Does Africa grow slower than Asia, Latin America and the Middle East? Evidence from a new data-based classification method

https://doi.org/10.1016/j.jdeveco.2004.05.001Get rights and content

Abstract

We address the question whether sub-Saharan African countries have lower average growth rates in real GDP per capita than countries in Asia, Latin and Middle America and the Middle East. In contrast to previous studies, countries are no a priori assigned to clusters based on geographical location. Instead, we propose a latent-class panel time series model, which allows a data-based classification of countries into clusters such that within a cluster countries have the same average growth rate. Our empirical results suggest that three clusters are sufficient to describe the different growth paths. Twenty-six African countries belong to the low growth cluster, but 8 African countries show growth rates comparable with many countries in Asia, Latin and Middle America and the Middle East.

Introduction

It is an often-heard statement that sub-Saharan Africa is the “lost continent”. Indeed, one of the most distinct and consistent findings from the economic growth literature appears to be that countries on the sub-Saharan African continent grow slower, in terms of real GDP per capita, than countries in Latin and Middle America, Asia and the Middle East and North Africa. Following Barro (1991), this conclusion is largely based on the significance of a 0–1 dummy variable for sub-Saharan Africa (hereafter, Africa) in the typical cross-country growth regression model.

Note that this essentially implies that, after conditioning on other relevant variables (such as investment ratios, initial GDP levels and school enrollment rates), all countries on the continent can be treated as a homogenous group with similar economic performance. However, even though on average African countries may have been showing worse economic performance than, say, Asian and Latin American countries, this obviously need not hold for all countries on the continent. It may very well be that certain African countries in some sense look more like Asian or Latin American countries in terms of economic development, and bear only little resemblance with neighboring countries on the same continent. The aim of this paper is to investigate whether grouping countries based on geography is appropriate or not. For this purpose, we develop a novel methodology for endogenous clustering of countries, where countries are grouped based on the similarity of their GDP growth rates directly, rather than on geographical proximity or other characteristics.

The data set we use concerns annual growth rates of real GDP per capita for 69 countries in Africa, Latin and Middle America, Asia and the Middle East over the 40-year period from 1961 to 2000. Our data set thus amounts to a panel of time series, where both the time series T and the cross-section dimension N are fairly large. Our approach aims to summarize the information in the panel of time series in a concise manner, while allowing for the possibility that countries show similar behavior. We assume that each country has some probability of getting assigned to a latent class (subpopulation), such that within each subpopulation countries have the same economic growth, while growth rates are different across the distinct classes. The data should tell us which countries belong to which class, and also how many clusters there are. We call the model we propose a latent-class panel time series model. The model parameters are estimated using the EM algorithm of Dempster et al. (1977). After parameter estimation, we can use the estimation output to classify countries into subpopulations with similar long-term growth performance.

Our approach is different from and complements previous applications of clustering techniques in the economic growth literature, including Durlauf and Johnson (1995), Desdoigts (1999) and Hobijn and Franses (2000). For example, Durlauf and Johnson (1995) employ classification and regression tree analysis, which relies upon identifying threshold values in particular economic variables to decide upon the appropriate grouping of countries. A similar argument applies to the techniques used in Desdoigts (1999). Our approach is potentially more flexible, in the sense that countries are clustered based on the similarity of their growth rates themselves. Finally, the clustering method developed in Hobijn and Franses (2000) generally leads to (too) many clusters. This might be due to the fact that their method involves many statistical tests, and it is therefore not easy to control the overall size.

It is important to understand the differences in growth rates across countries, as economic growth is of tantamount importance for poverty reduction, see for example Dollar and Kraay (2002). The literature on the sources of growth is extensive with the number of potential determinants that have been explored running in the hundreds, see Barro (1991), Levine and Renelt (1992), Sala-i-Martin (1997), and Temple (1999), among many others. Also, it is important to examine if poverty is increasing or not, see for example, Chen et al. (1994). Knowing which countries get poorer, relative to other countries, also allows one to study the origins of these negative developments. Explanations for the African “growth tragedy” that have been put forward in the literature are diverse, including geographic location, ethnic diversity, choice of institutions, political instability, bad public policy, insufficient infrastructure, limited openness to international trade, and the lack of social capital, see Easterly and Levine, 1997, Easterly and Levine, 2003, Sachs and Warner (1997), Bloom and Sachs (1998), Temple (1998), Collier and Gunning, 1999a, Collier and Gunning, 1999b, Acemoglu et al. (2001), Azam et al. (2002), and Artadi and Sala-i-Martin (2003), among others. In the present paper, we will not investigate the sources of economic growth in detail, although our model is well-suited for this purpose, especially for exploring the issue of parameter heterogeneity, which means to say that the marginal impact of certain growth determinants might be different across (groups of) countries. We return to this point in Section 2.1 below.

An important strand of the economic growth literature concerns the convergence hypothesis, which states that “the poor catch up with the rich”. The convergence literature describes a wide range of techniques, including regression-based methods and distribution-based methods, to examine this hypothesis for various types of data, see Barro and Sala-i-Martin (1992), Quah, 1996, Quah, 1997, Paap and van Dijk (1998), among many others. Durlauf and Quah (1999) and Islam (2003) provide excellent surveys of this extensive literature. It seems that the general conclusion from all these studies is that there is no worldwide convergence, and that there even are indications for divergence. If there is any evidence of convergence, then it is typically found that there are so-called “convergence clubs”. It is important to remark here that the model we propose below, to investigate whether Africa is growing slower than for example Asia, can be applied in case of convergence, in case of divergence and in case of convergence clubs. In that sense, our methodology is rather robust to the type and degree of convergence, if there is any. Indeed, we will examine if there are clusters of countries with similar growth rates. Whether there is convergence or not depends on the starting levels of the country-specific real GDP per capita. Our results shall be indifferent to these initial levels.

The outline of our paper is as follows. In Section 2, we introduce the model we use in our empirical work. For future purposes, we describe estimation and inference for a slightly more general version of it. In Section 3, we discuss the data and present the results of the empirical analysis. We find that the countries in our data set can be adequately clustered into three groups characterized as low, medium and high average growth, with annual growth rates equal to 0.4%, 2.2% and 4.3% per year, respectively. One of our key findings is that, while 26 African countries can be assigned to the low growth cluster, 8 African countries show growth rates which are comparable with many countries in Asia, Latin America and the Middle East. In Section 4, we conclude with some remarks.

Section snippets

Methodology

In this section we introduce the latent class panel time series model. We start with the representation and interpretation of the model, followed by a discussion of the method used for estimation of the model parameters. Throughout, we focus on the empirical application of the model in this paper. However, we discuss the estimation method for a more general version of the model, for future purposes.

Empirical results

In this section we apply the specific version of the model, that is Eq. (1) with Eq. (2) to our data set consisting of annual growth rates in real GDP per capita. We first discuss the data, and next we give detailed empirical results, for the full sample period as well as for sub-periods.

Real GDP per capita is obtained from the Penn World Tables version 6.1., for the 69 countries presented in Appendix A Table A.1. These are 34 countries in sub-Saharan Africa, 14 in Latin and Middle America, 13

Conclusion

In this paper we have proposed to analyze the relative economic performance, in terms of growth rates of real GDP per capita, of a large number of countries in Sub-Saharan Africa, Latin and Middle America, Asia, and the Middle East and North Africa. One of the main assumptions underlying our analysis is the possibility that there would be clusters of countries with similar economic performance. However, we did not want to decide a priori on the number and size of these clusters, let alone

Acknowledgements

We thank two anonymous referees and Rene Segers for helpful comments and suggestions, which substantially improved this paper. Any remaining errors are ours.

References (43)

  • J. Azam et al.

    Explaining slow growth in Africa

    African Development Review

    (2002)
  • C. Azariades et al.

    Threshold externalities in economic development

    Quarterly Journal of Economics

    (1990)
  • R.J. Barro

    Economic growth in a cross-section of countries

    Quarterly Journal of Economics

    (1991)
  • R.J. Barro et al.

    Convergence

    Journal of Political Economy

    (1992)
  • D.E. Bloom et al.

    Geography, demography, and economic growth in Africa

    Brookings Papers on Economic

    (1998)
  • D.E. Bloom et al.

    Geography and poverty traps

    Journal of Economic Growth

    (2003)
  • H. Bozdogan

    Mixture-model cluster analysis using model selection criteria and a new information measure of complexity

  • W.A. Brock et al.

    Growth empirics and reality

    The World Bank Economic Review

    (2001)
  • S.H. Chen et al.

    Is poverty increasing in the developing world?

    Review of Income and Wealth

    (1994)
  • D. Cochrane et al.

    Application of least squares regressions to relationships containing autocorrelated error terms

    Journal of the American Statistical Association

    (1949)
  • P. Collier et al.

    Explaining african economic performance

    Journal of Economic Literature

    (1999)
  • Cited by (0)

    View full text