Does a rising tide raise all ships? The impact of growth on inequality

Abstract

Using semiparametric methods and an up-to-date panel dataset on income inequality, the impact of past economic growth on current inequality is examined in a group of nations with widely differing initial incomes. Regardless of a nation's initial development, past economic growth stimulates inequality over short and medium-run periods. However, in the long-run, inequality in developed and developing nations reacts differently to comparable rates of economic growth. Specifically, inequality declines with growth in developing nations, while rising with growth in developed nations.

Acknowledgements

I would like to thank Jang-Ting Guo, Aman Ullah and seminar participants at Salisbury University for their helpful suggestions. Any remaining errors are my own.

Notes

¹ A total of three datasets are used, corresponding to the lagged growth horizon utilized. The ‘5-year’ dataset contains 252 observations on 55 countries, while the ‘10-year’ dataset contains 246 observations on 54 countries and the ‘20-year’ dataset contains 171 observations on 46 countries.

² The dependent and independent variables are expressed in deviations from country-specific means prior to estimation.

³ The authors of WIDER (2000) suggest making adjustments to inequality observations to correct for differences in observational units and/or income measures.

⁴ Interestingly, when controlling for differences in income measure (rather than reference unit) and repeating the above exercise, it was found that the average difference in individual and household based inequality measures was only − 0.120005 (on a 100-point scale). Thus, the data are not adjusted for differences in reference units.

Table 2. Average difference between inequality measures

⁵ Gaussian product kernels are used, with window widths obtained by minimizing asymptotic mean integrated square error (AMISE) (see Pagan and Ullah (1999), p. 25 for more details).

⁶ The weak statistical significance of the education regressors is common in the development literature, and is symptomatic of multicollinearity between the human capital measures.

Table 3. Linear coefficient estimates

⁷ Nations with 35, 50 and 65th percentile values of beginning of period GDP are referred to as ‘developing’ ‘typical’ and ‘developed,’ respectively.

⁸ This pattern is also repeated in the medium-run (10-year growth periods), and is provided in Fig. 2.

Fig. 1. Estimates of g(·) over previous 5-year growth period

Fig. 2. Estimates of g(·) over previous 10-year growth period

⁹ The Yatchew (1999) specification test cannot be applied to Equation 2.3(3) due to the requirement that the two subsets of data (over which the nonparametric function is being tested) must overlap. If this requirement is not met (as would be the case if data were sorted and grouped by lagged GDP in Equation 2.3(3)), the underlying asymptotically normal test statistic degenerates. As a result, Model (2.1) is used to conduct the test.