2.1. Empirical model
The empirical model used is inspired by the Kuznets hypothesis on the relationship from economic development to income inequality. It is a linear model as often presented in the empirical literature on the effect of migrant remittances on income inequality (Kousar et al. 2019; Sulemana et al. 2019; Anyanwu et al. (2016). Furthermore, following authors such as Adenutsi (2011), Mahmood and Noor (2014) and Vacaflores (2018), the model includes the previous period's inequality measure among the explanatory variables to account for the dynamic nature of inequality. The basic econometric model is then written as follows:
$${R}_{it}={\beta }_{0}+{\beta }_{1}{R}_{it-1}+{\beta }_{2}{GDP}_{it}+{\beta }_{3}{GDP}_{it}^{2}+{\beta }_{4}{Remit}_{it}+\gamma {X}_{it}+{u}_{i}+{ϵ}_{it}$$
1
In the specification (1), R denotes a measure of income inequality, GDP is GDP per capita which measures economic development, Remit represents migrant remittances per capita and X is a vector of control variables. The subscripts i and t, with i = 1, ..., N and t = 1, ..., T, represent country i and year t respectively. The coefficients β0, β1, β2, β3, β4 and γ (vector) are the parameters to be estimated and ε represents the error term.
Since the objective is to analyze the effect of migrant remittances on different income categories, a main modification is made to the model (1). The modification mainly concerns the dependent variable (R). Indeed, instead of the composite indicators usually used to measure income inequality (Gini index, Palma index, Atkinson index), this research uses the income shares held by three different groups of the population classified from the richest to the least rich. The three income categories used in this research are: the richest 10% of the population, the next 40% and the least rich 50% of the population. This categorization is relevant in that it allows for the analysis of the situation of specific groups. A similar approach was adopted by Azizi (2019) who analyzed the effects of migrant remittances on the richest 10% and 20% of the population and on the other hand on the least rich 10% and 20%. Thus, three specifications are described, each of which is relative to a particular category of the population:
$${R}_{it}^{1}={\beta }_{\text{0,1}}+{\beta }_{\text{1,1}}{R}_{it-1}^{1}+{\beta }_{\text{2,1}}{GDP}_{it}+{\beta }_{\text{3,1}}{GDP}_{it}^{2}+{\beta }_{\text{4,1}}{Remit}_{it}+\gamma {X}_{it}+{u}_{i}+{ϵ}_{it}$$
2
$${R}_{it}^{2}={\beta }_{\text{0,2}}+{\beta }_{\text{1,2}}{R}_{it-1}^{2}+{\beta }_{\text{2,2}}{GDP}_{it}+{\beta }_{\text{3,2}}{GDP}_{it}^{2}+{\beta }_{\text{4,2}}{Remit}_{it}+\gamma {X}_{it}+{u}_{i}+{ϵ}_{it}$$
3
$${R}_{it}^{3}={\beta }_{\text{0,3}}+{\beta }_{\text{1,3}}{R}_{it-1}^{3}+{\beta }_{\text{2,3}}{GDP}_{it}+{\beta }_{\text{3,3}}{GDP}_{it}^{2}+{\beta }_{\text{4,3}}{Remit}_{it}+\gamma {X}_{it}+{u}_{i}+{ϵ}_{it}$$
4
In these models, R1, R2 and R3 represent respectively the share of income held by the richest 10% of the population, that of the next 40% and finally that of the least rich 50%. These three variables can to some extent be understood in terms of a high-income class, a middle-income class and a low-income class respectively. There are at least two reasons for choosing these three variables. The first reason relates to the difficulty of having a large amount of data on the income share held by different quintiles of the population for several countries in the region covered by this research. The second reason relates to the high-income inequality in SSA itself. The top decile typically occupies about 40% of income in most countries in the region.
In the three models thus specified, the effect of migrant remittances is given by the coefficients \({\beta }_{\text{4,1}};{\beta }_{\text{4,2}} \text{e}\text{t} {\beta }_{\text{4,3}}\) respectively in specifications (2), (3) and (4). These coefficients will not only show how the different income groups identified are affected by migrant remittances, but also give an idea of which income groups receive the most migrant remittances and to some extent which migrate the most. For example, if the coefficient \({\beta }_{\text{4,1}}\) in Eq. (2) is positive and statistically significant, the conclusion will be that remittances increase the income share of the top 10% of the population. A comparison of the value of this coefficient with the values of the same coefficients \({\beta }_{\text{4,2}} \text{e}\text{t} {\beta }_{\text{4,3}}\) will allow us to draw the conclusion as to the general direction of variation of income inequalities. More explicitly, if the coefficients \({\beta }_{\text{4,1}};{\beta }_{\text{4,2}} \text{e}\text{t} {\beta }_{\text{4,3}}\) are all positive and statistically significant, with \({\beta }_{\text{4,1}}>{\beta }_{\text{4,2}}>{\beta }_{\text{4,3}}\), this would mean that migrant remittances increase income inequality in SSA. Furthermore, in line with the Jones (1998) hypothesis, it would mean that migrants from the richest families are the ones who send the most remittances.
In this research, the first decile of the income distribution, i.e. the share of national income going to the top 10%, is used to capture the share of wealth held by the upper class (high income class). The share of income going to the top 40% of income earners following the top 10% is used as a measure of the share of income held by the middle class (middle-income class). And finally, the share of income held by the bottom 50% of the population is used to assess the income share of the low-income class. All these three variables are taken from the World Inequality Database (WID.world). This is a multi-source database that combines tax, survey and national accounts data. It is the most comprehensive database available to date on the evolution of income distribution in the world.
In terms of explanatory variables, remittances are the main variable of interest. According to the International Monetary Fund (IMF), remittances are the portion of income that migrants (both resident and non-resident) send back to their country of origin, most often to family and acquaintances. In addition to migrant remittances, a set of other explanatory variables called controls is used. The level of development of the economy, as indicated by the Kuznets hypothesis (Kuznets 1955), is also essential in the distribution of wealth. Thus GDP per capita is used as a measure of economic development. The increase in GDP per capita is synonymous with the creation of wealth and therefore additional income. The distribution of income could be affected by the increase in GDP per capita. The work of authors such as Sulemana et al. (2019) and Vacaflores (2018) confirm the role of GDP per capita in the variation of income inequality.
The vector of control variables includes trade openness, foreign aid, urbanization, financial development and inflation. The degree of trade openness, as measured by the sum of exports and imports as a percentage of GDP, is thought to be a determinant of income inequality. For example, Anyanwu et al. (2016) find an increasing effect on income inequality in West Africa while Adenutsi (2011) finds a reducing effect. Empirical work conducted by authors such as Herzer and Nunnenkamp (2012) or Younsi et al. (2019) indicates that foreign aid contributes to increasing income inequality in the countries that receive it. In this research, foreign aid is measured by net official development assistance per capita. The degree of urbanization would also affect income inequality (Sulemana et al. 2019). The importance of urbanization in variations in income inequality had already been highlighted by Kuznets (1955) in the relationship between income inequality and the process of economic development. The literature also shows that financial development plays an important role in income distribution (Koechlin and Leon 2007; Adams and Klobodu 2016). Inflation, as a measure of the general level of price change, also appears to play a role in the income distribution (Younsi et al. 2019; Law and Soon 2020).
The descriptive statistics of the variables are provided in Table A1 in Appendix. This research uses a panel of data covering the period from 2005 to 2019 for a set of 34 SSA countries (see Table A2 in Appendix). The data used in this research come from two databases. Data on income distribution are taken from the World Inequality Database (WID.world 2021). The latter is the largest database on the evolution of wealth distribution in the world. All other variables are taken from the World Development Indicators (WDI) database (World Bank 2021).
The models (2), (3) and (4) that will be used for the empirical analyses have specific features that could prevent the use of standard econometric estimation methods. First, there is the presence of the lagged dependent variable among the explanatory variables. Second, there are potential endogeneity problems related to migrant remittances. Endogeneity here may arise from two main reasons. The first is the likely reverse causality between migrant remittances and income distribution. As much as remittances may change the distribution of income, inequalities in the distribution of income may increase the migration of people and incidentally trigger new remittances. The second reason for endogeneity is the undervaluation of remittances. The remittances used in this research are indeed those that pass through official transfer channels such as financial institutions. However, a good part of the remittances could, for example, reach the recipients through travelers, making it impossible to account for these funds. Finally, the models may suffer from the possible existence of heteroscedasticity of errors. In view of these particularities, the use of standard estimation techniques, such as ordinary least squares (OLS), could lead to biased results. The use of a method more adapted to such a situation is necessary.
The method used in this research is divided into two steps and takes into consideration the different particularities related to the specifications (2), (3) and (4). In the first stage, the Hausman (1978) test is implemented in order to choose between a fixed effects model and a random effects model. Secondly, the method of Driscoll and Kraay (1998) is used in this research. This estimation technique is used because of its many advantages. Firstly, Driscoll and Kraay (1998) method provide consistent standard errors unlike standard estimation methods (Kpodar et al. 2018). Second, as discussed by Cerra and Saxena (2008), this estimation method has the advantage of controlling for endogeneity problems. Finally, the Driscoll-Kraay (1998) estimator is also suitable for dealing with error heteroscedasticity problems. For the robustness analyses, the method of generalized moments in system was used. Indeed, this method makes it possible to deal with the problem of endogeneity of migrant remittances, resulting in particular from their undervaluation.