Structural differences in economic growth: an endogenous clustering approach

Abstract

This article addresses heterogeneity in determinants of economic growth in a data-driven way. Instead of defining groups of countries with different growth characteristics a priori, based on, for example, geographical location, we use a finite mixture panel model and endogenous clustering to examine cross-country differences and similarities in the effects of growth determinants. Applying this approach to an annual unbalanced panel of 59 countries in Asia, Latin and Middle America and Africa for the period 1971–2000, we can identify two groups of countries in terms of distinct growth structures. The structural differences between the country groups mainly stem from different effects of investment, openness measures and government share in the economy. Furthermore, the detected segmentation of countries does not match with conventional classifications in the literature.

Acknowledgements

We thank Jan Willem Gunning and seminar participants at the Tinbergen Institute Rotterdam, ESRC Development Economics Conference (2008) at University of Sussex, and 6th EUDN PhD Seminar at University of Göttingen for helpful comments.

Notes

¹ Alfo et al. (2008) also consider 10 and 25 year averages of the data and stress the problem of country heterogeneity being masked particularly in the latter case.

² This choice of regressors is by no means exclusive. Some regressors used in several growth regressions such as population density or squared inflation are not included in the model as a result of data availability or the presence of high multicollinearity with the other regressors.

³ The EM algorithm may converge to a local maximum. To prevent reporting local maximum results, we use 4000 different random starting cluster probabilities. For all considered models, the estimation results belonging to the highest log-likelihood value are reported.

⁴ The correlation between the human capital and real GDP growth rate is also negative when these two variables are analysed in a separate regression, that is, when real GDP per capita growth rates are regressed on a constant and secondary school enrolment rates.

⁵ From the countries considered, one can argue that the performance of Japan over the time span considered is rather different from the remaining countries. For this reason we also estimated the models excluding Japan from the sample. The general results in terms of the optimal number of clusters, the signs and the significance of the explanatory variables remain the same.

⁶ Initial openness measure is the trade percentage at the beginning of the sample period, and for initial schooling we use secondary school enrolment rates in the population over 25. Similar to Davis et al. (2009), we use European settler mortality rates by Acemoglu et al. (2001) (revised by Albouy (2008)) as a proxy for institutional quality.

⁷ The comparisons with other institutional quality proxies in Albouy (2008), such as latitude, also do not show a clear relationship with the cluster memberships.

⁸ The methodology by Lee and Kim (2009) can also be seen as an exogenous clustering approach using income levels as a threshold variable. They consider different growth patterns across countries depending on the World Bank criterion for the upper middle-income countries. However, our results do not indicate such a classification of countries either.

⁹ This group consists of Middle Eastern and Asian countries. We refer to this group as ‘Asia’ in the article.