A new paper by Kalimeris et al. is the third meta-analysis of the energy GDP causality literature. The previous two studies are Chen et al. and, of course, our own paper in press at the Energy Journal. This paper covers more studies in fact than either of the previous two papers, 158, but the number of individual tests analyzed is not much greater than in our study. Like Chen et al. the authors only classify results according to the direction of causality and not the magnitude or significance of the test statistics.
The authors attempted to use "Rough Set Data Analysis", which attempts to find decision rules in poorly defined data sets. The results show that there are no well supported conclusions about causal directions in the meta-sample. The main method used in the paper is like Chen et al. a multinomial logit regression analysis. Using this approach, Kalimeris et al. find, as we did too, that the cointegration techniques are more likely to find causality in some direction than are other techniques. This makes sense as there must be Granger causality in at least one direction in order to find cointegration between the variables. But they could not come to any general conclusions about the direction of causality.