying divergence. © 2012 Elsevier B.V. All rights reserved.

1. Introduction One of the most active debates in energy economics is an observed decline in energy intensity across developed countries over the last two decades. Works focusing on the causes of the decline have revealed that technological progress, economic structure, sectoral decomposition of energy use, fuel mix, efficiency in the conversion and end-use of energy account for the improvements in the use of energy resources across countries (Liddle, 2010).1 However, whether differences in energy intensity across countries diminish over time, achieving convergence, has received little attention in the literature, especially compared with related fields like environment and growth (Le Pen and Sévi, 2010). Within this literature, convergence in energy intensity could imply that technological differences across regions diminish over time. By contrasts the finding of divergence may be a motive to promote energysaving policies. Thus, examining this issue may provide new insights to the energy-economics literature.

⁎ Corresponding author at: The University of Nottingham, School of Contemporary Chinese Studies, Jubilee Campus, United Kingdom. E-mail address: [email protected] (M.J. Herrerias). 1 See also Chai et al. (2009), Cornillie and Fankhauser (2004), Fisher-Vanden et al. (2004), Garbaccio et al. (1999), and Sinton and Levine (1994). 0140-9883/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.eneco.2012.08.026

Energy intensity in the fast growing economies is of great importance due to the fact that these countries require a large amount of energy resources to fuel that process. Among them, the case of China is probably the most fascinating one since its economy has achieved an impressive performance over the last two decades. The rapid growth has resulted in a huge increase in demand for electricity in recent years. 2 Thus, an interesting question to investigate is the relationship between energy consumption and growth through the examination of energy intensity levels in this economy. The observed decline in energy intensity has also generated a large amount of empirical research in the Chinese case (Ma and Stern, 2008; Ma et al., 2009; Zhang, 2003). These works conclude that during the 1980s and 1990s the use of energy per unit of output at the national level improved. This trend changed, after that period until recent years, showing a deterioration of efficiency levels of energy (Zhao et al., 2010). However, the existence of these two trends makes necessary that in the empirical analysis either structural breaks or nonlinearities have to be introduced in the model. However, none of previous works takes into account this issue. In addition, these studies investigate the whole country leaving the regional dimension out of such analysis. 2 See Wang et al. (2010) and Zhang et al. (2011) for a review of China's energy situation.

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

We extend the previous works by introducing endogenous structural breaks and nonlinearities for each region, we consider not only the aforementioned two trends, but also the new improvement in the use of electricity in 2009–2010. Ignoring these changes, may cause bias in any estimation especially in the case of China that is continuously and gradually transforming its economy. On the other hand, the regional dimension presents new directions on energy aspects in this country. First, there are some energy exporting regions like Guangxi and Shanxi that display a sharp decrease in electricity intensity in the supply side in the nineties, but for the remaining regions a steady pattern is observed. In the energy importing regions, the behavior is quite irregular e.g. we observe some of these regions with a high level of energy intensity, or others that either has improved it or remains relatively stable over time. These differences in energy-intensity might show differences in economic structure and technologies. The Chinese government, regarding reduction in energy intensity as a desirable way of limiting the incremental environmental damage associated with the rapid economic growth, established in the 11th Five-Year Plan the objective of reducing energy intensity by 20% between 2005 and 2010. This objective is expected to achieve through optimization of the industrial structure and improvement of efficiency and reduction in consumption. However this expectation has not yet been fully realized. 3 It is well-known that energy intensity is a measure of the direct link between energy consumption and economic activity, which in turn is related to emissions and environmental protection. Thus, analyzing the convergence behavior in energy intensity across Chinese provinces may lead to new insights. This is the goal of this work. To carry out research, we focus on the concept of stochastic convergence, which has relied on unit-root tests in time-series analysis and apply it to the case of electricity-intensity across Chinese regions over the period 2003–2009 with monthly data. We perform several unit root test. First, we use Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test and Ng and Perron tests, which are some of the traditional test without breaks used in the literature, and then improve the power of such tests by introduction of endogenous structural breaks and nonlinearities according to Kapetanios et al. (2003) and Lee and Strazicich (2003). This allows us to consider a situation where each region has its own level of technology in China, which is a realistic assumption. This analysis is complemented by investigating the existence of club convergence through the application of the Hansen (2000) and Phillips and Sul (2007) tests, since the uneven distribution of energy resources in the vast territory in the Chinese economy makes us expect the presence of regional clusters. Analyzing energy-intensity convergence among Chinese regions is important because the decrease in energy intensity is mainly due to changes in economic structure and technological progress and these factors have possible impacts on institutions and flows comprising international trade. If the first influence predominates, namely, structural change- then, trade, may lead to divergence in energy intensity. However if technological progress dominates, then, trade by encouraging or facilitating best-efficiency practice, could lead to energy intensity convergence. 4 Furthermore, the finding that both developed regions and the less developed ones are converging toward a common pattern of energy use could be evidence of a “leapfrogging process” (Liddle, 2010). By contrast, finding divergence in energy intensity may cause damage to the credibility of local and national governments in their attempt to reduce the use of energy. In addition, it could imply a lack of diffusion of energy-related technologies. This is important not only for the ability of China to improve environment, but also for resources saving that is critical for sustainable development. In this 3 See Chai and Zhang (2010) and Zhang and Wang (2008) for the need to promote energy saving measures in China. 4 The finding of convergence could also be interpreted as convergence in technological progress across regions.

269

case, local and national governments may prioritize energy-saving measures across sectors i.e. providing incentives and technological policies aiming at encouraging knowledge diffusion, particularly in regions with lower energy efficiency. In existing research, there is little evidence on convergence in energy intensity across Chinese regions. 5 However, we can find it for other, developed and developing, countries in Ezcurra (2007), Le Pen and Sévi (2010), Liddle (2010) and Markandya et al. (2006) among others. 6 Our study distinguishes itself from previous work in many significant ways. First, we explore the novel aspect of stochastic electricity intensity convergence, which is the most important energy resource in the Chinese economy. Second, we perform our study at the provincial level, since one of the characteristics of the Chinese economy is its heterogeneity across regions. Besides, we use a unique and rich monthly data. Third, we use two traditional unit-root tests as well as the most recent developments that allows for structural breaks and nonlinearities in the data to take into account the significant transformation of Chinese regions over our sample period. 7 In addition, given that some of Chinese regions are much larger than many countries, we analyze convergence within a group of regions— in which we group regions or provinces into a ‘club’ according to their stage of economic development. This test is particularly relevant for economies in transition, since it allows a wide range of transition paths toward the steady state. Finally, we contribute to the literature in providing empirical evidence of notable convergence across regions, when unit-root tests that allow either structural breaks or nonlinearities in the data-generating process are introduced in the model, since the traditional ones (KPSS and Ng and Perron tests) support different conclusions. However, we also find that Chinese regions converge into clubs, namely, to their own steady state, except a few places like Liaoning, Tianjin, and Yunnan. These findings suggest that a common energy policy applied to Chinese provinces may provide unsatisfactory results, implying the need to design specific energy policies according to the clusters found. The plan of the paper is as follows. Section 2 reviews the notions of convergence and their applications in the energy intensity literature. Data and methodology are presented in Section 3. Results are explained in Section 4, while conclusions are drawn in Section 5. 2. The notion of convergence The concept of convergence in growth empirics is related to the reduction of inequality between countries or regions. However, convergence is not restricted to the growth literature, and has been applied recently to other fields, including energy economics. One of the most active lines of research in energy economics has focused on the factors of the decline in energy intensity (Chai et al., 2009; Cornillie and Fankhauser, 2004; Goldemberg and Prado, 2011; Ma and Stern, 2008; Sinton and Levine, 1994; Sun, 2003; Zhang, 2003; Zhao et al., 2010) and more recently has examined whether that decline has favored the convergence process across developed and developing countries (Duro et al., 2010; Ezcurra, 2007; Ian, 2008; Le Pen and Sévi, 2010; Liddle, 2010; Markandya et al., 2006). As stated by Islam (2003), different notions of convergence are linked to different methodological approaches. Among them, we can distinguish on the one hand, absolute versus conditional β-convergence and on the other hand σ-convergence according to the seminal paper by Barro and Sala-i-Martin (1992). Often these notions have been tested empirically, initially with cross-sectional data and latter with panel 5 Ito et al. (2010) investigated the energy demand in China from regional aspects, but these authors do not analyze the hypothesis of energy intensity convergence. 6 See additional works for other countries in Liddle (2009), Mielnik and Goldemberg (2000), Mulder and De Groot (2007) and Sun (2002). 7 See Kebede (2011) for the need to test for unit root test in an application of the efficiency of US natural gas market and by using a similar notion of convergence in the case of carbon dioxide emissions in Lee and Chang (2008).

270

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

data. However, this approach has been criticized due to doubts about the validity of the empirical methods that are used (Quah, 1993). Instead, Quah in 1993 suggests analyzing the entire shape of the distribution and changes in the dynamics of the distribution, since a negative coefficient in initial income is a necessary but not sufficient condition to conclude about the presence of convergence, especially when the distribution becomes multimodal. The last notion of convergence relies on time-series analysis, where it is tested through the application of unit-root tests. This paper takes this approach as its starting point in order to examine the convergence behavior of energy intensity across Chinese provinces over the period 2003 to 2009. This literature goes back to early works by Bernard and Durlauf (1995, 1996), Carlino and Mills (1993), Li and Papell (1999) and Loewy and Papell (1996), where the concepts of stochastic and deterministic convergence are introduced. Stochastic convergence in time-series analysis in energyeconomics literature implies that the energy intensity in one region relative to that of the economy as a whole is trend stationary. On the other hand, deterministic convergence requires the data-generating process to be level stationary, which in turn requires that both the deterministic and the stochastic trends are canceled out (Li and Papell, 1999, p. 274). An alternative definition of stochastic convergence is proposed by Bernard and Durlauf (1995), who defines convergence between two (or more) countries/regions as occurring when the long-run forecast of energy intensity differences tends to zero while the forecasting horizon tends to infinity. In the bivariate case, this requires the energy intensity to be cointegrated with a cointegrating vector (1,−1). These authors find little evidence of convergence among a large sample of countries, applying Augmented-Dickey Fuller test. Recent developments made by econometricians, introducing structural breaks in the datagenerating process or nonlinearities, find more evidence in favor of convergence (Kapetanios et al., 2003; Lee and Strazicich, 2003). However, the rejection of the convergence hypothesis across countries such as in Bernard and Durlauf (1995, 1996) does not necessary imply divergence since countries may still be in the transitional dynamics (Oxley and Greasley, 1995). Indeed, these authors refine the concept of stochastic convergence into long-run convergence and catching-up. The former refers to the attainment of long-run steady-state equilibrium in the energy intensity differential between two regions, while the latter refers to the situation in which the narrowing of the energy-intensity gap between the two regions is observed over time, but the convergence process has yet to be completed. From an empirical point of view, long-run convergence implies that the series are level stationary (without a trend) and caching-up is trend stationary. Although, the improvements in the power of unit root tests have been useful to test convergence in time-series analysis, within this framework the uneven distribution across space of the economic activities has been ignored. This fact is relevant since recent developments in economic geography have provided fresh evidence on the importance of the agglomeration of economic activities and its linkage with economic growth (Fujita and Thisse, 2002; Krugman, 1991). As economic activities require energy resources for the production process, the role of economic geography in the energy-economics literature provides interesting insights. In this regard, a panel-data test proposed by Phillips and Sul (2007) takes explicitly this issue into account by examining club convergence in a nonlinear time-varying framework. Besides, this test imposes a non-homogenous technology across regions and allows for a wide range of transition paths, which are especially feasible for economies in transition. We also complement this analysis by performing the Hansen (2000) test for each individual region to provide the full picture of clustering behavior. Previous evidence on energy-intensity convergence is quite limited in literature that there is no evidence in the case of Chinese provinces. However, there is evidence for other developed and developing countries such as in Liddle (2010) and Markandya et al. (2006) who investigate respectively energy intensity convergence in transition economies

and world energy intensity convergence in a similar fashion to Barro-type regressions. Their results show considerable convergence across countries. Ezcurra (2007) focusing on the dynamics of the distribution also finds similar results over the period 1971–2001. By contrast, Le Pen and Sévi (2010) find that stochastic convergence is rejected using a pair-wise test for a group of 97 countries in the period 1971–2003. Thus, this paper contributes to the literature by analyzing the stochastic energy-intensity convergence across Chinese provinces from 2003 to 2009 with monthly data. Analyzing stochastic convergence instead of other notions of convergence has several advantages. First, it allows us to investigate the behavior of each region individually that in turn make flexible the assumption of heterogeneous technology. Second, compared with other notions of convergence, the stochastic approach distinguishes not only convergence from divergence, but another possibility that may exist, that is, regions can be in the transitional dynamics. Third, by introducing endogenous structural breaks, threshold effects and non-linearities it is possible to take into account the significant transformation of the Chinese economy, making robust our results. Finally, considering the crosssectional dependence, by applying the club convergence test developed by Phillips and Sul (2007), it is possible to identify endogenously the cluster formation in the vast territory. The results of such analysis are relevant due to their implications for the sustainability of growth and for the energy and environmental policies across Chinese provinces as well as for the economy as a whole. In addition, knowing the behavior of electricity intensity of each region energy policies can be addressed in a more efficient way by local governments. If the cluster formation is present, it can be indicative that common policies applied to the whole country may fail since each cluster converges toward specific-steady state. In this circumstance, it is necessary to implement common energy policies for each cluster detected. 3. Data and methodology 3.1. Data We investigate whether differences in electricity intensity across the Chinese regions diminish over time. In doing this, we consider electricity intensity (Electricity Consumption/Industrial Output) from 2003(1) to 2009(12). The source of this monthly data is from the National Bureau of Statistics of China. In Figs. 1–6 in Appendix A, we can see that the efficiency of electricity-intensity has shown different stages. First, we can observe a smooth decline from 2003 to 2009, which is more pronounced in 2008. However, in 2009 two distinctive trends are observed. In the first half of 2009 the electricity-intensity of consumption increased, while in the second half of that year is found an improvement in the use of electricity. The regional dimension provides additional insights. We observe some notable differences across regions. For example, it seems that there are at least two groups of regions, one shows a steady pattern over the considered period and other displays an irregular behavior. Among the latter, Liaoning presents a decline in electricity-intensity, while Qinghai and Ningxia show the highest levels of electricityintensity with a modest decline over time until 2008, changing their trend after that period. There is a small group of provinces in an intermediate position, where appears regions such as Gansu, Guizhou, Inner Mongolia, Shanxi and Yunnan. Thus, in terms of these regional differences, our empirical strategy consists of several types of unit-root tests on convergence. Specifically, we use two well-known unit-root tests like KPSS and Ng and Perron tests. These tests allow us to use the results as benchmark since the former changes the null hypothesis to stationarity. Although the latter has a higher power than both the Augmented Dickey Fuller and Phillips–Perron tests, the absence of breaks in its estimate may

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

generate a misleading result when the rate of convergence varies over time. 8 The latest generation of unit-root tests considers the existence of one or two breaks or the presence of nonlinearities in the data-generating process. These tests provide more reliable results. Specifically, we use Lee and Strazicich (2003) for the former and Kapetanios et al. (2003) for the latter. However, such tests can uncover the convergence behavior of each individual region, but do not take into account the significant heterogeneity of Chinese provinces. Thus, by using recent tests developed by Hansen (2000) and Phillips and Sul (2007) in a nonlinear time-varying framework, we take explicitly into account the heterogeneity of regions as well as a novel aspect such as the identification of convergent clubs. With this set of techniques we attempt to capture the significant transformation of the Chinese economy over time. 3.2. Methodology 3.2.1. Lee and Strazich test Lee and Strazicich (2003) proposed a unit root test with endogenously-determined breaks. This test, unlike the traditional ones, takes into account the structural breaks in the data-generating process in a consistent manner, both under the null and the alternative hypotheses. The two-break minimum Lagrange multiplier can be estimated by means of the following regression: ΔY t ¼ d′ ΔZ t þ ϕ S˜ t−1 þ

p X

γ1 Δ S˜ t−1 þ εt

ð1Þ

t¼1

˜ x −Z t δ, ˜ t = 1, 2; δ˜ are coefficients in the regression where S˜ t ¼ Y t − ψ ˜ and Y1 and Z1 denote the first ˜ x is given by Y 1 −Z 1 δ, of ΔYt on Zt; ψ observations of Yt and Zt, respectively. The lagged terms Δ S˜ t−1 are included to correct for serial correlation. The selection of the lagged terms is carried out following the general-to-specific method of Perron and Ng (1996). The vector of exogenous variables Zt is defined according to the testing specification employed. Considering two breaks in both level and trend, the Zt vector is given by Zt = [1, t, D1t, D2t, DT1t, DT2t]′, where Djt for t ≥ TBj + 1,j = 1, 2 and 0 otherwise, and DTjt = t for t ≥ TBj + 1, j =1,2 with TBj denoting the points at which breaks occur. The unit-root hypothesis is tested using the t-ratio of ϕ in Eq. (1), being denoted as ˜τ. The break points are determined endogenously by considering all points   over the interval [0.1T, 0.9T]. Denoting the break fracTB tions as λj ¼ T j , the LM statistic is defined as: LMτ ¼ inf τ˜ ðλÞ:

ð2Þ

As shown in Lee and Strazicich (2003), critical values depend on the location of breaks. 3.2.2. Kapetanios et al. test Kapetanios et al. (2003) develop a non-linear unit root test with the aim of remedying the persistent failure of the standard Dickey– Fuller (DF) test to reject the null of a unit root. These authors provide an alternative framework for testing the null of a unit-root process against an alternative non-linear exponential smooth transition autoregressive (STAR) process, which is globally stationary.9 Consider a univariate smooth-transition autoregressive model of order 1 (STAR(1)): Y t ¼ ϕY t−1 þ ϕY t−1



n o 2 þ εt : 1− exp −γY t−d

As suggested by Kapetanios et al. (2003) – KSS hereafter – Eq. (3) can be conveniently re-parameterized as:  n o 2 ΔY t ¼ βY t−1 þ ϕY t−1 1− exp −γY t−d þ εt

ð4Þ

where β = ϕ − 1. Imposing β = 0 and d = 1, our specific STAR model is:  n o 2 þ εt ΔY t ¼ ϕY t−1 1− exp −γY t−1

ð5Þ

where εt ≈iid (0,σ 2) In order to test the null hypothesis of a unit root H0 : γ = 0 against H0 : γ b 0, Kapetanios et al. (2003) propose a Taylor approximation of the STAR model, given that in practice the coefficient γ cannot be identified under H0. Thus, under the null, the model becomes: 3

ΔY t ¼ δY t−1 þ ηt

ð6Þ

where ηt is an error term. We may now apply a t-test to analyze whether Yt is a non-stationary process, H0 : δ = 0, or whether it is a non-linear stationary process, such that. H0 : δ b 0. 3.2.3. Phillips and Sul log t test One of the advantages of tests based on panel data is that they are able to consider the heterogeneity across individuals and their evolution over time. Recently it has been developed a test based on a nonlinear time-varying form by Phillips and Sul (2007) that takes into account this issue as well as the formation of regional clusters. These two features, makes this test appropriate for the search of convergence across Chinese regions. The starting point of the test is a simple factor model as in Eq. (7): Y it ¼ δi μ t þ it

ð7Þ

where δi measures the idiosyncratic distance between some common factor μt and the systematic part of Yit. This model seeks to capture the evolution of the individual Yit in relation to μt by means of its two idiosyncratic elements, that is, the systematic element δi and the error it. Phillips and Sul (2007) modify this initial model by allowing the systematic idiosyncratic element to evolve over time, thereby accommodating heterogeneous agent behavior and evolution within that behavior by means of a time-varying factor-loading coefficient δit. Furthermore, they allow δit to have a random component, which absorbs it in Eq. (7) and allows for possible convergence behavior in δit over time in relation to the common factor μt. The time-varying behavior of δit is modeled in a semi-parametric form as follows: −1 −α

δit ¼ δi þ σ i εit Lðt Þ

t

ð8Þ

where δit is fixed, εit is i.i.d (0,1) across i but weakly dependent on t, and L(t) is a slowly-varying function (like log t) for which L(t) tends to infinity and t also goes to infinity. This formulation ensures convergence of the parameter of interest for all α ≥ 0, which is the null hypothesis of interest. 10 Then, the null hypothesis of convergence is set as: H 0 ¼ δi ¼ δ and α ≥ 0:

ð3Þ

Against the alternative: H 1 ¼ δi ≠δ f or all i or α b 0:

8

See Herrerías and Orts (2011) for further discussion on the unit root tests in this context. 9 Yt follows a non-linear but globally stationary process provided by −2 b γ b 0.

271

10

See additional technical details in Phillips and Sul (2007).

272

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

At the 5% level, the null hypothesis of convergence is rejected if the statistic has a value below − 1.65. However, the interesting aspect of this approach is that convergence patterns within groups can be examined using log t regressions, that is, testing for the existence of club convergence and then clustering. This fact is particularly relevant since the rejection of the null of convergence does not necessarily imply divergence, since different scenarios can be met, such as separate points of equilibrium or steady-state growth paths, as well as convergence clusters and divergent regions in the full panel. The existence of club convergence raises an important concern, that is, how to identify the regions that belong to each cluster. In this regard, Phillips and Sul (2007) suggest the following method. In the first step, individuals in the panel must be ordered according to the last observation in the panel. In the second step, the so-called “core group”, Gk, should be identified by selecting the first k with the highest value of the variable of individuals in the panel to form the subgroup Gk for some N > k ≥ 2, and then the log t regression is run and the convergence test statistic tk(Gk) is obtained for this subgroup. Then, the core group size k* is chosen by maximizing tk over k according to the criterion: 

function with I(qt ≤ γ) = 1 if the event qt ≤ γ occurs and I(qt ≤ γ) = 0 otherwise. The indicator function sets xt(γ) = xtdt(γ) so that model (9) equals: yt ¼ θ′ xt þ δ′n xt ðγ Þ þ et

ð10Þ

where δn = θ2 − θ1 denotes the “threshold effect” and θ = θ2. Eq. (10) allows all of the regression parameters to switch between regimes. The parameters of interest in the two-regime TR model can be estimated conveniently by sequential conditional least square. Under the additional assumption that the e's are distributed normally, the resulting estimates are equivalent to maximum likelihood estimates. Conventional tests of the null of a linear model against the TR alternative (H0 : θ2 = θ1) have non-standard distributions, as the threshold parameter is not identified under the null of linearity. Hansen (2000) proposed a heteroskedasticity-consistent F-test bootstrap procedure to test the null of linearity. The independent variables xt are supposed to be fixed, and the dependent variable is generated by a bootstrap ^ t Þ, where u ^ t is the OLS residual from the from the distribution Nð0; u estimated threshold model. Hansen (2000) showed that this procedure yields asymptotically correct p-values.

k ¼ argmaxðt k Þ; subject to minðt k Þ > −1:65:

4. Results

The latter condition ensures that the null hypothesis of convergence is supported for each k. The rule for classifying the groups of regions into clubs is straightforward. For example, if all the regions belong to the same group, then the size of the club will be N. In contrast, if there are regions that do not belong to that group, the clusters will have a size lower than N. More formally, this implies that if the condition min(tk) > − 1.65 does not hold for k = 2, then the highest individual in Gk can be dropped from each subgroup and new subgroups are created. This process is repeated as many times as necessary until the condition is satisfied. If at the end of this process there are subgroups that have been created (said to be club convergent), but there are others that do not satisfy the condition, then it is said that those individuals diverge.

Results based on KPSS and Ng and Perron tests are presented in Table 1. From this table, evidence on convergence is rather mixed. On the one hand, we find strong evidence of convergence in electricity intensity by using the KPSS test, however the opposite result is observed when the Ng and Perron test is performed. These unsatisfactory results are likely to be accounted for by the unequal power of these tests to detect the unit root. Thus, to improve those findings, first we

3.2.4. Threshold effects, Hansen test We extend previous analysis by performing Hansen test to analyze stochastic club convergence among Chinese regions. The threshold model allows the sample of provinces to be split up into different groups, as well as the construction of convergence clubs. Hansen (2000) proposed a test for the null hypothesis of a linear regression against the alternative of a Threshold Regression (TR) given by: yt ¼ θ′1 xt þ et yt ¼ θ′2 xt þ et

qt ≤γ qt >γ

ð9Þ

where yt is a dependent variable and xt corresponds to a vector of p × 1 independent variables. The random variable et is a regression error. The threshold variable qt may be an element of xt and is assumed to have a continuous distribution. The TR model assumes that the regime is determined by the value of qt relative to a threshold value, which we denote γ. The threshold variable is used to split the sample of provinces into different groups or clusters. When the explanatory variables xt are lagged values of the endogenous variable, the TR corresponds to Tong's (1983, 1990) Threshold Autoregressive (TAR) model. A special case arises when the threshold variable qt is taken to be a lagged value of the endogenous variable itself, qt = yt − d. In this case, the resulting model is called a Self-exciting TAR (SETAR) model. Hansen (2000) derived a useful asymptotic approximation to the ^ of the threshold parameter distribution of the least-square estimate γ γ. Following Hansen, the TR model can be written in a single equation by using a dummy variable dt = I(qt ≤ γ), where I(•) is the indicator

Table 1 KPSS test and Ng and Perron test, electricity intensity. Regions

Beijing Tianjin Hebei Shanxi Inner Mongolia Liaoning Jilin Heilongjiang Shanghai Jiangsu Zhejiang Anhui Fujian Jiangxi Shandong Henan Hubei Hunan Guangdong Guangxi Hainan Sichuan Guizhou Yunnan Shaanxi Gansu Qinghai Ningxia Provinces converging

KPSS

Ng–Perron test

Constant

Trend

Constant

Trend

0.26 0.12 0.34 0.16 0.32 3.92** 0.53* 0.97** 0.47* 0.45 0.90** 0.09 0.20 0.28 0.26 0.26 0.34 0.93** 0.30 0.79** 0.32 0.28 0.11 0.53* 0.76** 1.47** 1.85** 0.29 17

0.13 0.13 0.27** 0.13 0.09 0.24** 0.13 0.26** 0.10 0.13 0.25** 0.10 0.12 0.12 0.26** 0.12 0.16* 0.41** 0.11 0.16** 0.05 0.11 0.10 0.04 0.16** 0.45** 0.27** 0.14 17

0.57 0.84 0.28 0.37 0.33 0.60 0.53 0.87 0.35 0.31 0.20* 0.98 1.13 0.42 0.47 1.01 0.64 0.35 0.13** 1.10 1.61 0.84 0.12** 0.89 0.51 0.42 0.40 0.27 3

0.30 0.30 0.23 0.23 0.23 0.44 0.16* 0.34 0.22 0.24 0.19 0.95 1.14 0.40 0.25 0.65 0.23 0.23 0.63 0.70 1.29 0.11** 0.12** 0.10** 0.25 0.22 0.22 0.24 4

Note: ** denotes the rejection of the null hypothesis at 1% and * at 5%. KPSS refers to Kwiatkowski–Phillips–Schmidt–Shin test. See Perron and Ng (1996) for further details on the test.

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276 Table 2 Lee and Strazicich unit root test, electricity intensity by region related to the whole nation.

273

Table 3 KSS test, electricity intensity.

Regions

Break 1

Break 2

t-stat.

Results

Regions

t-stat.

t-stat trend

Results

Beijing Tianjin Hebei Shanxi Inner Mongolia Liaoning Jilin Heilongjiang Shanghai Jiangsu Zhejiang Anhui Fujian Jiangxi Shandong Henan Hubei Hunan Guangdong Guangxi Hainan Sichuan Guizhou Yunnan Shaanxi Gansu Qinghai Ningxia Provinces converging

2007:9 2007:9 2005:10 2005:1 2006:10 2005:2 2007:10 2006:9 2007:10 2007:10 2005:4 2007:11 2007:11 2004:10 2006:8 2005:1 2006:9 2006:12 2007:11 2007:11

2008:11 2008:11 2007:9 2008:12 2009:1 2007:12 2008:11 2008:11 2008:11 2008:11 2008:11 2008:12 2008:12 2008:12 2008:11 2009:1 2008:11

−8.89** −7.12** −5.50*** −7.41** −8.44** −5.88* −7.65** −6.88** −9.69** −8.32** −6.39** −8.04** −9.07** −6.87** −6.22** −7.79** −7.98** −3.74 −8.59** −9.72** −6.80** −6.57** −6.81** −7.73** −6.16** −4.67 −5.17 −5.74*

C C C C C C C C C C C C C C C C C D C C C C C C C D D C 25

Beijing Tianjin Hebei Shanxi Inner Mongolia Liaoning Jilin Heilongjiang Shanghai Jiangsu Zhejiang Anhui Fujian Jiangxi Shandong Henan Hubei Hunan Guangdong Guangxi Hainan Sichuan Guizhou Yunnan Shaanxi Gansu Qinghai Ningxia Provinces converging

−4.89** −4.80** −3.16*** −6.60** −7.02** −6.15** −4.68** −4.85** −4.95** −4.64** −5.36** −4.73** −5.43** −6.01** −4.59** −6.06** −4.80** −4.30** −5.51** −6.40** −7.35** −7.81** −5.52** −6.42** −5.56** −3.89* −2.73 −4.07**

−0.76 −0.52 −0.67 −2.07 −2.50 −5.44** −1.28 −1.24 −0.84 −1.24 −2.53 −1.87 −2.80*** −3.15*** −1.31 −1.90 −1.56 −2.51 −0.12 −1.97 −0.41 −2.59 −1.44 0.07 −2.84*** −1.60 −0.65 −1.51

LRC LRC LRC LRC LRC CUP LRC LRC LRC LRC LRC LRC CUP CUP LRC LRC LRC LRC LRC LRC LRC LRC LRC LRC CUP LRC D LRC 27

2005:1 2007:8 2004:7 2005:1 2004:11 2005:4 2005:8

2009:1 2009:3

2008:3 2007:4 2008:12 2007:11 2008:9 2008:7

Note: ** denotes the rejection of the null hypothesis at 1% and * at 5%. C = Convergence and D = Divergence.

Note: ** denotes the rejection of the null hypothesis at 1% and * at 5% and *** at 10%. D means divergence, LRC means long-run convergence, and CUP means caching up. Critical values were taken from Chong et al. (2008).

use a unit-root test that takes into account the structural breaks in the data-generating process such as the one developed by Lee and Strazicich (2003) and second, the one that introduces nonlinearities, developed by Kapetanios et al. (2003). Besides, the latter allows us to distinguish convergence from divergence, but also an additional state, namely, the catching-up process. These two tests may help us in capturing the significant transformation of the Chinese economy in the recent decades. Table 2 reports the results based on the test with structural breaks. From there, we can observe that our findings provide empirical evidence in favor of the convergence in the majority of the regions in terms of electricity intensity. There are few exceptions such as in Hunan, Gansu and Qinghai. Moreover, we find that the global crisis (2007–2008) influenced the majority of regions since it represents the main structural break found. Besides, through the examination of the structural breaks detected it is possible to observe the two distinctive trends in energy intensity from the early 2000 up to date. 11 Examples of these cases are shown in Figs. 2 and 4 where a deterioration of electricity intensity in Zhejiang, Qinghai and Ningxia in recent years is detected. However, at the same time the presence of groups of regions that share similar characteristics or by contrast regions displaying a non-homogenous behavior is detected. We will address this issue next by the identification of the different clubs. However, we can improve this finding by analyzing this issue through the application of the Kapetanios et al. (2003) test. The goal of this test is that it allows us to consider the transitional dynamics in addition to the common finding of convergence/divergence. Results

of this analysis are shown in Table 3. Our findings indicate that 23 regions display long-run convergence, that is, the majority of regions in China. Besides, we find that 4 regions (Liaoning, Fujian, Jiangxi and Shaanxi) are in the transitional dynamics, and Qinghai diverges from the national level. Here, the most important consumers of electricity display long-run convergence, but those four regions along with Qinghai need special energy saving policies to help them converge to the national benchmark in terms of energy intensity. Finally, the size of Chinese regions is often larger than many countries, and regional clusters can be found. Indeed, as we stated earlier in the figures in Appendix A, it is possible to see that this empirical observation such as in Liaoning and Shanxi (Fig. 1), Zhejiang and Fujian (Fig. 2), Qinghai and Ningxia (Fig. 4) and Yunnan and Guizhou (Fig. 5) displays a different behavior than for example in those regions that appear in Fig. 3, which show a more homogenous trends. In addition to the different economic policies applied to each region, the transmission grid may also create small regional clubs. If this is the case, a common policy on these aspects may fail since clubs display very distinctive behavior, and it is necessary to design specific sets of policies for each cluster. Results of this analysis are presented in Table 4, where the Phillips and Sul (2007) log t regression is reported. From there, we can conclude that the hypothesis that all the regions form a unique group is rejected, implying that regional clubs are present. We find 3 clubs and a divergent group that comprises Tianjin, Liaoning and Yunnan. More specifically, a dominant club (club 2) is identified and belongs to 20 regions. Although this is a satisfactory result since such club contains the majority of the regions, the own existence of clubs makes necessary the implementation of a set of measures to improve electricity intensity such as the promotion of technological progress, changes in economic structure and reforms oriented to a more market economy. To conclude the empirical analysis, we performed the Hansen test to investigate the existence of threshold effects across the Chinese

11 From econometric point of view, these two distinctive trends make that a simple linear regression analysis fails to account the structural transformation of the Chinese economy. For this reason, the only econometric tool capable to capture these trends is on the one hand, the introduction of endogenous structural breaks and on the other hand, allowing for a non-linear form in the model with time varying coefficients.

274

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

Table 4 Convergence club, electricity intensity. Convergence club 1 log t 0.23 (0.16) Convergence club 2 log t- 0.26 (−19.47) Convergence club 3 log t 0.53 (3.79) Divergence club

Zhejiang, Hainan Beijing, Hebei, Shanxi, Shanghai, Jiangsu, Anhui, Fujian, Jiangxi, Shandong, Henan, Hunan, Guangdong, Sichuan, Shaanxi, Gansu, Qinghai, Ningxia, Inner Mongolia, Guangxi, Guizhou Jilin, Heilongjiang, Hubei Tianjin, Liaoning, Yunnan

Note: See further details on the test in Phillips and Sul (2007). In the below map appears the clubs identified according with our empirical results.

Table 5 Hansen test. First threshold

Shanghai Beijing Tianjin Zhejiang Guangdong Shandong Fujian Hubei Inner Mongolia Shaanxi Jiangsu Henan Sichuan Liaoning Jilin Heilongjiang Jiangxi Qinghai Ningxia Hebei Hainan Anhui Gansu Guizhou Yunnan Guangxi Shanxi Hunan China

Second threshold

F-test

P-value

Year

F-test

P-value

Year

11.10 16.09 11.84 17.46 24.46 20.08 9.52 9.13 12.57 9.92 7.24 21.42 42.80 10.22 9.91 11.15 24.36 7.39 14.79 8.37 13.16 23.47 20.88 8.11 6.75 25.47 25.32 11.85 9.61

0.20 0.02 0.14 0.01 0.00 0.00 0.34 0.38 0.10 0.31 0.63 0.00 0.00 0.24 0.27 0.20 0.00 0.63 0.04 0.43 0.07 0.00 0.00 0.44 0.60 0.00 0.00 0.16 0.29

2006:10 2007:5 2007:1 2008:10 2004:1 2007:8 2008:9 2006:5 2004:5 2008:3 2007:7 2007:5 2008:11 2008:9 2008:9 2006:8 2008:4 2008:6 2004:11 2005:7 2004:2 2006:9 2007:9 2008:10 2004:1 2004:3 2007:2 2004:12 2006:3

3.61 6.12 3.08 6.07 7.48 6.53 3.01 2.35 4.00 1.92 1.41 7.92 16.72 2.50 2.63 2.32 8.76 1.69 4.67 2.52 3.28 9.25 6.16 2.05 1.36 9.77 9.90 4.13 2.21

0.09 0.03 0.15 0.01 0.00 0.00 0.17 0.31 0.06 0.47 0.70 0.00 0.00 0.26 0.23 0.32 0.00 0.57 0.04 0.25 0.13 0.00 0.01 0.39 0.66 0.00 0.00 0.06 0.33

2005:12 2005:5 2005:9 2003:12 2008:9 2005:2 2004:1 2006:5 2008:5 2004:7 2005:3 2005:5 2003:11 2004:1 2004:1 2006:2 2004:6 2004:4 2007:11 2007:3 2008:8 2006:1 2005:1 2003:12 2008:9 2008:7 2005:8 2007:10 2006:7

Note: The bold numbers indicate that the existence of the threshold is significant.

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

steady state. However, the own existence of clusters, some of them comprising small number of regions, also is indicative that a common energy policy for all the regions may fail, since different groups of regions show different time paths. From an economic policy point of view, it is necessary to design specific energy saving policies for each of the clusters detected to improve the efficiency of electricity consumption. This should allow that small clusters are able to strengthen their reduction in their differences from the efficiency-frontier clubs over time, and therefore improving energy efficiency as a whole for the Chinese economy.

Appendix A

0,2300 0,2100 0,1900 0,1700 0,1500 0,1300

2009-9

2009-1

2009-5

2008-9

2008-5

2008-1

2007-5

2007-9

2006-9

2007-1

2006-1

2006-5

2005-9

2005-5

2005-1

2004-5

2004-9

2004-1

2003-9

China

Fig. 1. Electricity intensity, China national.

1.0000 0.9000 0.8000 0.7000

Beijing

0.6000 Tianjin

0.5000 0.4000

Hebei

0.3000

Shanxi

0.2000

Liaoning

0.1000

Jilin 2009-9

2009-4

2008-6

2008-11

2008-1

2007-8

2007-3

2006-10

2006-5

2005-7

2005-12

2005-2

2004-9

2004-4

2003-11

2003-6

2003-1

0.0000

Fig. 2. Electricity intensity, Chinese provinces.

0.2500

0.2000 Heilongjiang 0.1500 shanghai Jiangsu

0.1000

Zhejiang 0.0500

Anhui Fujian 2009-9

2009-4

2008-11

2008-6

2008-1

2007-8

2007-3

2006-5

2006-10

2005-7

2005-12

2005-2

2004-9

2004-4

2003-11

0.0000 2003-1

There is a large piece of empirical research on an observed decline in energy intensity across developed and developing countries. Most of this literature has focused on the factors behind such decline, but whether differences in energy intensity diminish over time, achieving convergence, has received little attention among economists. However, analyzing the convergence process across countries or regions is of great importance, since the finding of convergence could imply a diffusion of energy-related technologies across regions, while divergence could be a reason to promote energy-saving measures, especially in those regions which display lower levels of energy efficiency. In this context, the objective of this paper was to investigate the convergence behavior in energy intensity across Chinese provinces over the period 2003–2009. Specifically, we focused on the notion of stochastic convergence that relies on the use of unit-root tests. We applied it to the case of electricity intensity that is the most important source of energy in this country. In the best of our knowledge, there is no previous empirical evidence that explores this issue across Chinese regions. In addition, we use a rich provincial and monthly data set that allows us to analyze this notion of convergence through the application of unit-roots tests, which provide information for each region on energy intensity. This approach relaxes the hypothesis of homogeneous technology across regions in China. Finally, in order to take into account the well-known economic transformation of the Chinese economy, we introduce in the model endogenous structural breaks and nonlinearities in the data-generating process. Although the results of this analysis by themselves are rich, we complement them by investigating the existence of club convergence. The latter aspect is especially relevant in the case of the Chinese economy due to the uneven distribution of energy resources across regions. Our results show considerable empirical evidence in favor of convergence across the Chinese regions in terms of electricity intensity. Besides, the structural changes detected capture adequately the two distinctive trends in energy intensity in the case of each Chinese region, that is, a deterioration of efficiency levels of energy in more recent years. However, our results also suggest that this convergence occurs within clubs, although some regions still show divergence. Among the clubs detected, we find a predominant cluster, which contains a large number of regions that are converging toward a specific-

2003-1

0,0900

5. Conclusions

2003-5

0,1100

2003-6

regions. This information is presented in Table 5. From there, we can observe that only in 14 regions were detected such threshold effects, which implies that during the period analyzed these provinces undergone significant changes in terms of electricity intensity, but also reveal that half of the Chinese regions need further efforts to reform the energy sector. Mainly these changes are linked with energy reforms during 2005 and 2006 and the global crisis in 2007–2008. This result could be related with the aforementioned clubs, and therefore confirms the idea that specific policies should be addressed to each cluster identified to improve electricity intensity levels across the vast territory of China. To sum up, from these results three essential conclusions emerge. First, due the existence of clubs, a common energy policy for the whole country will provide less satisfactory results. Secondly, a convergent club (club 2) contains a large number of regions, which means that previous energy policies have helped reduce the differences across regions, but given that some other regions belong to small clusters, more effort is needed to make them join the dominant clubs. Finally, those regions that display divergence from any of the clubs, especially Liaoning, require active regional policies to join the leading cluster. This is important since the finding of convergence in energy intensity could imply convergence in technological progress across regions, which can have spillovers in other fields in the domestic economy and raise the credibility of China and its regions in the national and international commitments on environmental issues.

275

Fig. 3. Electricity intensity, Chinese provinces.

276

M.J. Herrerias, G. Liu / Energy Economics 36 (2013) 268–276

0.3500 0.3000 0.2500 Jiangxi 0.2000

Shandong

0.1500

Henan

0.1000

Hubei Hunan

0.0500

Guangdong 2009-4

2009-9

2008-11

2008-1

2008-6

2007-8

2007-3

2006-5

2006-10

2005-12

2005-2

2005-7

2004-9

2004-4

2003-6

2003-11

2003-1

0.0000

Fig. 4. Electricity intensity, Chinese provinces.

1.2000 1.0000 Sichuan 0.8000

Shaanxi Gansu

0.6000

Qinghai 0.4000 Ningxia 0.2000

Inner Mongolia

2009-7

2009-1

2008-7

2008-1

2007-7

2007-1

2006-7

2006-1

2005-7

2004-7

2005-1

2004-1

2003-1

2003-7

0.0000

Fig. 5. Electricity intensity, Chinese provinces.

1.2000 1.0000 0.8000 Guangxi

0.6000

Hainan 0.4000

Guizhou Yunnan

0.2000

2009-9

2009-4

2008-6

2008-11

2008-1

2007-8

2007-3

2006-5

2006-10

2005-7

2005-12

2005-2

2004-9

2004-4

2003-11

2003-6

2003-1

0.0000

Fig. 6. Electricity intensity, Chinese provinces.

References Barro, R., Sala-i-Martin, X., 1992. Convergence. J. Polit. Econ. 100, 223–251. Bernard, A.B., Durlauf, S.N., 1995. Convergence in international outputs. J. Appl. Econ. 10, 97–108. Bernard, A.B., Durlauf, S.N., 1996. Interpreting tests of convergence hypothesis. J. Econ. 71, 161–173. Carlino, G., Mills, L., 1993. Are US incomes converging? A time series analysis. J. Monet. Econ. 32, 335–346. Chai, Q., Zhang, X., 2010. Technologies and policies for the transition to a sustainable energy system in China. Energy 35, 3995–4002. Chai, J., Guo, J., Wang, S., Lai, K., 2009. Why does energy intensity fluctuate in China? Energy Policy 37, 5717–5731.

Chong, T.T., Hinich, M.J., Liew, V.K., Lim, K.P., 2008. Time series test of nonlinear convergence and transitional dynamics. Econ. Lett. 100, 337–339. Cornillie, J., Fankhauser, S., 2004. The energy intensity of transition countries. Energy Econ. 26, 283–295. Duro, A., Alcántara, V., Padilla, E., 2010. International inequality in energy intensity levels and the role of production composition and energy efficiency: an analysis of OECD countries. Ecol. Econ. 69, 2468–2474. Ezcurra, R., 2007. Distribution dynamics of energy intensities: a cross-country analysis. Energy Policy 35, 5254–5259. Fisher-Vanden, K., Jefferson, G.H., Liu, H., Tao, Q., 2004. What is driving China's decline in energy intensity? Resour. Energy Econ. 26, 77–97. Fujita, M., Thisse, J.-F., 2002. Economics of Agglomeration. Cambridge University Press, Cambridge. Garbaccio, R.F., Ho, M.S., Jorgenson, D.W., 1999. Why has the energy-output ratio fallen in China? Energy J 20, 63–91. Goldemberg, J., Prado, L., 2011. The decline of the world's energy intensity. Energy Policy 39, 1802–1805. Hansen, Bruce E., 2000. Testing for structural change in conditional models. J. Econometrics 97 (1), 93–115. Herrerías, M.J., Orts, V., 2011. The driving forces behind China's growth. Econ. Transit. 19, 74–124. Ian, S.W., 2008. Explaining the declining energy intensity of the US economy. Resour. Energy Econ. 30, 21–49. Islam, N., 2003. What have we learnt from the convergence debate? J. Econ. Surv. 17, 309–362. Ito, T., Chen, Y., Ito, S., Yamguchi, K., 2010. Prospect of the upper limit of the energy demand in China from regional aspects. Energy 35, 5320–5327. Kapetanios, G., Shin, Y., Snell, A., 2003. Testing for a unit root in the non-linear STAR framework. J. Econ. 112, 359–379. Kebede, Y., 2011. Testing for unit roots, causality, cointegration, and efficiency: the case of the northwest US natural gas market. Energy 36, 3489–3500. Krugman, P., 1991. Geography and Trade. MIT Press, London. Le Pen, Y., Sévi, B., 2010. On the non-convergence of energy intensity: evidence a parwise econometric approach. Ecol. Econ. 69, 641–650. Lee, C., Chang, C., 2008. New evidence on the convergence of per capita carbon dioxide emissions from panel seemingly unrelated regressions augmented Dickey–Fuller test. Energy 33, 1468–1475. Lee, J., Strazicich, M., 2003. Minimum LM unit root test with two structural breaks. Rev. Econ. Stat. 85, 1082–1089. Li, Q., Papell, D., 1999. Convergence of international output, time series evidence for 16 OECD countries. Int. Rev. Econ. Finance 8, 267–280. Liddle, B., 2009. Electricity intensity convergence in IEA/OECD countries: aggregate and sectoral analysis. Energy Policy 37, 1470–1478. Liddle, B., 2010. Revisiting world energy intensity convergence for regional differences. Appl. Energy 87, 3218–3225. Loewy, M., Papell, D., 1996. Are US incomes converging? Some further evidence. J. Monet. Econ. 38, 587–598. Ma, C., Stern, D., 2008. China's changing energy intensity trend: a decomposition analysis. Energy Economics 30, 1037–1053. Ma, H., Oxley, C., Gibson, J., Kim, B., 2009. Modeling China's energy consumption behavior and changes in energy intensity. Environ. Model. Softw. 24, 1293–1301. Markandya, A., Pedroso-Galinato, S., Streimikiene, D., 2006. Energy intensity in transition economies: is there convergence toward the EU average? Energy Economics 28, 121–145. Mielnik, O., Goldemberg, J., 2000. Converging to a common pattern of energy use in developing and industrialized countries. Energy Policy 28, 503–508. Mulder, P., De Groot, H., 2007. Sectoral energy- and labour productivity convergence. Environ. Resour. Econ. 36, 85–112. Oxley, L., Greasley, D., 1995. A time-series perspective on convergence: Australia, UK and USA since 1870. Econ. Rec. 71, 259–270. Perron, P., Ng, S., 1996. An Autoregressive Spectral Density Estimator at Frequency Zero for Nonstationarity Tests. Cahiers de recherche 9611, Centre interuniversitaire de recherche en économie quantitative, CIREQ. Phillips, P., Sul, D., 2007. Transition modeling and econometric convergence tests. Econometrica 75, 1771–1855. Quah, D.T., 1993. Empirical cross-section dynamics in economic growth. Eur. Econ. Rev. 37, 426–434. Sinton, J., Levine, M., 1994. Changing energy intensity in Chinese industry, the relative importance of structural shift and intensity change. Energy Policy 239–255 (March). Sun, J.W., 2002. The decrease in the difference of energy intensities between OECD countries from 1971 to 1998. Energy Policy 30, 631–635. Sun, W., 2003. Three types of decline in energy intensity—an explanation for the decline of energy intensity in some developing countries. Energy Policy 31, 519–526. Tong, H., 1983. Threshold Models in Nonlinear Time Series Analysis: Lecture Notes in Statistics. Springer, Berlin, p. 21. Tong, H., 1990. Non-Linear Time Series: A Dynamical System Approach. Oxford University Press, Oxford. Wang, J., Bloyd, C., Hu, Z., Tan, Z., 2010. Demand response in China. Energy 35, 1592–1597. Zhang, Z., 2003. Why did the energy intensity fall in China's industrial sector in the 1990s? The relative importance of structural change and intensity change. Energy Economics 25, 625–638. Zhang, J., Wang, G., 2008. Energy saving technologies and productive efficiency in the Chinese iron and steel sector. Energy 33, 525–537. Zhang, N., Lior, N., Jin, H., 2011. The energy situation and its sustainable development strategy in China. Energy 36, 3639–3649. Zhao, X., Ma, C. And, Hong, D., 2010. Why did China's energy intensity increase during 1998–2006: decomposition and policy analysis. Energy Policy 38, 1379–1388.