![Sample our Global Development journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5929/TOC-Subject-Sample-2021-Global-Development.jpg"})
Abstract
That researchers look for the inverted-U shape in inequality in the arbitrary periods of arbitrary countries underlies the divergent empirical evidence across studies. To point to the right context for the pattern, this paper establishes a formal mechanism in line with Kuznets' explanation that relates to the industrialization-cum-urbanization phases of closed trade regimes. The mechanism involves an interaction among urban–rural sectoral size differences, agricultural tastes/income, and migration, and predicts an inverted-U shape in inequality in the following way: (i) widening differences in the sizes of urban and rural sectors due to exogenous shocks affect negatively the agricultural tastes/income, worsening inequality; (ii) increasing sectoral size differences and decreasing agricultural tastes/income jointly foster intersectoral migration; (iii) migration acts, in turn, as an equilibrating effect, improving the income distribution. Empirically testing these predictions, non-Sub-Saharan developing countries' data support the mechanism, while data from developed and Sub-Saharan African countries provide little support, as per our prior expectations. This highlights a contrasting evidence on the inverted-U shape across country groups of differing development stages.
Acknowledgements
The authors are grateful to Kaushik Basu, Mustafa Caglayan, Nancy Chau, Gary Fields, Ira Gang, Phillip Hone, Cem Karayalcin, M. Ali Khan, Naci Mocan, Debraj Ray and Dimitrios Thomakos for their invaluable inputs.
Notes
1. There are, however, a few studies that proposed channels to explain the inverted-U shaped pattern of inequality. See, among others, Williamson (1985), Lindert (1986), Anand and Kanbur (1993a), Aghion and Bolton (1997) and Acemoglu and Robinson (2002).
2. Some exceptional studies have investigated the pattern in the specific context that Kuznets suggested − see Anand and Kanbur (1993a). Ahluwalia (1976a, 1976b) do so as well, but by using a log per capita income in an inequality regression. Anand and Kanbur (1993b) criticize Ahluwalia.
3. Sachs and Warner (1995) specify five criteria for a country to be considered as open: (i) average tariffs rates being 40% or less; (ii) nontariff barriers covering 40% or less of trade; (iii) black market premium on the exchange rate being less than 20%; (iv) no state monopoly on major exports; and (v) not being a socialist system. Wacziarg and Welch (2003) explore this classification further.
4. In a great majority of developing countries, import-substitution policies lasted until at least a few decades ago (see Rodrik 1998, 1999a, 2001).
5. Technological enhancements in agriculture have not impacted the farm size much during the period we focus on; Table 12.2 in Ray (1998, 418) verifies that the typical farm in the world was still owner cultivated (i.e., was still a family farm) in 1970.
6. Many studies take the number of workers as the proxy for the size of enterprises; not surprisingly, there are large size differences among countries.
7. Analysing the inverted-U pattern for open trade regimes is clearly an extension of this paper, with potentially important implications. In order to stick to Kuznets' domain, we restrict ourselves to closed trade regimes.
8. The evolution of a large and extensive migration literature can be traced back to Lewis (1954), Todaro (1969), Harris and Todaro (1970), Stiglitz (1974), Calvo (1978), Bhatia (1979), Gang and Gangopadhyay (1987), and Quibria (1988). It must be noted that while some exogenous elements of the Lewis and Harris-Todaro frameworks have been endogenized (e.g., the rural–urban wage gap), prices of agricultural and manufacturing products are assumed to be mostly exogenous due to small open economy assumptions. In addition, agents' labour supply and leisure decisions are typically bypassed in most frameworks.
9. A detailed description of k will be given below.
10. Adam Smith (1937, 5–6) stated: ‘The nature of agriculture, indeed, does not admit of so many subdivisions of labour, nor of so complete a separation of one business from another, as manufactures. It is impossible to separate up entirely, the business of the grazier from that of the corn-farmer. … The spinner is almost always a distinct person from the weaver; but the ploughman, the horrower, the seed sower, and the corn reaper, are often the same.'
11. It is well-known that in Cobb-Douglas utility functions, the portions of income spent on different goods are proportional to the exponents of those goods. Thus, α i ’s portion of each agent i’s income is spent on the agricultural good and portion (1 − α i ) is spent on the manufacturing good.
12. Apart from the obvious simplifying advantages that this assumption buys us, the interruptions experienced by many developing countries in the imports of the essential capital goods provide another justification for this assumption.
13. Although most of the migration literature assumes that individuals compare their pre- and post-migration incomes to decide about migration, it is clear that most people in real-life are motivated by concerns other than income − such as their leisure levels as well as the relative cost of living in rural and urban areas.
14. When pre- and post migration incomes and/or utilities of farmers are equal, we will assume that the farmers will choose to migrate because of weak altruistic preferences. A farmer is said to have weak altruistic preferences if these preferences enter the decision process if and only if the farmer is undecided to migrate or not on the basis of income or utility. The positive social mobility prospect of their offspring will only matter in case they are undecided in terms of their own pre- and post-migration welfare.
15. Corollary 1 does not touch upon the fact that an increase in k also widens the income ratio (and gap) between entrepreneurs and workers.
16. One might consider relating G to α and k only since Theorem 1 relates A to α and k. But recall that Theorem 1 states only the equilibrium conditions among A, α and k. As α and k keep changing, A will have to keep adjusting to the changes in α and k. Hence, Theorem 2 does not only consider inequality at equilibrium levels of A, α and k; rather, it considers inequality given all possible levels of A, α and k.
17. Although for the sake of the model's parsimony, we have taken the levels of α f and k fixed, they are hardly fixed in real life.
18. The countries and the periods are: Bangladesh 65–69, 70–74, 75–79, 80–84, 85–89; Bolivia 80–84; Brazil 70–74, 80–84, 85–89; Chile 65–69, 70–74; Colombia 65–69, 70–74, 75–79, 80–84; Costa Rica 80–84; Dominican Republic 75–79, 80–84, 85–89; Ecuador 65–69, 85–89; Egypt 65–69, 75–79, 80–84, 85–89; El Salvador 65–69, 75–79; Fiji 75–79; Guatemala 80–84; Guyana 80–84; Honduras 65–69, 80–84, 85–89; India 80–84, 85–89; Indonesia 60–64, 65–69; Iran 65–69, 70–74, 80–84; Jamaica 75–79, 80–84, 85–89; Mexico 80–84; Morocco 65–69, 80–84; Nepal 75–79, 80–84; Pakistan 65–69, 70–74, 80–84, 85–89; Panama 65–69, 70–74, 75–79, 80–84; Peru 70–74, 80–84, 85–89; Philippines 60–64, 65–69, 70–74, 80–84; South Korea 60–64; Sri Lanka 65–69, 70–74; Sudan 65–69; Trinidad and Tobago 70–74, 75–79, 80–84; Tunisia 65–69, 75–79, 80–84; Turkey 70–74, 75–79, 80–84; Venezuela 70–74, 75–79, 80–84, 85–89.
19. The countries and the periods are: Burundi 90–94; C. Afr. Rep. 90–94; Cote d'Ivoire 85–89; Ghana 80–84; Kenya 80–84, 85–89; Lesotho 85–89; Nigeria 60–64, 85–89; Senegal 90–94; South Africa 65–69, 80–84, 85–89; Tanzania 65–69, 75–79, 90–94; Uganda 85–89; Zambia 60–64, 75–79, 80–84; Zimbabwe 90–94.
20. The countries and the periods are: Australia 60–64; Bahamas 75–79, 80–84; Canada 60–64; Finland 60–64; Germany 60–64; Greece 60–64; Ireland 60–64; Japan 60–64; Netherlands 60–64; New Zealand 70–74, 75–79, 80–84; Spain 60–64; Sweden 60–64.
21. Rodriguez and Rodrik (2000) closely and excellently examined the dichotomous openness variable of Sachs and Warner by partitioning it into its original components. They conclude, ‘the Sachs and Warner indicator serves as a proxy for a wide range of policy and institutional differences,’ the qualification that we look for in our analysis.
22. Not every country provides a continuous series that can be presented graphically, although point observations of some countries that are not in the graphics have been used in the estimations.
23. Depending on the focus of the papers, it is reported that even a slight change in the state variables, control variables, time span or functional form result in different implications about the Kuznets hypothesis. See Spilimbergo et al. (1999), Schultz (1998) and Barro (2000) for different findings. Li et al. (1998) report that Kuznets' hypothesis finds support in the cross-country dimension of the data, rather than within countries over time.
24. For example, Barro (2000) adds some Gini data to his data set that was classified non-acceptable by D-S, and then uses a dummy variable in the regression to control for this. In our practice, any potential measurement error in Gini would also be captured by the error term, because Gini is a dependent variable.
25. Ginis are based on expenditures versus incomes in the D-S dataset. As suggested by D-S (1996), we add 6.6 Gini points to the expenditure-based Ginis to obtain a consistent series.
26. The squares of the correlations between the residuals of each equation are multiplied by the corresponding sample size, and their sum provides a chi-squared test statistic with degrees of freedom equal to the number of equations.
27. With number of instruments greater than the number of parameters in our equations, the system is said to be over-identified; we conduct the suggested over-identification tests and the null is accepted in every case.
28. Papageorgiou (2003) finds that primary schooling contributes to the productive capacity of the economies, while post-primary schooling adds to their innovative capacity. We experiment with both current level of schooling and schooling in 1960 and find very similar results.
29. For arguments on the influence of natural resources and latitude on the business environment, see Acemoglu et al. (2001) and Easterly and Levine (2003).
30. The control dummies on the construction of the income Gini never had any explanatory power in the regressions, except NA in a few cases, in which case all three are tested to be jointly equal to zero.
31. Note that net income should be distributed more evenly than gross income due to progressive taxation systems, which implies that Iran would have a higher gross income Gini.
32. It is evident that per capita income in economies is an important and strong determinant of agricultural tastes (Engel-law idea). We do not use this variable in the α equation as it would be highly correlated with k (both theoretically and empirically).
33. In an auxiliary regression, k is regressed on a set of exogenous variables and the residuals from this regression are then inserted in the original regressions. The significance of the residuals indicates endogeneity. We employed various sets of exogenous variables in the auxiliary regressions and evidence on endogeneity varies. For consistency, we use only the lagged k in the auxiliary regressions, and the test results are based on this instrument.
34. This might be due to the existence of heteroskedasticity related to k, which is detected, in a single equation context, in equations (1) and (2) through White's (1980) test.
35. The econometric implementation with GMM does not converge due to the need to estimate N+m parameters (where N is the number of countries with more than one observation and m is the number of time-variant variables on the right-hand side of our system), but does with 3SLS. With the GMM and 3SLS results being quite similar, as found above, this is unlikely to cast doubt on our results.
36. It turns out that, with fixed effects analysis, α changes its sign to positive in equation (3) for the base and base+developed sample, while remaining strongly significant and negative in the samples with Sub-Saharan Africa. This implies that it is the dominance of Between-variation in α that results in a negative sign in the former group of countries. We should re-iterate that we can control for fixed effects for only part of the data set.
37. Only the base sample is used for this check. Endogeneity and the Breusch-Pagan tests (unreported) show that MV/AV is endogenous to the dependent variables in all equations. The p-value for the Breusch-Pagan test is 0.15.
38. We also regress MV/AV on k along with the other controls of the respective equations (i.e., equations (1), (2) and (3)), and in the first two equations, we obtain a positive and significant relationship.