Meta-Analysis of Energy-GDP Causality Literature

An interesting paper by Chi-Chung Chen of National Chung-Hsing U., Taiwan and coauthors was recently published in Energy Policy that carries out a meta-analysis of the energy-GDP causality literature.

I've written about this literature on the blog before and originally worked on the topic in my PhD dissertation resulting in a paper published in Energy Economics. I also have a paper I am working on that investigates this issue in our long-run Swedish dataset.*

There is a very large literature on the topic, but as yet no agreed consensus. Studies in this tradition use dynamic time series models to test for Granger Causality or cointegration between energy use and economic output. Studies vary in their methods, the samples they use, what other variables are included in the model etc. Outcomes are either no causality, GDP causes energy, energy causes GDP, or bidirectional or mutual causation.

Chen et al. use a multinomial logit model to assess the effects of various variables on the probability of a study resulting in each of these four outcomes. Here are the key results:



I've edited their table for clarity... The numbers indicate the change in probability of the given outcome for a unit increase of the variable. So researchers are more likely to find that GDP causes energy in developing countries (Develop) and less likely to find that result in OPEC and Kyoto Annex 1 countries where energy causes GDP is more likely to be found. Larger economies though (GDP) are more likely to have GDP causes energy as are studies with more recent data (that including Year 2000) but higher total energy use (EC) is likely to result in a finding of energy causes GDP.

The method variable takes the value of one when a standard Granger Causality test was used and is zero for other methods. Not using the Granger Causality method is more likely to result in a finding of no causality.

I can certainly buy that using different methods can result in different findings if one method is more appropriate than another. But is there really a difference of this sort between countries of these types? Or is something else going on here? My collaborators, Christian Gross and Stephan Bruns are going to be presenting related work at the IAEE Conference in Perth in June. So I'll be writing more about this topic in future...

* This version of the paper that I presented at a workshop in Michigan last year is only an early draft.