Club Convergence in the Energy Intensity of China.

Author:Zhang, Dayong

    The recent high speed of economic growth in China has generated an enormous growth in the demand for energy. Minimizing the energy intensity (1) of China, which is normally defined as energy consumption per unit of GDP, is therefore a crucial objective in order for the country to maintain a sustainable long-term economic growth path. In recent years the Chinese government has realized the importance of this issue and put forward a number of stringent initiatives to help keep growth in energy consumption to a minimum. The 'Twelfth Five-Year Plan' for example contains clear targets for reducing energy intensity, setting energy consumption out as one of the most important strategic challenges facing China over the coming years. Figure 1 shows the trend in nationwide energy intensity from 1953 to 2008 using data from the China Statistical Yearbook, and re-affirms several features highlighted in previous studies: the increased policy debate in recent years has helped support a clear trend in favor of less intensive energy consumption since 1980 (see Zhou et al., 2010; Wu, 2012); though there have been some fluctuations in the nationwide trend from 2003 onwards as noted by Liao et al. (2007). In the existing literature, the aggregate pattern of China's energy intensity has attracted significant attention, with numerous studies trying to explain how the decline has been achieved, and why there have been fluctuations in recent years (for example Chai et al., 2009; Zhao et al., 2010).

    Energy structure is among the various factors that can affect the energy intensity of an economy. The primary source of energy in China is coal, which accounts for more than 60% of total energy used (Feng et al., 2009). One feature of coal production in China is that it is not evenly distributed across the provinces and has clear regional patterns, consequently regional energy intensities will differ to account for issues such as the additional 'energy costs' of energy supply arising from moving coal from where it is extracted to where it is needed as a power/heat source. While it is important to look at the macro-level behavior of energy intensity in China, given the arguments above it might also be necessary to look deeper, at a disaggregated level. China is a vast country and has clear regional differences both geographically and economically. Energy intensity at the regional level has shown a strong heterogeneity across provinces. A recent study by Yu (2012) indicates the huge gaps which exist between the most and the least energy intensive regions in China: the province with the highest energy intensity is approximately three times that of the province with the lowest intensity. Yu (2012) goes on to suggest that acknowledging regional disparities and using provincial level data to study the factors influencing energy intensity in China are an important direction for future research.

    Economic development across China has been demonstrated in numerous studies to have clear geographical patterns. Provinces in the eastern coastal areas are generally more economically developed and, according to Yu (2012), tend to have lower energy intensity than the western regions, which are far less developed, even though they are richest in natural resources. During the recent years of rapid economic growth, the gap between these regions has narrowed. Yang and Fang (2008) provide evidence of convergence in the level of energy productivity (known as Beta convergence) and suggest that the overall disparity has gradually decreased. However, they found no significant evidence of convergence in the variability of energy productivity across the regions (known as Sigma convergence). Many regional level studies for China tend to divide the provinces into three core geographical regions, namely, the eastern, western and central regions. This type of division has become commonly used in the literature (Wu, 2012), though is rather arbitrary in nature. In this paper, we adopt a new panel data method proposed by Phillips and Sul (2007) to study provincial level energy intensity in China, which enables us to cluster members of the panel (the provinces) into convergent groups. The method provides a simple but effective way to study the behavior of energy intensity in transition (i.e. over time) allowing for common groups to emerge while maintaining individual specific heterogeneity. Thus, using a nonlinear time-varying framework, our work re-visits the question/nature of energy intensity in China, allowing provincial groupings to be empirically determined rather than choosing them purely on geography. Hence we can question if the east/west/center division commonly used is actually supported by the data.

    Our analysis reveals three convergence clubs but that are different with the simple geographical divisions that have been applied in previous related studies. Thus, in order to try and better understand what commonalities or disparities do lay behind their grouping we undertake a regression analysis of the determinants of energy intensity. The regression results reveal significant club specific differences, even after controlling for club-specific fixed effects, which give rise to a number of policy relevant implications. Not all determinants are common across the three clubs, and where they are common, they can differ both in magnitude and sign, reflecting the fundamental differences across the groups. As a closing feature of our analysis we question if the identified clubs are likely to further converge towards a single group in the coming decade, using simple forecasting methods. We find no evidence that such convergence will occur, thereby reinforcing the value of identifying and understanding club specific features.

    The paper is structured as follows: Section 2 reviews the relevant literature on studies in convergence and their application to energy; Section 3 explains briefly the methodology; Section 4 describes the source and characteristics of our data and reports empirical results; the last two sections discuss the main findings, with policy implications, and then conclude the paper.


    The concept of convergence was originally introduced to economics within the economic growth literature. In one of the earliest related studies, Baumol (1986) suggested that the economic growth of the US at the time was slowing down, allowing other countries to catch up with their economic position. There are three types of convergence which have emerged in the literature: Beta convergence, Sigma convergence, and Stochastic convergence, for a detailed survey of the related literature on convergence refer to Islam (2003), here we merely summarize some core aspects. Beta convergence refers to convergence in the value of something, such as per-capita income, and is sometimes also referred to as the catch-up effect. For instance, poor economies grow faster than the rich allowing them to catch-up and converge in the level of income. Barro and Sala-i-Martin (1992) first introduced the related concept of Sigma convergence, which refers to a reduction of the 'dispersion' in the associated variables of interest, or convergence in variance as opposed to the level. Finally, Stochastic convergence, pays attention to convergence in the time series properties of variables, i.e. whether they share common dynamics or not (Quah 1990, 1993; and Carlino and Mills, 1993).

    Quah (1996) gives a general review of the early literature/methods on convergence, in particular those looking into club convergence and 'twin peak' convergence. The leading conclusion of this study was that existing research had placed too strong a reliance on one set of countries converging towards another, and that methods trying to establish a more general type of convergence e.g. twin-peaking or multi-modal peaking "... continue to miss the principal important features..." (p.13). While Quah (1996) offered a possible solution he also concluded that "Many issues remain to be researched in this alternative approach." (p.13). Phillips and Sul (2007) ultimately follow in this line of research, offering an approach to club-convergence that borrows many of the same underlying concepts e.g. utilizing transition dynamics and reviewing the full cross-sectional distribution of convergence, but in such a way that empirical testing of the convergence is possible. This offered an innovative twist on the topic of club convergence, providing a more defendable way of classifying a set of panel data into two or more clubs. In addition to its strong statistical properties, the Phillips and Sul (2007) procedure is also simple to implement, increasing its appeal.

    There have been a considerable number of studies extending the idea of convergence beyond the growth literature and applying it to other fields, including energy and environmental topics. Aldy (2006), Ezcurra (2007a) and Panopoulou and Pantelidis (2009) for example studied the convergence of carbon dioxide emissions using cross country data. Barassi et al. (2011) alternatively studies the stochastic convergence of cross country emissions, taking explicit account of potential long-memory dynamics. A number of other studies look at the cross country level convergence of energy data including: Markandya et al. (2006), Ezcurra (2007b), Liddle (2010), Le Pen and Sevi (2010), Duro and Padilla (2011), Mohammadi and Ram (2012), Meng et al. (2013). (2) Cumulatively, these studies consider all three concepts of convergence mentioned above (Beta, Sigma and Stochastic). However they fail to discern a common conclusion, and find what can only be described as mixed evidence. Notwithstanding this, a common belief supported by most of these studies is that there is no common global convergence, but that it might exist when looking across specific regions. A natural extension of the empirical work was then to...

To continue reading