Poverty Convergence Or Divergence? No, Convergence Clubs!

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Poverty convergence or divergence? No, Convergence clubs!

Gustavo A. Marrero1 Departamento de Economía, Centro de Estudios Desigualdad Social y Gobernanza (CEDESOG), Universidad de La Laguna, Spain. Ángel S. Marrero Departamento de Economía, Centro de Estudios Desigualdad Social y Gobernanza (CEDESOG), Universidad de La Laguna Darío Teixido Centro de Estudios Desigualdad Social y Gobernanza (CEDESOG), Universidad de La Laguna February 2017, very preliminary and incomplete version (please, do not quote)

Ravallion (2012) reopened the debate on the existence or not of poverty convergence. Despite finding evidence in favour of income convergence, this author concludes that there is no poverty convergence, an outcome questioned by the analyses of Sala-i- Martín (2006) and also by more recent studies. Using the most up-to-date data from PovcalNet, in this paper we review the discussion on poverty convergence. Firstly, we conduct a basic analysis of β-convergence, showing that the results are sensitive to taking logarithms on poverty data. However, there are evidence of geographical and non-random associations among countries that suggest the existence of groups of countries with similar convergence patterns. Thus, secondly, we employ the methodology of Phillips and Sul (2007) to analyse the existence of poverty convergence clubs, and results show evidence of six poverty convergence clubs. We also estimate convergence clubs in income and inequality, and we show that the correspondence between poverty and income clubs is much greater than that between poverty and inequality clubs, and that the former is particularly close among countries with high levels of poverty, which would suggest the existence of a poverty trap.

Keywords: Headcount poverty, Convergence clubs, Development, Inequality. JEL Classification:

1 Corresponding author. A: Facultad Ciencias Económicas y Empresariales, Camino la Hornera s/n, 38071 La Laguna, Tenerife, Spain. E: [email protected]

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

1. Introduction

According to recent information published in October of 2016 by the World Bank based on an update by Ferreira et al. (2016), the people who live in extreme poverty represents, still in 2012, 12.7% of the world population. The countries belonging to the East Asia and Pacific zone have reduced their poverty headcount rates from 61% in 1990 to 7% in 2012, in South Asia from 51% to 19% and in Latin American and the Caribbean from 18% to 6%.2 However, in Europe and Central Asia the rate has practically remained unchanged (≈2%) and in Sub-Saharan Africa the poverty rate has only decreased by 25% from 1990, meaning that 43% of its population in 2012 lives below the poverty line. These different trajectories in the development of poverty rates would not be worrying in the long term if they took part in a process of absolute convergence, that is, a process where time would lead all countries to reduce, sooner or later, their poverty until they reach zero or close to zero levels. Ravallion (2012) reopened the debate on the existence of poverty convergence. Using data from PovcalNet - for the base period of 2005 and a poverty line of 1.25 US$ - and considering a sample of 90 developing countries, he concludes that, despite the fact that there are signs of convergence in income levels, there is no evidence of absolute poverty convergence. This result contrasts with the aforementioned evidence by Sala- i-Martín (2006) who, using data from WIID, basically conclude the opposite, though warning of a certain divergence in some sub-Saharan African countries (see also Pinkovskiy and Sala-i-Martin, 2014). Recently, Cuaresma et al. (2016), using the same sample as Ravallion (2012), conclude that their lack of convergence result is owing to the author taking natural logarithms for the poverty series, thus distorting values for low poverty levels, and also including in the sample a set of Eastern European countries whose behaviour is unusual due to their process of political system change. In this paper we use the more recent data from October of 2016 published in PovcalNet and reconsider the discussion of poverty convergence. The construction of these new poverty levels is based on recent updates derived from Ferreira et al. (2016), which re- estimates a poverty line of $1.90 per person per day at 2011 PPP and updates the previous line of $1.25 at 2005 PPP. The new data base is composed of a total of 158 countries, mixing developing countries with developed countries. In this article we follow the same criteria used by Ravallion (2012) and we pick only those developing countries that have an initial poverty level different from zero and at least have two surveys with real data and not extrapolated. Accordingly, the final sample reduces to 120 countries and it includes 89 of the 90 countries used by Ravallion. The information covers a period of time from 1981 to 2013. In a first glance at the data, a lack of significant evidence of convergence is observed when poverty data are expressed in logarithms, in line with Ravallion (2012). At the same time, we find significant evidence of convergence when we use the data in levels, in line with Cuaresma et al. (2016). The appropriateness of taking natural logarithms or not for the poverty series is discussed by these latter authors and, previously, also by López and Servén (2009), who opt for the use of poverty levels, which is what we

2 Poverty data used refer to a poverty line of $1.90 base 2011. The reference year to measure the achievement of the objectives of the millennium regarding poverty is 1990.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs! do in the subsequent sections (more regarding this matter will be discussed in Section 2). However, either if we take logarithms or not and in any robustness check to be conducted with the sample, a significant dispersion of countries around the regression line is noted and, more importantly, certain groups of countries, which are clearly non- randomly grouped and belong to different geographical zones, are detected. The existence of these groupings could be showing the existence of convergence clubs, which means that poverty convergence within each club exists but not between them. If this was to be the case, depending on the weight of each type of convergence (between or within), the evidence of absolute convergence for the whole sample could be positive or negative, and results would be in general erroneous. The formation of convergence clubs derives from the theory of multiple equilibria (Abramovitz, 1986; Baumol, 1986), whose analysis has been applied mainly to individual income or to GDP per capita for a wide set of countries (Quah, 1996; Bartkowska and Riedl, 2012). In fact, there is agreement in the literature when considering that income per capita among countries shows signs of grouping into clubs more than a path of common growth. For instance, Quah (1996) found evidence of two convergence clubs of rich and poor economies in the global income distribution, Pittau and Zelli (2006) showed a very slow process of convergence across EU-12 regions and a process of distancing of a group of rich regions while Phillips and Sul (2009), using data income from 152 countries from 1970 to 2007, reported evidence of 5 income convergence clubs. However, the existence of income convergence clubs does not necessarily mean the existence of poverty convergence clubs. This is owing to, among other things, poverty levels not only being dependant on mean income, but also on other factors, such as inequality (Bourguignon, 2003; 2004). Thus, countries within the same income convergence club and with very different inequality levels could belong to different poverty clubs. Moreover, the existence of income convergence clubs and non-existence of poverty convergence clubs could also be possible if we were to analyse a sample of countries with different high income levels and all of them with very low poverty levels.3 Thus, in the second part of this article, the existence of poverty convergence clubs is analysed, which a novel contribution of the paper. To achieve this, we use the methodology proposed by Phillip and Sul (2007; 2009) based on the calculation of the cross-sectional variance ratio of poverty ratios over time, which has been recently applied to numerous cases of macroeconomic variables.4 This method has a clear advantage over other alternatives such as unitary root tests or cointegration, as it allows to detect convergence patterns even during transitory periods of divergence. Likewise, it allows to detect different behaviours among countries such as convergence, divergence, or convergence clubs. In order to conduct the convergence club analysis we use the most recent interpolated poverty series (every three years from 1981 to 2013) provided by PovcalNet and which are constructed from the surveys

3 This possibility is reduced in our case as our sample is composed of developing countries with, in most cases, clearly positive poverty levels. 4 See e.g. Camarero et al., (2013) for CO2 emissions, Apergis et al., (2016) for electricity prices, or Fritsche and Kuzin, (2011) for price levels.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs! conducted.5 Thus, the first significant outcome is the null evidence of absolute convergence; the second outcome is the existence of evidence in the formation of 6 poverty clubs. Once poverty convergence clubs are estimated, we apply the same methodology to the mean income series and to the Gini inequality index, and study the similarities and correlations in the formation of poverty, income and inequality convergence clubs. The main conclusion is that correlation between poverty and income clubs is much more evident than that between poverty and inequality, and the former is specially close among countries with high levels of poverty, which would suggest the existence of poverty trap. The rest of the paper is structured as follow. In Section 2, we present the poverty data utilized and conduct a basic beta-convergence analysis. In Section 3 we present the convergence clubs methodology, starting from a simple neoclassical model that relates income to poverty and to the existence of clubs. In Section 4 we show the results of poverty convergence clubs, and Section 5 analyses its correlation with income and inequality convergence clubs. Finally, Section 6 presents the main conclusions.

2. Data description

The data utilized in this work have been obtained from PovcalNet, a World Bank tool used for the measurement of absolute poverty in the whole world. Absolute poverty is measured through headcount ratio, an index that represents the proportion of population in an specific country with a consumption per capita (or income when consumption is not available) below the poverty line. Poverty line is set at $1.90 per person per day at 2011 PPP, which updates the previous threshold utilized of $1.25 per person per day at 2005 PPP (Ferreira et al., 2016). The latest information published in October of 2016 covers a total of 138 developing countries and 21 countries with high incomes for different years between 1981 and 2013. Poverty levels estimates are made from over two million surveys conducted in homes randomly selected, which are homogenized using the PPP exchange rates for household consumption from the 2011 International Comparison Program. The number of surveys conducted is not the same for all of the countries, nor they are conducted periodically. This means that for the period 1981-2013 each country will have a different number of surveys conducted in different years. In order to be able to make comparisons, PovcalNet offers estimates of the poverty levels of each country aligned to a reference year (2011 for the new data base). In this section, we will only use data obtained from these surveys. Besides, Povcalnet allows working with time series, which were built through interpolations in those cases where there were no data, but there were data before and/or after. Obviously, the greater the number of surveys per country, the more precise the interpolation process will be. These time series will be used in the second part of the article, in Section 4.

5 PovcalNet (online analysis tool), World Bank, Washington, DC. Retrieved Nov 24, 2016 from: http://iresearch.worldbank.org/PovcalNet/

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

From the 159 PovcalNet countries and applying the same selection criteria as in Ravallion (2012), we choose those countries that have at least two surveys and also have a poverty level different from zero in the earliest survey. Besides, we do not consider the most developed countries in any case as they have poverty levels very close to zero during the whole period of time analysed. By applying these criteria, our sample reduces to 120 countries with their respective percentage of population under the poverty threshold of $1.90 per day in 2011PPP, a sample that is clearly higher than that of 90 countries used by Ravallion (2012) and Cuaresma et al. (2016).

2.1. Preliminary exploration of data: is there evidence of absolute convergence?

For the 120 countries selected, next we analyse the existence of a negative relation between poverty growth during the period considered and their initial levels. As in Ravallion (2012), we select for each country the longer periods of time available between surveys (the selected sample with the countries and the years of the surveys can be found in Appendix 1). Note that in the selected sample, for poverty levels of the first and last periods, years do not coincide among countries, so in order to calculate the growth rate or the variations in percentage points, annualised data are used. In this case, given a country i in a year t with a poverty level Hit and a time difference between the earliest and last survey of τ, the variation rates will be ∆퐻푖 = (퐻푖푡 − 퐻푖푡−τ)/τ if we work with levels series, or ∆ln퐻푖 = (푙푛퐻푖푡 − 푙푛퐻푖푡−τ)/τ if we do it with logarithm variables. By making a first division by geographical zones, in Table 1 we show the main descriptive statistics associated to poverty in levels and its average annualized growth rate in percentage points.

Table 1. Poverty descriptive statistics (levels)a Earliest survey Latest survey Zone ∆푯 푯 푯 푯 푯 푯 푯 푯 푯 풊 (no. countries) Year 풊 풊 풊 풊 Year 풊 풊 풊 풊 mean mean mean Std Min Max mean mean Std Min Max EAP (18) 1990 40.15 29.61 2.83 95.59 2012 9.71 13.37 0.04 46.76 -1.08 ECA (27) 1995 9.86 16.14 0.02 54.37 2012 5.63 15.05 0.00 66.79 -0.14 LAC (22) 1988 20.30 16.88 0.46 55.59 2011 7.66 11.24 0.30 53.91 -0.56 MENA (5) 1992 10.66 7.56 0.28 20.63 2010 5.56 9.57 0.08 22.52 -0.19 SA (9) 1990 41.01 20.09 9.99 62.16 2011 12.26 8.35 1.92 24.83 -1.52 SSA (39) 1994 56.10 25.45 0.42 94.05 2010 42.05 21.19 0.53 77.84 -0.95

a Own elaboration from PovcalNet. Sample size of 120 countries (Appendix 1). Following the division used by the World Bank, the geographical zones are East Asia and Pacific (EAP), Europe and Central Asia (ECA), Latin America and the Caribbean (LAC), Middle East and North Africa (MENA), South Asia (SA) and Sub-Saharan Africa (SSA). In bracket the number of countries of our sample pertaining to each zone. Year (mean) is the average year by zone in which the earliest and latest survey

were conducted, 퐻푖 (mean) is the average headcount ratio by zone and ∆퐻푖 (mean) is the average annualized growth rate by zone

Table 1 highlights that zone SSA is the one with the greatest poverty averages, both in the earliest and latest surveys, although the high values of its standard deviations show great differences within countries, besides being the third zone with a greater poverty reduction out of the 6 considered. On the opposite side, MENA zone stands out for having the lowest average poverty rate at the end of the sample, also presenting

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs! a low standard deviation within its countries. For its part, even having the second highest average poverty rate at the end of the sample, SA zone is the one with a highest reduction in its poverty levels, very much conditioned by countries like China. Besides, it has also the highest reduction of standard deviation, thus showing a greater convergence process among the countries that comprise it. Figure 1 shows the relation between poverty annualised growth and its initial level, taking natural logarithms (analogous to Figure 1 in Ravallion, 2012), while Figure 2 shows that same relation but in levels (β-convergence graphics).6 On the one hand, for data in logarithms (Figure 1) we find the same result as Ravallion (2012), that is, lack of evidence in favour of absolute convergence. On the other hand, for data in levels (Figure 2) the opposite evidence is found, being the slope negative and significantly different from zero at 1% (as in Cuaresma et al., 2016), thus suggesting that countries with initial higher levels of poverty tend to reduce (increase) their poverty rates more (less) than countries with lower initial levels.7 Making a thorough study of both figures, regardless of the existence or not of evidence of beta-convergence, certain groupings of countries related with their geographical zones are identified. For instance, zones ECA ( ) and LAC ( ) are mainly found in the south-western part of both graphics, with low initial poverty levels and reduction in their poverty rates, while the SSA ( ) zone is mainly found in the eastern side of both figures, with high initial poverty levels but with a strong variability among countries, as it was shown in Table 1. These groupings, apparently non-random, could be a symptom of the existence of poverty convergence clubs, that is, groups of countries converging towards equilibrium points at common poverty levels within the group, but different between each other. Contrasting the existence of poverty convergence clubs and understanding the factors correlated with the formation of clubs can shed light on the existence of nonlinearities in the shape of poverty traps (Azariadis and Stachurski, 2005), which would hinder the reduction of poverty and subsequent development of certain countries and regions.

6 Using poverty data in levels seems more appropriate to estimate a relationship between poverty and average income. On the one hand, taking logarithms of data very close to zero can distort the sample. On the other hand, the relation between H (in levels) and average income (푦̅) per capita (in logs) is more linear than the resultant when both variables are in logs. Because of that, it is more appropriate to estimate models of the kind 퐻 = 훼 + 훽 ln(푦̅) + 휀 than those of the kind ln⁡(퐻) = 훼 + 훽 ln(푦̅) + 휀. Appendix 2 shows the scatter plot between H and ln(푦̅), and between ln(H) and ln(푦̅), thus being evident that the relation is clearly more linear in the former case than in the latter. 7 Cuaresma et al. (2016) also finds that, even using logarithms, evidence of lack of poverty convergence disappears when Eastern Europe countries with anomalous behaviour are excluded from the sample. Besides, these results stand when using new data in PPP2011 and when restricting the sample of countries to the 90 considered by Ravallion (2012).

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Figure 1. Poverty rate changes vs. initial values (logs)

y = -0.0057x - 0.0477 R² = 0.0074 0.3 t-test (β) = -0.94

0.1

-0.1

-0.3

Annualized growth in the poverty rate poverty the in growth Annualized -0.5

-0.7 -4 -3 -2 -1 0 1 2 3 4 Log initial poverty rate

regresion EAP ECA LAC MENA SA SSA Lineal (regresion)

Figure 2. Poverty rate changes vs. initial values (levels)

5 y = -0.0224x + 0.0287 4 R² = 0.3024 t-test (β) = -7.15 3

-1

Annualized growth in the poverty rate poverty the in growth Annualized -2

-3

-4

-5 0 10 20 30 40 50 60 70 80 90 100

Initial poverty rate

regresion EAP ECA LAC MENA SA SSA Lineal (regresion)

3. A model to analyse poverty convergence clubs

This section attempts to derive a convergence equation for the headcount poverty rate allowing the possibility of accounting for convergence clubs. To this purpose, we merge two existing results in the literature. The first one establishes the relationship between poverty, mean income and inequality (Bourguignon, 2003, 2004 or Ferreira,

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

2010), while the second, from an extension of the neoclassical growth model, derives an equation for income growth that allows for cross-section transition heterogeneity among economies and thus the possibility to analyse income convergence clubs (Phillips and Sul, 2007 and 2009).

3.1. Poverty, mean income and inequality

The first equation simply considers the incidence of poverty – the headcount poverty rate 퐻 – as a function of both the mean income and the income distribution shape. We start from a widely used accounting identity in the inequality literature,

푦(푝) 퐿 (푝, 휋) = , (1) 푝 푦̅ where 퐿(푝, 휋) is the Lorenz curve for a given distribution evaluated at a percentile 푝; 휋 is the vector of parameters that determines the functional form of the Lorenz curve; 퐿푝 is the derivative of the Lorenz curve at percentile 푝; 푦(푝) is the income at that percentile; and ⁡푦̅ is the overall distribution income mean. The headcount 퐻 is simply the share of population with incomes no higher than the poverty line, 푧 (i.e., 1.90 2011PPP-US$ in our case), as for instance, 퐻 = 퐹(푧), where 퐹(· ) is the cumulative distribution of income. Therefore, evaluating equation (1) at the headcount level, 푝 = 퐻, and solving for 퐻 yields:

−1 퐻 = 퐿푝 (푧/푦̅, 휋), (2) which exactly relates the headcount poverty rate at poverty line 푧 to the income mean and to the shape of the Lorenz curve, which is strongly associated with overall inequality. Moreover, assuming that income follows a log-normal distribution, which is widely accepted in the literature, the headcount level 퐻 can be expressed as an exact function of the log of mean income and the Gini coefficient,⁡퐺,8

ln(푧)−ln(푦̅) 휎 퐻 = 훷[ + ], (3) 휎 2 where 휎 is the standard deviation of the log of income, which can be expressed as 휎 = √2훷−1(1 + 퐺⁄2), being 훷(.) the standard normal cumulative distribution.

3.2. Mean income and cross-country heterogeneity

The second piece of evidence follows the works of Phillips and Sul (2007, 2009). These authors argue that, in order to analyse convergence in income from an econometric point of view, the construction of a model designed to generate heterogeneous transitory growth patterns among countries is required. Based on an extended

8 Lopez and Serven (2004) compare the theoretical quintile shares according to a log-normal distribution with their empirical counterparts using data from 794 household surveys and conclude that the log normal approximation fits the empirical data extremely well.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs! neoclassical growth model, they derive an equation that accommodates heterogeneity growth patterns across economies i at a time t as follows:

−훽푖푡푡 ln푦̅푖푡 = 푙표𝑔푦̃푖 + 푙표𝑔퐴푖0 + (푙표𝑔푦̃푖0 − 푙표𝑔푦̃푖)푒 + 푋푖푡푡, (4) where 푙표𝑔푦̃푖 is the steady-state level of the income; 푙표𝑔퐴푖0 and⁡푙표𝑔푦̃푖0 denote the technology and income initial levels; 훽푖푡 is the speed of convergence; and 푋푖푡 is the technological progress. The parameters 훽푖푡 and 푋푖푡 allow for cross-section heterogeneity, and hence transitional growth inequality. Equation (4) can provide an explanation on how poor economies with low initial technology levels learn faster (푋푖푡), thus accelerating the speed of convergence (훽푖푡) and leading to a process of catching up with more developed economies. However, certain poor economies that respond slowly to the diffusion of technology might have a divergent behaviour and are likely to fall into a poverty trap. Additionally, this model presumes the existence of a common growth component common to all countries (e.g., information and communications technologies, or a common technological progress factor). Accordingly, from (4), the dynamics of the income can be written as follows:

ln푦̅ = 푙표𝑔푦̃ + 푙표𝑔퐴 + (푙표𝑔푦̃ − 푙표𝑔푦̃)푒−훽푖푡푡 + 푋 푡, 푖푡 ⏟ 푖 푖0 푖 0 푖 푖푡 (5) 푎푖푡 where the technological progress parameter (푋푖푡푡) is assumed to have a growth component (µ푡) that is common among economies. Then, we can write equation (5) in the following form:

푎푖푡+푋푖푡푡 푙푛푦̅푖푡 = ( )µ푡 = 푏푖푡µ푡, (6) µ푡 where 푏푖푡 measures the extent to which an economy shares the common growth component µ푡. Particularly, 푏푖푡 can be used to measure the transition path of an economy towards the common growth path represented by µ푡. During the transition period, the parameter 푏푖푡 may depend on the 푎푖푡 component, that is, on the speed of convergence, technological progress, steady-state level of income and technology and income initial levels.

3.3. A model for poverty convergence clubs

This part of the section merges the two previous models, more concretely, equations (3) and (4). From (3), we assume the following reduced-form specification, which relates the headcount poverty rate to the log of the income mean and considers the Gini coefficient as an overall measure of inequality,9

퐻푖푡 = 훼푖 + 훾푖푡푙푛푦̅푖푡 + 휑푖푡퐺푖푡 + 푣푖푡, (7)

9 A similar linear (or log-linear) specification has been considered by Ravallion (2012) and Cuaresma et al. (2016) to analyse econometrically the relationship between poverty and mean income.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

where 훼푖 is a country-specific fixed effect and 푣푖푡 is an error term. Combining (6) with (7) and considering that poverty dynamics also has a common trend µ푡, we can rewrite equation (7) as:

훼푖µ푡 (휑푖푡퐺푖푡+푣푖푡)µ푡 훼푖+훾푖푡푏푖푡+푤푖푡 퐻푖푡 = + 훾푖푡(푏푖푡µ푡) + = µ푡 = 훿푖푡µ푡, (8) µ푡 µ푡 µ푡 where 푤푖푡 = (휑푖푡퐺푖푡 + 푣푖푡), allowing us to measure poverty as the fraction 훿푖푡 that a particular country i at a time t is sharing with the common trend µ푡. This poverty equation includes the aforementioned growth component common to all countries, but also idiosyncratic factors incorporated included in 훿푖푡 such as the geographical location, institutional quality, financial development, etc.

3.4. The implementation of the quantitative approach

In order to implement the quantitative strategy to analyse and test the convergence club hypothesis, a particular form needs to be assumed for the modelling of 훿푖푡. Following Phillips and Sul (2007), this can be done by constructing the following transition coefficient:

퐻푖푡 훿푖푡 ℎ푖푡 = 1 = 1 , ∑ 퐻 ∑ 훿 (9) 푁 푖푡 푁 푖푡 which eliminates the common trend µ푡 by scaling the component 훿푖푡 from an economy i in relation to the cross-section average. The transition parameter measures both the country behaviour relative to the average and the country deviations from the common path. When all countries converge to the same steady state equilibrium, we have ℎ푖푡 = ℎ푡 and ℎ푖푡 → 1 as 푡 → ∞. This would imply, in our case, overall convergence in poverty headcount ratios. Then, the cross-sectional variance of ℎ푖푡 converges to zero and can be expressed as:

1 휎2 = ∑푁 (ℎ − 1)2 → 0⁡푐푢푎푛푑표⁡⁡푡 → ∞ , (10) 푡 푁 푖:1 푖푡

In order to test the existence of convergence, Phillips and Sul (2007) specify a semiparametric model assuming the following general form for 훿푖푡:

휎 훿 = 훿 + 휎 휖 ; 휎 = 푖 ⁡⁡; ⁡⁡푝푎푟푎⁡푡 ≥ 1⁡⁡⁡⁡⁡⁡⁡휎 > 0⁡ , 푖푡 푖 푖푡 푖푡 푖푡 퐿(푡)푡훼 푖 (11) where 휖푖푡 is i.i.d (0,1); 퐿(푡) is a slowly varying function (converges to ∞ as 푡 → ∞); and 훼 is the speed of convergence. The null and alternative hypotheses of convergence can be written as:

퐻0:⁡⁡⁡훿푖 = 훿 ⁡⁡⁡푎푛푑⁡⁡⁡훼 ≥ 0 , (12) 퐻퐴:⁡⁡⁡훿푖 ≠ 훿 ⁡⁡⁡표푟⁡⁡⁡훼 < 0 .

Note that this formulation allows for some interesting possibilities. It permits to test the existence of overall convergence or divergence across economies, and additionally, it also allows for the possibility of any subgroup of economies converging to their own

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

steady state equilibrium. In the latter case, the alternative hypothesis is that 훿푖 ≠ 훿 and 훼 ≥ 0. Further, this formulation can accommodate periods of transitional divergence, presenting a clear advantage over other conventional methods such as cointegration and unit root tests, which fail to detect convergence during these periods (see Phillips and Sul, 2007, p. 1778). In order to test the null hypothesis of convergence against the alternative hypothesis, a testing procedure is proposed involving a one-side t-test based on the following simple time series regression:

2 휎1 ̂ log ( 2) − 2푙표𝑔퐿(푡) = 푎̂ + 푏 log 푡 + 푢̂푡⁡⁡⁡푝푎푟푎⁡푡 = [푟푇], [푟푇] + 1, … , 푇⁡푐표푛⁡푟 > 0 , (13) 휎푡

2 2 where the 휎1 ⁄휎푡 ratio is the cross-sectional variance (equation 10) at the initial period in relation to the cross-sectional variance of each time period and 푟 is some fraction used to disregard the first r% of the time series. Based on Monte Carlo simulations, the authors suggest using the function 퐿(푡) = log⁡(푡 + 1) and 푟 = 1/3 for 푇 < 50. Considering that 푏̂ = 2훼̂, to test the inequality of the null hypothesis 훼 ≥ 0 the one side t-test is constructed using HAC standard errors. In this way, if 푡푏 < −1.65 (at 5% significance level), the null hypothesis of convergence is rejected on the understating that a rejection of the overall sample does not imply absence of convergence among subgroups of countries. More precisely, in order to check for the existence of convergence clubs, the authors propose a clustering algorithm based on repeating iteratively the log 푡 regression. The basic procedure is to first arrange the panel in descending order according to the last period of observation, then run the log 푡 regression and check for overall convergence. If the hypothesis of overall convergence is rejected, the first two highest units of the panel are selected and countries are added one by one, running the log 푡 until a 푡푏 larger than -1.65 is found. The group of countries that maximize 푡푏 comprises the so- called core group. If 푡푏 > −1.65 does not hold for the first two countries, the process is started again with the next two ones. Add each of the remaining countries at a time to the core group and run the log 푡 regression. All units with 푡푏 > −1.65 are included in the core group, thus constituting the first convergence club. Then the process is repeated for all the countries apart from the convergence club, in order to classify them either as convergence clubs or divergent units.10

4. Estimation of poverty convergence clubs

In this section, we apply the methodology of Phillips and Sul (2007; 2009) to contrast the existence of poverty convergence clubs. The convergence clubs methodology requires a balanced data panel. For the 120 selected countries, we use the interpolated data provided by PovcalNet, thus being able to have data on the poverty rate of each

10 See Phillips and Sul (2007) for further details on this process.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs! country for the period 1981-2013 with three years intervals11. Table 2 shows the main statistics associated to poverty for each year according to geographical zones and for the whole sample, while Figure 3 shows these data graphically.

Table 2. Annual poverty rate mean by zoneb Zone 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2013(*) 49.69 45.4 41.77 38.9 35.5 29.94 27.42 21.76 17.4 15.54 10.65 7.93 EAP (33.34) (30.78) (29.28) (28.57) (26.61) (22.35) (19.32) (17.18) (15.08) (14.33) (10.59) (8.86) 3.05 2.62 2.55 3.74 9.26 10.59 11.28 9.04 6.12 3.41 3.33 2.84 ECA (10.45) (9.04) (8.83) (7.86) (15.12) (16.95) (16.47) (15.73) (12.33) (8.88) (7.44) (6.19) 23.07 24.57 22.97 18.87 19.47 15.15 14.43 13.94 12.43 9.3 8.24 7.39 LAC (23.29) (22.91) (22.86) (15.27) (21.57) (11.43) (11.44) (11.7) (11.99) (11.24) (11.35) (11.13) 9.64 7.64 7.88 6.47 7.15 7.87 7.57 6.85 6.1 5.01 4.48 5.05 MENA (5.81) (4.72) (3.78) (2.74) (5.13) (6.94) (7.88) (7.93) (8.07) (7.75) (7.92) (9.78) 55.94 53.23 50.26 46.64 43.15 36.7 33.75 30.96 23.97 18.24 12.31 9.2 SA (23.79) (23.9) (23.39) (22.63) (20.5) (19.88) (16.44) (13.7) (12.92) (12.61) (8.04) (6.4) 55.7 56.67 56.53 56.59 58.28 54.97 52.35 50.95 47.29 44.5 42.05 39.71 SSA (23.61) (23.84) (24.01) (23.04) (22.03) (23.05) (22.89) (22.54) (21.86) (21.9) (21.51) (22.16) 35.07 34.63 33.52 32.3 33.45 30.6 29.15 27.02 23.69 20.84 18.64 16.99 Global (31.41) (30.84) (30.39) (28.94) (28.73) (26.63) (25.18) (24.75) (23.5) (22.96) (21.88) (21.42) (*) See note in Table 1. b Own Elaboration from PovcalNet. Interpolated data of 120 countries from 1981-2013. The table shows the cross-section mean of the headcount ratio by each geographical zone. Standard deviation in parenthesis

Figure 3. Time trend of the poverty rate for alternative geographical areas

30 Poverty ratio Poverty

0 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2014

Global EAP ECA LAC MENA SA SSA

11 Reference years are 1981, 1984, 1987, 1990, 1993, 1996, 1999, 2002, 2005, 2008, 2010, 2011, 2012 and 2013, although we exclude 2010 and 2012 for consistency. For the last period we use 2013 as approximation to 2014. Whether to include or not the year 2013 does not modify the results attained.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

As it can be observed both in the table and the figure, poverty rates have been reduced in all geographical zones considered in the sample. Nevertheless, behaviours over time are different. For instance, zones MENA and ECA reduce their poverty rates to a lesser degree than the rest of the zones, even in periods of growth (1990-1999) in the second case. For their part, zones EAP and SA are the ones that experience a more pronounced reduction of poverty over these almost 30 years. Given the high standard deviations found in most cases within each zone and year, although the geographical region to which each country belongs would be relevant, this does not need to be by no means the most determining factor for the formation of clubs. For this reason, it is relevant to have a methodology, like the one used in this work, which establishes convergence clubs using the whole sample and without including priors that could distort the results of the analysis.

4.1. Convergence club estimation

When applying the regression log-t (Equation 13) to the headcount ratio of the 120 countries analysed for the period 1981-2013, the absolute convergence hypothesis is clearly rejected at the 1% of significance12. Namely, the estimated equation is 푙표𝑔 ℎ1⁄ℎ푡 − 2 log log 푡 = 2.20 − 1.86 log 푡, with a t-statistic for the slope of -13.5, clearly below the reference level of -1.65. Therefore, we can conclude that, conditional on the selected sample, the 120 countries are not converging towards the same long-run equilibrium of poverty. Notice that this conclusion contradicts that reached using a simple beta-convergence graphic (Figure 2). As commented above, under the existence of convergence clubs, a simple beta-convergence analysis leads easily to misleading conclusions. Next, we study the possibility of the existence of poverty convergence clubs using the cluster algorithm explained in the previous section. The successive regressions log-t considered show the existence of 6 poverty convergence clubs, with evidence of 3 countries with divergent behaviors (Jamaica, Dominican Republic and Thailand)13 The list of member countries of each club can be seen in Appendix 3. The results of the log-t regressions for each club are presented in Table 3, where the coefficients 훽 considered, the associated t statistics and the mean poverty rate for each club in 2013 (last year of sample) are shown. Clubs are arranged in descending order regarding the mean poverty rate they are converging to. It is worth mentioning that clubs with the most poverty (clubs 1, 2 and 3) tend to group countries from the sub-Saharan Africa, while clubs with medium and medium-low poverty levels (clubs 4 and 5) mix more countries from different geographical zones. Finally, in club 6, with the lowest poverty levels, there is again a high concentration of countries from the European and Central Asian zones, although it not as high as in the case of the three first clubs. These geographical groupings regarding the poverty clubs considered can be visually examined in Figure 4.

12 The original MATLAB code was kindly provided us by Dra. Monika Bartkowska. 13 The divergent behavior is caused because these countries have a high poverty ratio at the initial period and a sharp fall afterwards.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Table 3. Identified clubs Number of countries by zone 푯 푯 푯 푯 휷 풕 풊 풊 풊 풊 Club EAP ECA LAC MENA SA SSA Total (푿̅ퟐퟎퟏퟑ) (Sd.) (Min) (Max) Club1 0 1 0 0 0 12 13 0.05 1.52 59.67 18.68 23.170 80.71 Club2 0 1 1 1 0 7 10 0.27 2.71 43.98 11.94 22.520 56.13 Club3 2 1 2 0 1 13 19 0.12 1.43 26.13 10.06 7.700 41.06 Club4 6 2 5 0 4 3 20 0.19 1.50 10.63 5.42 0.080 18.93 Club5 6 8 9 0 4 2 29 -0.85 -1.59 3.35 2.46 0.001 8.93 Club6 3 14 3 4 0 2 26 -6.15 -1.61 0.50 0.45 0.001 1.35

Figure 4. Poverty convergence clubs. World map

Additionally, Figure 5 shows, on the one hand, the time evolution of each club's poverty rates, and on the other hand, the relative transition paths with the cross sectional means associated to each club. The transition curves show a stable behaviour of club 6 (lowest poverty rate) over time, and an apparent tendency of clubs 4 and 5 to converge towards it, being this tendency more noticeable in the last years of the period considered. Likewise, it is observed a convergent behaviour of clubs 1 and 2, which have the highest poverty rates both at the beginning and the end of the period, and a relatively divergent behaviour of club 3. In summary, evidence suggests that the poverty rate within the sample of selected countries does not converge over time. Nevertheless, a certain evidence of convergence among the 6 subgroups of countries, 3 of them with high poverty levels, and the other 3 with relatively low levels, is visually observed. However, this visual evidence is not yet corroborated with the statistical significance analysis when applying the regressions log-t among these sets of clubs.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Further, it is worth remarking that these results are quite solid regardless of whether we use our sample of 120 countries or Ravallion's restricted sample of 90 countries. 14

Figure 5. Poverty rates and relative transition paths

70 4

3 50

40 2

Poverty rates by club (mean) clubby rates Poverty 30 Relative transition paths (mean) paths transition Relative

20 1

0 0 1981 1986 1991 1996 2001 2006 2011 1981 1986 1991 1996 2001 2006 2011

5. Do income and inequality determine poverty convergence clubs?

The poverty headcount index is well known to be closely linked with the existing mean income and inequality in a country or region (Bourguignon, 2003; 2004). In fact, if income distribution is log-normal, this would be an exact relation (Cowell, 2000). However, the existence of nonlinearities between poverty and these variables, for example, the nonlinearity between income and inequality (Barro, 2000), poverty traps (Azariadis and Stachurski, 2005) or the changing relation between poverty and income according to the different levels of poverty (Ravallion, 2012) can cause the correspondence among poverty, income and inequality convergence clubs not to be exact. In this section we study the existing relation between the formation of poverty, mean income and inequality clubs. To achieve this, we first apply the methodology described in section 3.1 to the mean income (taking logarithms) and the Gini time series obtained from PovcalNet in the latest update from October of 2016.

5.1. Income and inequality convergence club estimation

For the 120 countries in our sample and during the same period (1981-2013, every 3 years), we use the interpolated data from PovcalNet on income and Gini. Income is the average monthly income expenditure per capita in 2011 PPP. For the case of the Gini,

14 Results are available upon request

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

it is worth mentioning that in some cases, interpolated data for Gini index has the problem of only showing the value of the years when surveys were conducted, or else no data is shown. In the first case, in order to obtain a value for a year when no surveys were conducted, we have decided to weight according to the previous and next surveys.15 In the second case, we have recovered the Gini from the exact condition which relates it to the headcount poverty rate and the mean income under log- normality of the income distribution (Equation 3). Table 4 shows, for the global sample, the main statistics for income and Gini throughout the selected period. The table highlights that, in average, income has increased over time (about 1.3% per year), while inequality has remained more or less stable around a level of 41%. For this selection of developing countries, it is also observed that then coefficient of variation (the standard deviation divided by the mean) shows a downward trend, thus now the differences in mean income among countries (on average) are now relatively smaller than 30 years ago. Moreover, it is also observed that a higher number of countries are now concentrated around a Gini of about 41% than 30 years ago (the standard deviation is clearly smaller during the twenties than during the eighties), and also the difference between the maximum levels of inequality (above 75% before 1995 and below 65% after 2005) and the minimum (about 20% before 1995 and around 25% after 2005) has been reduced along these 30 years.

Table 4. Income and Gini descriptive statisticsc Series Stats 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2013 Mean 197.77 189.52 191.46 183.82 172.85 179.14 185.37 196.01 219.32 252.40 266.88 283.27 Sd 186.08 163.68 163.02 143.77 130.79 131.39 136.13 137.81 157.91 186.93 193.89 205.15 Income Min 5.83 8.71 13.01 19.43 13.75 29.84 24.62 21.20 23.28 31.32 38.86 46.40 Max 1084.88 764.28 714.78 731.76 686.39 728.96 730.15 745.91 933.09 1046.71 1045.44 959.90 Mean 41.27 40.87 40.69 41.23 42.15 42.37 41.94 41.55 41.35 41.12 40.75 40.49 Sd 12.29 11.69 11.67 11.12 10.14 9.59 8.93 9.05 8.70 8.37 8.55 8.49 Gini Min 18.46 17.79 20.20 22.30 21.60 27.01 24.76 17.36 16.64 23.72 24.55 24.55 Max 80.71 75.28 75.38 75.66 75.54 75.55 63.42 64.73 63.62 63.20 63.38 63.38 c Own Elaboration from PovcalNet. Interpolated data of 120 countries from 1981-2013. The table shows the cross-section mean of the income and Gini measures and its associated statistics.

By applying regressions log-t to the mean income and Gini time series, we reject the absolute convergence hypothesis in both cases with a significance level of 5%. Specifically, the estimated 훽 coefficients are -0.88 and -0.61 and associated t-statistics are -28.1 and -34.3 for mean income and Gini respectively. Next, by applying the same algorithm as for poverty, we find 3 convergence clubs both for income and for Gini, with evidence of 2 divergent countries only in the first case (Dem. Rep. Congo and Madagascar).16 The list of member countries of each club can be seen in Appendix 4 for income and Appendix 5 for Gini. Table 5 reports the estimated 훽 coefficients, associated t statistics, income mean rate and Gini for each club in the year 2013.

15 For instance, if the Gini value is only shown for 1990 and 1996, the value for 1993 will be half of the value for 1990 plus half of the value for 1996. 16 Congo has the lesser income value from 1996 to 2011 while Madagascar has the lesser income value in 2013 despite pertaining to the second quartile of the series at the initial period.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Likewise, Figure 6 and 7 show the geographical representation of the clubs while Figure 8 shows the relative transition paths of identified clubs for each of the series.

Table 5. Income and Gini identified clubs Number of countries by zone 휷 풕 푿̅ퟐퟎퟏퟑ 푺풅. 푴풊풏. 푴풂풙. Series Club EAP ECA LAC MENA SA SSA Total Club1 12 15 17 4 4 8 60 0.04 1.17 412.86 206.12 86.92 959.90 Income Club2 4 4 4 0 2 6 20 -0.06 -1.14 224.08 101.82 70.02 364.64 Club3 2 8 1 1 3 23 38 -0.03 -0.78 122.26 59.31 48.29 291.82 Club1 5 8 16 2 3 17 51 0.08 1.16 46.12 7.35 32.13 63.38 Gini Club2 6 4 5 0 1 10 26 0.33 3.32 42.02 6.05 30.55 51.45 Club3 7 15 1 3 5 12 43 0.10 2.53 32.88 4.32 24.55 42.65

Notice that the income clubs are arranged in descending order, whereas the Gini clubs are arranged from high to low inequality. The club 1 of high income is mostly comprised of countries from East Asia and Pacific, Europe and Central Asia and Latin America and the Caribbean, while the club 3 of low income is mainly composed of countries from Sub-Saharan Africa. As shown in Figure 8, clubs 1 and 3 of income have opposite paths, increasing the gap between high and low income countries. In relation to the Gini clubs, Table 5 shows a concentration of countries with high levels of inequality (club 1) in the LAC zone and, on the contrary, a concentration of low inequality countries (club 3) in the ECA zone. It is interesting the case of the SSA zone, with countries covering a broad spectrum of inequality levels despite their low income status. Further, Figure 8 also shows a process of distancing between high and low inequality countries.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Figure 6. Income. World map

Figure 7. Inequality. World map

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Figure 8. Income and Gini relative transition paths

1.2 1.2

1.1 1

0.8

0.9

Gini relative transition paths transitionrelative Gini Income relative transition paths transition relative Income

0.8 0.6 1981 1986 1991 1996 2001 2006 2011 1970 1980 1990 2000 2010 2020

Club1 Club2 Club3 Club1 Club2 Club3

5.2. How do poverty, income and inequality convergence clubs correlate?

Once poverty, mean income and inequality convergence clubs have been estimated, in this section we compare them with each other. Table 6 shows conditional frequencies of a country belonging to a certain income club (1, 2 or 3) conditional of also belonging to a particular poverty club (1 up to 6). Results show that there is a greater probability of being located on the matrix secondary diagonal, which suggests the evidence of a significantly negative correlation between poverty and income clubs. For instance, 92.3% of countries belonging to the highest poverty rate club (Club 1) belong, at the same time, to the lowest income club (Club 3), while 92.5% of countries with the lowest poverty rate (Club 6) belong at the same time to the highest income club (Club 1). Moreover, belonging to the highest poverty clubs (Clubs 1 and 2) practically guarantees belonging to the lowest income club (Club 3), 20 out of 33 countries. Nevertheless, as poverty levels decrease, distribution related to income clubs evens out. For example, of the 51 countries from poverty clubs 4 and 5, 22 of them (43%) belong to medium and low income clubs. Therefore, this analysis of conditioned frequencies demonstrates the existence of nonlinearity which are consistent with the existence of a poverty trap. Table 6 also shows evidence of a non-perfect correlation between poverty and income clubs even for the extremes (highest and lowest poverty clubs). For example, the 3 countries that belong to the high poverty clubs (1 and 2) and the medium income club (2) are Cote d'Ivoire, Mali and Burkina Faso, all of which belong to the SSA zone and with poverty and income rates for 2013 of 25.3, 51.7 and 45.1, and 112.3, 73.2 and 79.4 respectively. Likewise, the 3 countries that belong to low poverty clubs (5 and 6) and the low income club (3) are Albania, Romania and Armenia, all of them countries from

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Eastern Europe with poverty and income rates in 2013 of 1, 0.001 and 2.4, and 229.8, 266.5 and 182.3 respectively. In these apparently anomalous behaviour countries where other factor such as inequality could be influencing.

Table 6. Poverty Clubs vs. Income Clubs

+ Income - Club1 Club2 Club3 Total N % N % N % N % + Club1 0 0.00% 1 7.69% 12 92.31% 13 100%

Club2 0 0.00% 2 20.00% 8 80.00% 10 100% Club3 6 31.58% 2 10.53% 11 57.89% 19 100%

Club4 10 50.00% 4 20.00% 6 30.00% 20 100% Poverty Club5 19 61.29% 10 32.26% 2 6.45% 3117 100% - Club6 25 92.59% 1 3.70% 1 3.70% 27 100%

Table 7 shows the conditional frequency of a country belonging simultaneously to a poverty club and to an inequality club. The table shows that there is a slightly higher probability of being located on the matrix main diagonal, thus bringing to light a positive correlation between poverty and inequality. Nevertheless, in this case there is a much greater uniformity in the distribution of the frequencies. Thus, of the 23 countries in the highest poverty clubs 1 and 2, there are 11 (47%) which belong to low or medium inequality clubs, and the others 53% to the high inequality club. Conversely, of the 58 countries comprising the low poverty clubs 5 and 6, there are 33 (56%) that belong to medium or high inequality clubs, and the others 44% to the low inequality club. Thus, although the correlation is positive (as the theory would suggest), the evidence found suggests that belonging to a low (high) poverty club does not guarantee belonging to a low (high) inequality club. As an example, two of the aforementioned countries, Mali and Burkina Faso, belong to the second highest poverty, medium income and low inequality clubs.

Table 7. Poverty Clubs vs. Inequality Clubs

+ Inequality - Club1 Club2 Club3 Total N % N % N % N % + Club1 7 53.85% 3 23.08% 3 23.08% 13 100%

Club2 5 50.00% 2 20.00% 3 30.00% 10 100% Club3 6 31.58% 6 31.58% 7 36.84% 19 100%

Club4 12 60.00% 3 15.00% 5 25.00% 20 100% Poverty Club5 14 45.16% 5 16.13% 12 38.71% 31 100% - Club6 7 25.93% 7 25.93% 13 48.15% 27 100%

Table 8 attempts to compare the 3 type of clubs. The results show expected behaviours. For instance, of the 23 countries which comprise high poverty clubs 1 and 2, 11 (47%)

17 Due to comparison reasons, we include the 3 divergent countries considered in section 3.2 in clubs 5 and 6 according to their poverty rate in 2013.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

belong to the low income club 3 and the high inequality club 1. Likewise, 11 of the 27 countries (40%) belonging to club 6, with a less poverty impact, belong to the high income club 1 and the lowest inequality club 3. It is worth mentioning, however, that there are 11 countries that belong to the highest income club and to a low poverty club (Club 5) and at the same time have high inequality rates. Of these 11 countries, 7 belong to the LAC zone (Costa Rica, Panama, Brazil, Peru, Dominican Republic, Ecuador and Jamaica).

Table 8. Poverty Clubs vs. Income and Inequality Clubs Income Club1 Club1 Club1 Club2 Club2 Club2 Club3 Club3 Club3 Clubs Total Gini Club1 Club2 Club3 Club1 Club2 Club3 Club1 Club2 Club3 Clubs N % N % N % N % N % N % N % N % N % N % Club1 0 0.00% 0 0.00% 0 0.00% 1 7.69% 0 0.00% 0 0.00% 6 46.15% 3 23.08% 3 23.08% 13 100%

Club2 0 0.00% 0 0.00% 0 0.00% 0 0.00% 0 0.00% 2 20.00% 5 50.00% 2 20.00% 1 10.00% 10 100% Club3 1 5.26% 3 15.79% 2 10.53% 2 10.53% 0 0.00% 0 0.00% 3 15.79% 3 15.79% 5 26.32% 19 100% Club4 7 35.00% 1 5.00% 2 10.00% 3 15.00% 1 5.00% 0 0.00% 2 10.00% 1 5.00% 3 15.00% 20 100% Poverty Club5 11 35.48% 3 9.68% 5 16.13% 3 9.68% 2 6.45% 5 16.13% 0 0.00% 0 0.00% 2 6.45% 31 100% Club6 7 25.93% 7 25.93% 11 40.74% 0 0.00% 0 0.00% 1 3.70% 0 0.00% 0 0.00% 1 3.70% 27 100%

6. Conclusions

Using the most op-to-date data from PovcalNet, in this paper we review the discussion on poverty convergence. We have employed the methodology of Phillips and Sul (2007) to analyse the existence of convergence clubs, and results show evidence of six poverty convergence clubs. The first three clubs with higher poverty levels tend to group countries from the Sub-Saharan Africa, while clubs with medium and medium- low poverty levels mix countries from different geographical zones. In the club with the lowest poverty levels there is again a high concentration of countries from the European and Central Asian zones, although it is not as high as in the case of the clubs with high poverty levels. We have also estimated convergence clubs in income and inequality, finding evidence of three clubs in both cases. The club with higher income levels is mostly comprised of countries from East Asia and Pacific, Europe and Central Asia and Latin America and the Caribbean, while the club of lowest income is mainly composed of countries from Sub-Saharan Africa. These two clubs have opposite paths, increasing the gap between high and low income countries. In relation to the inequality convergence clubs, it is shown a concentration of countries with high levels of inequality in the LAC zone and, on the contrary, a concentration of low inequality countries in the ECA zone; moreover, it is interesting the case of the SSA zone, with countries covering a broad spectrum of inequality levels despite their low-income status. Further, we have shown strong evidence of a significant negative correlation between poverty and income clubs; 92.3% of countries belonging to the highest poverty club belong, at the same time, to the lowest income club, while 92.5% of countries with the

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs! lowest poverty rate belong at the same time to the highest income club. Moreover, belonging to the highest poverty clubs practically guarantees belonging to the lowest income club. However, as poverty levels decrease, distribution related to income clubs evens out. Thus, these results suggest the existence of nonlinearity which are consistent with the existence of a poverty trap.

7. References

 Abramovitz, M. 1986. Catching Up, Forging Ahead, and Falling Behind”. Journal of Economic History 46(2), 385-406.  Apergis, N., Fontini, F., Inchauspe, J. 2016. Integration of regional electricity markets in Australia: A price convergence assessment. Energy Economics. In Press, Corrected Proof.  Azariadis, C. and Stachurski, J. 2005. Poverty traps. Handbook of economic growth, 1, 295-384.  Barro, R. J. 2000. Inequality and Growth in a Panel of Countries. Journal of economic growth, 5(1), 5-32.  Bartkowska, M. and Riedl, A. 2012. Regional convergence clubs in Europe: Identification and conditioning factors, Economic Modelling, 29(1), 22-31.  Baumol, W. J. 1986. Productivity Growth, Convergence, and Welfare: What the Long-Run Data Show”. American Economic Review 76(5), 1073-85.  Bourguignon, F. 2003. The growth elasticity of poverty reduction: explaining heterogeneity across countries and time periods. Inequality and growth: Theory and policy implications, 1(1).  Bourguignon, F. 2004. The Poverty-Growth-Inequality Triangle. Working Paper 125. Indian Council for Research on International Economic Relations, New Delhi, India.  Camarero, M., Picazo-Tadeo, A. J., Tamarit, C. 2013. Are the determinants of CO 2 emissions converging among OECD countries? Economics Letters, 118(1), 159-162.  Cowell, F.A. 2000. Measurement of Inequality. Atkinson, Anthony B. And Bourguignon, Francois ed., Handbook of Income Distribution, North Holland: Amsterdam, 80 – 166.  Cuaresma, J. C., Klasen, S., Wacker, K. M. 2016. There is poverty convergence. Working paper available at SSRN.  Ferreira, F. H. 2010. Distributions in motion: economic growth, inequality, and poverty dynamics. Working paper available at SSRN  Ferreira, F. H. G., S. Chen, A. Dabalen, D. et al. (2016). A global count of the extreme poor in 2012: data issues, methodology and initial results. The Journal of Economic Inequality, 14(2), 141-172.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

 Fritsche, U., and Kuzin, V. 2011. Analysing convergence in Europe using the non- linear single factor model. Empirical Economics, 41(2), 343-369.  Lopez, H. and Serven, L. 2004. The mechanics of growth-poverty-inequality relationship. Mimeo, The World Bank.  Lopez, H., and Servén, L. 2009. Too poor to grow World Bank Policy Research Working Paper 5012. Washington, DC: World Bank.  Phillips, P. C. and Sul, D. 2007. Transition modeling and econometric convergence tests. Econometrica, 75(6), 1771-1855.  Phillips, P. C., and Sul, D. 2009. Economic transition and growth. Journal of Applied Econometrics, 24(7), 1153-1185.  Pinkovskiy, M., and Sala-i-Martin, X. 2014. Africa is on time. Journal of Economic Growth, 19(3), 311-338.  Pittau, M. G., and Zelli, R. 2006. Empirical evidence of income dynamics across EU regions. Journal of Applied Econometrics, 21(5), 605-628.  Quah, D. T. 1996. Twin peaks: growth and convergence in models of distribution dynamics. The economic journal, 1045-1055.  Ravallion, M. 2012. Why don't we see poverty convergence? The American Economic Review, 102(1), 504-523.  Sala-i-Martin, X. 2006. The world distribution of income: falling poverty and… convergence, period. The Quarterly Journal of Economics, 121(2), 351-397.

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Appendix 1. Sample

Table 9. Sample

Id Country Zone t1 푯풊ퟏ t2 푯풊ퟐ ∆푯풊 Id Country Zone t1 푯풊ퟏ t2 푯풊ퟐ ∆푯풊 1 Albania ECA 1996 1.1 2012 1.06 0.00 61 Latvia ECA 1988 0.03 2012 1.38 0.06 2 Angola SSA 2000 32.28 2009 30.13 -0.26 62 Lesotho SSA 1987 74.77 2010 59.65 -0.64 3 Armenia ECA 1999 16.88 2014 2.31 -0.97 63 Lithuania ECA 1988 0.08 2012 1.03 0.04 4 Azerbaijan ECA 1995 7.32 2008 0.49 -0.53 64 Macedonia ECA 1998 0.8 2008 1.33 0.05 5 Bangladesh SA 1984 38.48 2010 18.52 -0.75 65 Madagascar SSA 1993 69.36 2012 77.84 0.45 6 Belarus ECA 1998 11.33 2014 0 -0.71 66 Malawi SSA 1998 63.63 2010 70.91 0.59 7 Belize LAC 1993 10.09 1999 13.92 0.64 67 Malaysia EAP 1984 2.88 2009 0.28 -0.10 8 Benin SSA 2003 48.85 2011 53.11 0.51 68 Maldives SA 2003 9.99 2010 7.26 -0.39 9 Bhutan SA 2003 35.22 2012 2.17 -3.67 69 Mali SSA 1994 84.88 2010 49.25 -2.24 10 Bolivia LAC 1991 8.57 2014 6.81 -0.07 70 Mauritania SSA 1987 40.05 2014 5.93 -1.26 11 Bosnia and Herzegovina ECA 2001 0.31 2011 0.07 -0.02 71 Mauritius SSA 2007 0.42 2012 0.53 0.02 12 Botswana SSA 1986 42.56 2009 18.24 -1.03 72 Mexico LAC 1984 7.93 2014 3.04 -0.16 13 Brazil LAC 1981 24.32 2014 3.66 -0.63 73 Micronesia EAP 2005 11.42 2013 17.37 0.74 14 Bulgaria ECA 1989 0.05 2012 2.03 0.09 74 Moldova ECA 1997 16.07 2014 0 -0.95 15 Burkina Faso SSA 1994 83.06 2014 43.73 -1.99 75 Mongolia EAP 1995 13.86 2014 0.22 -0.72 16 Burundi SSA 1992 81.12 2006 77.65 -0.25 76 Montenegro ECA 2005 0.23 2014 0 -0.03 17 Cabo Verde SSA 2002 16.01 2007 8.07 -1.43 77 Morocco MENA 1985 11.07 2007 3.12 -0.35 18 Cambodia EAP 1994 30.06 2012 2.17 -1.55 78 Mozambique SSA 1996 85.36 2009 68.74 -1.34 19 Cameroon SSA 1996 48.08 2014 23.98 -1.34 79 Namibia SSA 1994 52.87 2010 22.6 -1.92 20 Central African Republic SSA 1992 84.27 2008 66.26 -1.16 80 Nepal SA 1996 61.9 2010 14.99 -3.20 21 Chad SSA 2003 62.94 2011 38.43 -3.06 81 Nicaragua LAC 1993 36.29 2014 6.22 -1.43 22 Chile LAC 1987 8.46 2013 0.92 -0.29 82 Niger SSA 1993 78.19 2014 45.7 -1.54 23 China* EAP 1981 88.32 2013 1.85 -2.70 83 Nigeria SSA 1985 45.27 2010 53.47 0.33 24 China--Rural EAP 1981 95.59 2013 3.38 -2.88 84 Pakistan SA 1987 62.16 2014 6.07 -2.12 25 China--Urban EAP 1981 59.43 2013 0.51 -1.84 85 Panama LAC 1989 23.75 2014 3.77 -0.80 26 Colombia LAC 1992 8.14 2014 5.68 -0.11 86 Papua New Guinea EAP 1996 53.23 2010 39.31 -1.02 27 Congo, Dem. Rep. SSA 2005 94.05 2012 77.08 -2.26 87 Paraguay LAC 1990 1.19 2014 2.77 0.07 28 Congo, Republic of SSA 2005 50.2 2011 36.97 -2.21 88 Peru LAC 1986 16.58 2014 3.13 -0.47 29 Costa Rica LAC 1981 35.6 2014 1.61 -1.03 89 Philippines EAP 1985 28.08 2012 13.11 -0.55 30 Cote d'Ivoire SSA 1985 6.81 2008 29.02 0.97 90 Poland ECA 1996 2.57 2014 0.01 -0.14 31 Djibouti MENA 2002 20.63 2013 22.52 0.17 91 Romania ECA 1998 2.11 2013 0 -0.14 32 Dominican Republic LAC 1986 38.25 2013 2.32 -1.33 92 Russian Federation ECA 1993 2.37 2012 0.04 -0.12 33 Ecuador LAC 1987 23.16 2014 3.82 -0.72 93 Rwanda SSA 1985 62.46 2014 60.43 -0.07 34 El Salvador LAC 1989 18.98 2014 2.97 -0.64 94 Sao Tome and Principe SSA 2001 29.84 2010 32.28 0.26 35 Estonia ECA 1988 0.06 2012 0.99 0.04 95 Senegal SSA 1991 68.43 2011 37.98 -1.52 36 Ethiopia SSA 1981 69.26 2011 33.54 -1.22 96 Serbia ECA 2002 0.17 2013 0.19 0.00 37 Fiji EAP 2003 5.49 2009 4.07 -0.24 97 Seychelles SSA 2000 0.58 2013 1.06 0.04 38 Gambia, The SSA 1998 70.46 2003 45.29 -4.76 98 Sierra Leone SSA 1990 65.49 2011 52.33 -0.62 39 Georgia ECA 1996 6.04 2014 9.77 0.21 99 Slovenia ECA 1998 0.02 2012 0.03 0.00 40 Ghana SSA 1988 39.95 2006 25.19 -0.81 100 South Africa SSA 1993 29.29 2011 16.56 -0.71 41 Guatemala LAC 1987 50.94 2014 9.32 -1.51 101 Sri Lanka SA 1985 13.27 2013 1.92 -0.41 42 Guinea SSA 1991 92.31 2012 35.27 -2.72 102 Swaziland SSA 1995 81.66 2009 42.03 -2.75 43 Guinea-Bissau SSA 1991 42.96 2010 67.08 1.27 103 Tajikistan ECA 1999 54.37 2014 19.51 -2.32 44 Guyana LAC 1993 33.19 1998 14 -3.49 104 Tanzania SSA 1992 70.42 2012 46.6 -1.20 45 Haiti LAC 2001 55.59 2012 53.91 -0.15 105 Thailand EAP 1981 19.57 2013 0.04 -0.61 46 Honduras LAC 1989 38.6 2014 15.96 -0.91 106 Timor-Leste EAP 2001 44.22 2007 46.76 0.42 47 Hungary ECA 1987 0.06 2012 0.26 0.01 107 Togo SSA 2006 55.55 2011 54.18 -0.27 48 India* SA 1983 53.86 2012 21.23 -1.14 108 Tonga EAP 2001 2.83 2009 1.09 -0.22 49 India--Rural SA 1983 60.03 2012 24.83 -1.24 109 Trinidad and Tobago LAC 1988 0.67 1992 3.41 0.69 50 India--Urban SA 1983 34.2 2012 13.39 -0.73 110 Tunisia MENA 1985 13.93 2010 1.99 -0.47 51 Indonesia* EAP 1984 70.31 2014 8.25 -2.07 111 Turkey ECA 1987 1.61 2013 0.33 -0.05 52 Indonesia--Rural EAP 1984 78.12 2014 8.77 -2.31 112 Turkmenistan ECA 1988 39.4 1998 42.26 0.29 53 Indonesia--Urban EAP 1984 47.19 2014 7.79 -1.31 113 Uganda SSA 1989 87.11 2012 34.64 -2.24 54 Iran, Islamic Rep. MENA 1986 7.4 2013 0.08 -0.27 114 Ukraine ECA 1996 5.37 2014 0.01 -0.30 55 Jamaica LAC 1990 4.56 2004 1.7 -0.20 115 Uruguay LAC 1989 0.46 2014 0.3 -0.01 56 Kazakhstan ECA 1996 6.33 2013 0.04 -0.37 116 Uzbekistan ECA 1998 45.47 2003 66.79 4.26 57 Kenya SSA 1992 23.08 2005 33.6 0.79 117 Venezuela LAC 1981 1.27 2006 9.24 0.32 58 Kosovo ECA 2003 1.74 2013 0.78 -0.10 118 Vietnam EAP 1993 49.21 2014 3.06 -2.17 59 Kyrgyz Republic ECA 1993 44.31 2014 1.29 -2.05 119 West Bank and Gaza MENA 2004 0.28 2009 0.11 -0.03 60 Lao People's Dem. EAP 1992 22.85 2012 16.72 -0.31 120 Zambia SSA 1991 54.05 2010 64.42 0.55

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Appendix 2. Relation between poverty and income

Figure 9. Headcount Ratio vs. Income (logs). Earliest Survey (120 countries)

100

Headcount Ratio Headcount 30

0 3 3.5 4 4.5 5 5.5 6 6.5 7 log income

Figure 10. Headcount Ratio (logs) vs. Income (logs). Earliest Survey (120 countries)

0 3 3.5 4 4.5 5 5.5 6 6.5 7 -1

log Headcount Rario Headcountlog -2

-3

-4

-5 log income

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Appendix 3. Poverty club identification

Club 1 (13 countries) EAC Zone: Uzbekistan (UZB) SSA Zone: Burundi (BDI); Central African Republic (CAF); Congo, Democratic Republic of (ZAR); Cote d'Ivoire (CIV); Guinea-Bissau (GNB); Madagascar (MDG); Malawi (MWI); Mozambique (MOZ); Niger (NER); Rwanda (RWA); Togo (TGO); Zambia (ZMB) Club 2 (10 countries) ECA Zone: Tajikistan (TJK) LAC Zone: Haiti (HTI) MENA Zone: Djibouti (DJI) SSA Zone: Benin (BEN); Burkina Faso (BFA); Gambia, The (GMB); Lesotho (LSO); Mali (MLI); Nigeria (NGA); Tanzania (TZA) Club 3 (19 countries) EAP Zone: Papua New Guinea (PNG); Timor-Leste (TLS) ECA Zone: Georgia (GEO) LAC Zone: Bolivia (BOL); Venezuela, Bolivarian Republic of (VEN) SA Zone: India--Rural (IND-R) SSA Zone: Angola (AGO); Cameroon (CMR); Chad (TCD); Congo, Republic of (COG); Ethiopia (ETF); Guinea (GIN); Kenya (KEN); Sao Tome and Principe (STP); Senegal (SEN); Sierra Leone (SLE); South Africa (ZAF); Swaziland (SWZ); Uganda (UGA) Club 4 (20 countries) EAP Zone: China--Rural (CHN-R); Indonesia (IDN); Indonesia--Rural (IDN-R); Lao People's Democratic Republic (LAO); Micronesia, Federated States of (FSM); Philippines (PHL) ECA Zone: Kyrgyz Republic (KGZ); Moldova (MDA) LAC Zone: Belize (BLZ); Colombia (COL); Guatemala (GTM); Honduras (HND); Paraguay (PRY) SA Zone: Bangladesh (BGD); India (IND); India--Urban (IND-U); Nepal (NPL) SSA Zone: Botswana (BWA); Ghana (GHA); Namibia (NAM) Club 5 (29 countries) EAP Zone: Cambodia (KHM); China (CHN); Fiji (FJI); Indonesia--Urban (IDN-U); Mongolia (MNG); Vietnam (VNM) ECA Zone: Armenia (ARM); Bulgaria (BGR); Kazakhstan (KAZ); Kosovo (XKX); Macedonia, former Yugoslav Republic of (MKD); Montenegro (MNE); Romania (ROM); Turkmenistan (TKM) LAC Zone: Brazil (BRA); Costa Rica (CRI); Ecuador (ECU); El Salvador (SLV); Guyana (GUY); Mexico (MEX); Nicaragua (NIC); Panama (PAN); Peru (PER) SA Zone: Bhutan (BTN); Maldives (MDV); Pakistan (PAK); Sri Lanka (LKA) SSA Zone: Cabo Verde (CPV); Mauritania (MRT) Club 6 (26 countries) EAP Zone: China--Urban (CHN-U); Malaysia (MYS); Tonga (TON) ECA Zone: Albania (ALB); Azerbaijan (AZE); Belarus (BLR); Bosnia and Herzegovina (BIH); Estonia (EST); Hungary (HUN); Latvia (LVA); Lithuania (LTU); Poland (POL); Russian Federation (RUS); Serbia (SRB); Slovenia (SVN); Turkey (TUR); Ukraine (UKR) LAC Zone: Chile (CHL); Trinidad and Tobago (TTO); Uruguay (URY) MENA Zone: Iran, Islamic Republic of (IRN); Morocco (MAR); Tunisia (TUN); West Bank and Gaza (PSE) SSA Zone: Mauritius (MUS); Seychelles (SYC) Diverging countries EAP Zone: Thailand (THA) LAC Zone: Dominican Republic (DOM); Jamaica (JAM).

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Appendix 4. Income club identification

Club 1 (60 countries) EAP Zone: China (CHN); China--Rural (CHN-R); China--Urban (CHN-U); Indonesia (IDN); Indonesia--Rural (IDN-R); Indonesia--Urban (IDN-U); Malaysia (MYS); Mongolia (MNG); Thailand (THA); Tonga (TON); Vietnam (VNM); Timor-Leste (TLS) ECA Zone: Belarus (BLR); Bosnia and Herzegovina (BIH); Bulgaria (BGR); Estonia (EST); Hungary (HUN); Latvia (LVA); Lithuania (LTU); Montenegro (MNE); Poland (POL); Russian Federation (RUS); Serbia (SRB); Slovenia (SVN); Turkey (TUR); Turkmenistan (TKM); Ukraine (UKR) LAC Zone: Bolivia (BOL); Brazil (BRA); Chile (CHL); Colombia (COL); Costa Rica (CRI); Dominican Republic (DOM); Ecuador (ECU); Guatemala (GTM); Guyana (GUY); Honduras (HND); Jamaica (JAM); Panama (PAN); Paraguay (PRY); Peru (PER); Trinidad and Tobago (TTO); Uruguay (URY); Venezuela, Republica Bolivariana de (VEN) MENA Zone: Iran, Islamic Republic of (IRN); Morocco (MAR); Tunisia (TUN); West Bank and Gaza (PSE); SA Zone: Bhutan (BTN) ; Maldives (MDV); Nepal (NPL); Sri Lanka (LKA) SSA Zone: Botswana (BWA); Cabo Verde (CPV); Ghana (GHA); Guinea (GIN); Mauritius (MUS); Swaziland (SWZ); Seychelles (SYC); Uganda (UGA) Club 2 (20 countries) EAP Zone: Cambodia (KHM); Fiji (FJI); Lao People's Democratic Republic (LAO); Philippines (PHL) ECA Zone: Azerbaijan (AZE); Kazakhstan (KAZ); Kosovo (XKX); Macedonia, former Yugoslav Republic of (MKD); LAC Zone: Belize (BLZ); El Salvador (SLV); Mexico (MEX); Nicaragua (NIC) SA Zone: India--Urban (IND-U); Pakistan (PAK); SSA Zone: Burkina Faso (BFA); Chad (TCD); Cote d'Ivoire (CIV); Mali (MLI); Mauritania (MRT); South Africa (ZAF) Club 3 (38 countries) EAP Zone: Micronesia, Federated States of (FSM); Papua New Guinea (PNG) ECA Zone: Albania (ALB); Armenia (ARM); Georgia (GEO); Kyrgyz Republic (KGZ); Romania (ROM); Moldova (MDA); Tajikistan (TJK); Uzbekistan (UZB) LAC Zone: Haiti (HTI) MENA Zone: Djibouti (DJI) SA Zone: Bangladesh (BGD); India (IND); India--Rural (IND-R) SSA Zone: Angola (AGO); Benin (BEN); Burundi (BDI); Cameroon (CMR); Central African Republic (CAF); Congo, Republic of (COG); Ethiopia (ETF); Gambia, The (GMB); Guinea-Bissau (GNB); Kenya (KEN); Lesotho (LSO); Malawi (MWI); Mozambique (MOZ); Namibia (NAM); Niger (NER); Nigeria (NGA); Rwanda (RWA); Sao Tome and Principe (STP); Senegal (SEN); Sierra Leone (SLE); Tanzania (TZA); Togo (TGO); Zambia (ZMB) Diverging countries SSA Zone: Congo, Democratic Republic of (ZAR); Madagascar (MDG)

Marrero, Marrero and Teixido (2017) Poverty Convergence or Divergence? No, Convergence Clubs!

Appendix 5. Inequality club identification

Club 1 (51 countries) EAP Zone: China--Rural (CHN-R); China--Urban (CHN-U); Fiji (FJI); Indonesia--Urban (IDN-U); Philippines (PHL) ECA Zone: Bulgaria (BGR); Georgia (GEO); Latvia (LVA); Lithuania (LTU); Macedonia, former Yugoslav Republic of (MKD); Russian Federation (RUS); Turkmenistan (TKM); Uzbekistan (UZB) LAC Zone: Belize (BLZ); Bolivia (BOL); Brazil (BRA); Colombia (COL); Costa Rica (CRI); Dominican Republic (DOM); Ecuador (ECU); Guatemala (GTM); Haiti (HTI); Honduras (HND); Jamaica (JAM); Mexico (MEX); Panama (PAN); Paraguay (PRY); Peru (PER); Uruguay (URY) MENA Zone: Djibouti (DJI); Morocco (MAR) SA Zone: Bangladesh (BGD); India--Urban (IND-U); Sri Lanka (LKA) SSA Zone: Benin (BEN); Botswana (BWA); Cameroon (CMR); Central African Republic (CAF); Chad (TCD); Congo, Republic of (COG); Cote d'Ivoire (CIV); Gambia, The (GMB); Ghana (GHA); Lesotho (LSO); Mozambique (MOZ); Namibia (NAM); Rwanda (RWA); Seychelles (SYC); South Africa (ZAF); Togo (TGO); Zambia (ZMB) Club 2 (26 countries) EAP Zone: Lao People's Democratic Republic (LAO); Malaysia (MYS); Micronesia, Federated States of (FSM); Papua New Guinea (PNG); Tonga (TON); Vietnam (VNM) ECA Zone: Bosnia and Herzegovina (BIH); Estonia (EST); Hungary (HUN); Poland (POL) LAC Zone: Chile (CHL); El Salvador (SLV); Guyana (GUY); Nicaragua (NIC); Venezuela, Republica Bolivariana de (VEN) SA Zone: Nepal (NPL) SSA Zone: Angola (AGO); Cabo Verde (CPV); Congo, Democratic Republic of (ZAR); Guinea- Bissau (GNB); Kenya (KEN); Malawi (MWI); Nigeria (NGA); Swaziland (SWZ); Tanzania (TZA); Uganda (UGA) Club 3 (43 countries) EAP Zone: Cambodia (KHM); China (CHN); Indonesia (IDN); Mongolia (MNG); Thailand (THA); Timor-Leste (TLS) ECA Zone: Albania (ALB); Armenia (ARM); Azerbaijan (AZE); Belarus (BLR); Kazakhstan (KAZ); Kosovo (XKX); Kyrgyz Republic (KGZ); Moldova (MDA); Montenegro (MNE); Romania (ROM); Serbia (SRB); Slovenia (SVN); Tajikistan (TJK); Turkey (TUR); Ukraine (UKR) LAC Zone: Trinidad and Tobago (TTO) MENA Zone: Iran, Islamic Republic of (IRN); Tunisia (TUN); West Bank and Gaza (PSE) SA Zone: Bhutan (BTN); India (IND); India--Rural (IND-R); Maldives (MDV); Pakistan (PAK) SSA Zone: Burkina Faso (BFA); Burundi (BDI); Ethiopia (ETF); Guinea (GIN); Madagascar (MDG); Mali (MLI); Mauritania (MRT); Mauritius (MUS); Niger (NER); Sao Tome and Principe (STP); Senegal (SEN); Sierra Leone (SLE)