The Effects of International Remittances on Poverty and Inequality in Ethiopia

Abstract

The article studies the effects of international remittances on poverty and inequality in Ethiopia using an urban household survey from 2004. Counterfactual consumption in the hypothetical case of no remittance is estimated in a selection-corrected estimation framework. Inequality and poverty values in the hypothetical and actual cases are then compared. There is a significant reduction in poverty, while inequality does not change. The head count, the poverty gap and the squared poverty gap ratios decreased from 0.41 to 0.38, 0.14 to 0.13, and 0.064 to 0.057 respectively.

Acknowledgments

I thank Halvor Mehlum, Kalle Moene, Jo Thori Lind and Fredrik Willumsen for comments and suggestions.

Notes

1. It is worth noting that the World Bank estimate is based on formal transfers and hence is bound to underestimate the actual size of remittances.

2. The trend of total remittance flow to Ethiopia from 2001 to 2010 is shown in Figure A1a, while the trend of remittance flow as percentage of GDP is presented in Figure A1b, in the Online Appendix.

3. There is also some evidence that remittances serve as insurance against crop failure (Yang & Choi, 2007) and are invested on physical and human capital (Adams, 2006; Adams & Cuecuecha, 2010; Edwards & Ureta, 2003). But remittances may also create a moral hazard problem of increasing unemployment or lowering work efforts (Acosta, 2011; Grigorian & Melkonyan, 2008) thereby reducing the positive effects.

4. The choice of consumption over income as a measure of welfare is motivated by the fact that information on consumption is more reliable than information on income in a developing country context. Consumption is also less volatile than income and hence measures average welfare of households better than income (Deaton, 1997).

5. For a discussion on selection models, see Cameron and Trivedi (2005).

6. Ethiopian emigrants have a norm of sending money to their families during major holidays, though the money is not necessarily spent only for holiday festivities. Thus, it could be argued that a large fraction of them receive remittance regularly.

7. It could be argued that remittance might also affect the household-level variables, particularly the two education variables. The reason why the probability of not receiving remittance is positively associated with the number of household members who have completed primary education and negatively with the number of household members who have completed secondary education might be because remittances are invested on human capita, and hence some of the household members who would have just completed primary education in the absence of remittance may now complete secondary education. Similarly, the positive effect of the number of employed members on the likelihood of not receiving remittance might be reflecting potential negative effect of remittances on employment. But given the fact that most of the remitters stayed abroad only for a few years, the average being six years, it is believed that there will not be a big endogeneity bias in general.

8. An attempt was made to use the instruments that have been used by other researchers, which include age of household head, wealth, regional fraction of remittance recipients and district-level concentration of major ethnic groups. Age of the household is an exogenous variable (though it might affect consumption directly) and hence could be used as an instrument; but it does not have significant effect on the selection equation. While there is no information on how much wealth households own, there is information on the ownership of some consumer durables from which a wealth index could be constructed. However, wealth is likely to be affected by remittance and it will also affect consumption directly. And the data which are collected from seven major cities in Ethiopia does not allow to use fraction of remittance recipients and concentration of major ethnic groups; these variables could meaningfully be aggregated at the city level, and hence there is not enough variation, especially considering that about 60 per cent of the households come from Addis Ababa.

9. The definitions of the three poverty measures are presented in the Online Appendix.

10. The Gini and poverty indices were also computed using consumption per capita without adjusting by equivalence scale and the result does not change, though poverty increases in general as expected.

11. For a detailed discussion on types of migrants’ selection, see Borjas (1989).

12. It was also assumed that those remitters who are close relatives of the household head (children and spouse) would be part of the household if they did not migrate. But it is possible that some of the remitters were not part of the household even if they are close relatives of the household head. On the other hand, some of the remitters who are not close relatives of the household head might have been part of the household. Thus, as an alternative, similar estimation was done assuming every remittance-receiving household has one migrant member abroad – an approach used by other researchers in the absence of any information about the migrant (Acosta et al., 2008; Gubert et al., 2010). The effects on poverty and inequality of remittances do not change. The detailed results on the counterfactual regression estimation and the effects on poverty and equality could be provided upon request.

13. The counterfactual estimation for the main result does not include household income as an explanatory variable. Income is an important determinant of consumption and can also affect the selection equation. But including income in the counterfactual estimation has some problems. The estimated consumption equation is used to construct the counterfactual consumption values for the remittance-receiving households in the absence of remittances. But total household income includes remittances, and given that household income has a positive effect on consumption, this will overestimate the counterfactual consumption for the remittance-receiving households, which in turn might underestimate the poverty reduction effect of remittances. On the other hand, total household income does not include what the migrants would contribute to household income (over and above their consumption) if they did not migrate, and that might lead to an over estimation of the poverty reducing effect of remittances. But the first effect is likely to outweigh because migrants remit significant amount of money. When income is included in the counterfactual estimation, the poverty-reducing effect of remittances decreased as expected. The head-count ratio decreased only from 0.39 to 0.38 and the squared poverty-gap ratio only from 0.061 to 0.057, while there is no reduction in the poverty-gap ratio. Remittances do not affect inequality as before. The regression equations and the effects on poverty and inequality are given respectively in Tables A6a and A6b in the Online Appendix. It might be argued that this problem could be solved by excluding remittances from the household income. But this also has its own problems. First, there will likely be indirect effects of remittances. Second, what the migrants would contribute to household income if they were at home will not be included as before. Whether the poverty-reducing effect of remittances will be overestimated or underestimated will depend on the relative strengths of the two effects. When non-remittance income is included in the counterfactual consumption estimation, the poverty-reducing effects of remittances decreased, suggesting that the indirect benefits of remittances are larger than the contribution of migrants to household income. The head-count ratio fell from 0.40 to 0.38 and the squared poverty-gap ratio decreased from 0.062 to 0.057. The poverty-gap ratio does not change. The detailed results could be provided upon request.