World energy intensity revisited: a cluster analysis

Abstract

The aim of this article is to empirically identify convergence clubs in energy intensity among 109 countries from 1971 to 2010 by using a recently developed methodology, i.e., a new regression-based convergence test, introduced by Phillips and Sul (2007). This log t test allows us to endogenously identify the groups of countries that converge to different equilibriums and those that do not converge to any convergence clubs. We mainly find that, first, world countries do not seem to converge at the same steady-state level; instead, they form four separate clubs converging to their own steady-state paths and few countries are found to converge to no group at all. In addition, although the world as a whole shows the evidence of convergence, economic and geographic groups seem to converge at different speeds. Last, estimates from an ordered-logit model reveal that initial energy intensity level and openness are mainly responsible for the formation of the world convergence clubs, whereas industry share and R&D share are not.

Keywords:

• energy intensity
• club convergence
• log t regression test
• nonlinear time-varying factor model

JEL Classification:

• C15
• C23
• Q53
• Q54

Notes

¹ Gamma convergence, or intra-distribution mobility, examines whether the individual countries with the highest intensity and lowest intensity remain the same (Liddle, 2009, 2010).

² For more details about Equation 4, see Phillips and Sul (2007, pp. 1786–1787).

³ It is typical to use both TPES and PPP-converted GDP in energy intensity studies (Mielnik and Goldemberg, 2000; Ezcurra, 2007; Jobert et al., 2010; Liddle, 2010; Camarero et al., 2013).

⁴ Moran (1950) proposed a test statistic to assess the degree of spatial autocorrelation between adjacent locations. It is defined in a matrix form as I = (ZʹWZ)/(ZʹZ), where Z is the variable of interest (energy intensity in this study) and W is a symmetric matrix that can take several forms. In this study, the spatial weight matrix is based on the centroid distance between each pair of country i and country j.

⁵ See Cameron and Trievedi (2005), Greene (2003), and Jung (1993) for more details on the ordered-logit approach.

⁶ In the ordered-logit model, the probability (or odds ratio) is an exponential function of the estimated logistic coefficients and hence can be computed by exponentiating the logistic coefficients. For instance, the odds ratio of the initial energy intensity is: exp(–1.735) = 0.18.

Additional information

Funding

This work was supported by the Fundamental Research Funds for the Central Universities and the Research Funds of Renmin University of China [grant number 13XNJ017].