Evidence on Income Convergence: A Global Analysis

Abstract

The theoretical debate on the question of the poor becoming the rich, or income convergence between countries, was mainly initiated with the introduction of the neoclassical growth model by Solow (1956) and Swan (1956). However, the convergence empirics are more recent and span just over a quarter of a century. This thesis aims at analyzing the multiple notions of income convergence for the world group of countries and within its various geographic and income clusters. The geographic clusters are Africa, Asia, Europe and Latin America & Caribbean, while, high, upper middle, lower middle and low income are the four income clusters in the study. Both β-convergence and σ-convergence are examined in their absolute and conditional forms. Absolute β-convergence is defined as the convergence towards similar levels of per capita income across countries in the long run. Therefore, absolute β-convergence involves a negative relationship between the growth rate of income and initial income while considering the cross-country steady state levels of income as constant. On the other hand, conditional β-convergence is defined as convergence towards respective steady state levels of income of countries. Conditional β-convergence takes into account the country specific geographic, structural and socio-economic variables and, thus, requires the equality of income growth rates rather than income levels. In other words, conditional β-convergence entails a negative relationship between income growth and initial income after controlling for differences in steady state income levels of countries. In contrast, σ-convergence is defined as a reduction in income dispersion among countries over time.

The absolute β-convergence is estimated by applying the non-linear least squares technique both with cross-sectional and panel data sets using the variables of GDP per capita and GDP per worker for the period 1950-2008. The estimation of a logarithmic trend regression for the income dispersion is the underlying methodology for analyzing σ-convergence. In addition, the dynamic panel data system GMM estimator is utilized for the study of conditional β-convergence. This estimation is based on five-yearly panel data spanning 1960-2008. The conditioning variables in the augmented Solow model include physical capital, population growth and human capital. Whereas, a Barro style income growth regression additionally includes the fertility rate, life expectancy, institutional quality, a measure of democracy, trade openness, government consumption share, inflation rate and regional dummies. Further analysis of conditional β-convergence includes the study of the sources of convergence and the estimation of steady state levels of income for the sample countries. In addition conditional σ-convergence, also known as growth convergence, is examined utilizing the recently developed methodology by Phillips and Sul (2007a).

Results confirm no absolute β-convergence for the world countries and among its geographic and income categories, except for Europe, the high income and upper middle income countries. However, an important finding is the significant absolute β-convergence for the ‘world excluding the Sub Saharan African countries’. Because of contradictory results, the conclusions on σ-convergence are dependent on the two measures of dispersion utilized in the study. The analyses suggest that the extensively cited relationship between β and σ convergence is only pertinent when the standard deviation of log income is the measure of σ-convergence; implying that σ-convergence is plausible in the absence of β-convergence. Contrary to the infrequent evidence for absolute β-convergence in the sample categories, conditional β-convergence is confirmed for the world sample and for each of its geographic and income categories for both the augmented Solow model and Barro style income growth regressions, with the exception of low income countries. Europe, Latin America and Asia have higher rates of convergence compared to the African continent, in which GDP per worker is converging at a higher speed than GDP per capita, because of higher dependency ratios. Moreover, convergence rates for the world sample are lower than those for various regions. Finally, in the growth convergence analysis based on a time-varying factor model, Europe again has shown more convergence. But, the African, Asian and Latin American regions are divided into further convergence clubs with 5, 2 and 5 clubs respectively. Similarly, there is no evidence of growth convergence for the world sample, but for 6 global clubs comprising varying numbers of countries.

Some of the original contributions of the thesis are, firstly the reconsideration of the relationship between absolute β and σ convergence. Secondly, varying rates of conditional β-convergence for GDP per worker and GDP per capita are found. In this context, the presence or absence of conditional β-convergence in the workers to population ratio has a key role; therefore, these results indicate that the demographic structure of countries and the record of population growth have played an important role in the income convergence of countries. Thirdly, a further analysis of conditional β-convergence for low income countries shows institutional quality to be relatively more important than initial human capital for income convergence. Specifically at low initial levels of development, institutional quality has a greater role in income growth and convergence.

Date of Award	30 Nov 2012
Original language	English
Awarding Institution	University of Bath
Supervisor	John Hudson (Supervisor) & Atanu Ghoshray (Supervisor)