ABSTRACT
Empirical macroeconomics has highlighted the existence of technology clubs in the world. Are such clubs converging in terms of technological progress, and if so, at what speed? And, what is the role of technology convergence in output convergence? To answer these research questions, we here propose a pragmatic nonparametric test. Using a sample of 81 countries over the 1965–2014 period, we find that the ‘Advanced’ technology club defines the best practice technology in the world, and that the club of the ‘Followers’ but not the ‘Marginalized’ one shows technology convergence. Only in the ‘Followers’ indeed is technology convergence related to output convergence.
Acknowledgements
We thank the Editor Mark Taylor and the three referees for their comments that have improved the paper substantially.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 *We thank the Editor Mark Taylor and the three referees for their comments that have improved the paper substantially.
†HEC Liège, Université de Liège, Belgium. email: [email protected].
At this point it is fair to highlight that working with aggregated data presents some drawbacks such as bias due to structural breaks in the data and cyclicality.
2 An alterative here is to use the approach developed by (Phillips and Sul Citation2009). This approach uses income data to construct the clubs. We prefer here to use variables directly related to technology intensity.
3 In practice, distance functions are easily computed using linear programmings (see e.g (Walheer Citation2016)).
4 For better readability, we give the percentage changes for the technical and technological indicators in .
5 Another option is to select a world technology as the benchmark technology. In that case, a natural question is how to define such world technology. A common practice is to rely on a (convex or non-convex) envelopment of the club technologies (see e.g (Walheer Citation2021). for more discussion). Note that the relative technology gaps between clubs would not been impacted in that case as the benchmark technology is the same for each club. It implies that our results would remain correct.
6 H: 1965 and 2014 distributions are equal; H: 2014 distribution is smaller than 1965 distribution. Note that similar conclusions hold true when relying on a Kuiper test or a Student test for the averages.
7 Note that both ratios contain counterfactual concepts. This might be computationally challenging but not so when using linear programmings (see e.g (Walheer Citation2016)).