Inequality and Growth: Uncovering the Main Conclusions from the Empirics

Abstract

This paper is a critical survey of the empirical literature on the effects of inequality on economic growth. We conclude that it is most likely that the disparities found in the results are due to differences in the type of countries and time periods included in the samples, the variable used to measure inequality, the structure of the data, and the estimation techniques. These findings suggest that the mechanisms that link inequality to growth are likely to operate differently in different circumstances, an element that may offer important guidelines for both policy makers and researchers.

Notes

^1. Although the literature often refers to causal effects and that theoretical models can posit causal mechanisms, it is rarely possible in empirical work to identify a relationship as causal. Thus, causation here should be interpreted loosely.

^2. Although our focus is on size distribution, in order to avoid misleading interpretations regarding the relationship between inequality and growth, when relevant, we will briefly refer to other above-mentioned dimensions of inequality throughout our analysis.

^3. For the sake of simplicity, throughout the text we use the term ‘inequality’ to denote size distribution inequality, which is the focus of our study.

^4. More specifically, Bénabou (2000) predicts a globally U-shaped relationship between inequality and redistribution.

^5. Note that this theory considers that increasing savings benefit economic growth through increased investment. However, this is not a unanimously accepted fact. In particular, in Keynesian economics, an increase in aggregate savings, especially during recessions, may lead to a decrease in aggregate demand, thereby hampering economic growth (see, for example, Bhaduri & Marglin, 1990).

^6. Regarding the estimation technique, several papers examine whether changing the way reverse causation is dealt with has a significant impact on the final results, especially because the use of a lagged explanatory variable may not perform so well when variables are highly persistent. In particular, they estimate the growth regression in another way – using two-stage least squares and instrumenting for the inequality measure – and find that the results do not change considerably. Perotti (1996), however, argues that this procedure is not advisable, as it is hard to come up with plausible instruments for income distribution. In fact, many of the instruments that are used in the literature (such as school enrolment ratios, literacy rates and fertility rates) appear also as endogeneous variables.

^7. The D&S dataset has been continuously augmented with more observations. In particular, a remarkable effort has been developed by The United Nations University - World Institute for Development Economics Research (UNU-WIDER), at Helsinki. The UNU-WIDER World Income Database includes the D&S dataset, augmented with information from other sources as they become available (such as the Luxembourg Income Study and Transmonee), thus covering more countries and a longer time span.

^8. For those seeking alternatives to the D&S dataset, the Luxembourg Income Study (LIS) and the University of Texas Inequality Project (UTIP) provide attractive options, as they deal more accurately with problems of data comparability and consistency. The LIS offers a harmonised transnational dataset built from surveys, presenting a rigorous picture of the cross-sectional pattern of variations in income inequality within the OECD. It is, however, mainly restricted to a few of the wealthiest countries and to recent years, making it inadequate for large-scale purposes. UTIP has focused on measuring and explaining movements of inequality in wages and earnings and patterns of industrial change around the world. This work has been based on the use of Theil T statistic of manufacturing wages as an instrument for estimating measures of household income inequality, for a large panel of countries from 1963 to 1999. Given that manufacturing pay is a major source of total income and has been measured with reasonable accuracy using standard procedures in most countries, it can be used to provide stable, reliable and comparable indicators of income inequality (Galbraith & Kum, 2002). Moreover, the Theil T statistic, by assessing the contribution of between-group and within-group inequality to overall income inequality, has unique properties that make it a powerful instrument for producing data and analysing patterns and dynamics of inequality (Conceição & Ferreira, 2000). The UTIP dataset currently has more than 3000 country-year observations.

^9. Although the release of the D&S dataset triggered the development of panel studies, the same did not happen with time-series studies. In fact, time-series research on the inequality-growth effect is very scarce, as, to our knowledge, only two articles of this type have so far been published in journals. Gobbin and Rayp (2008) examine the inequality effect on growth for Belgium, Finland and the United States, finding quite different results in each case. Frank (2009) explores the interaction between income inequality, human capital attainment, and income growth in a sample of US states over the period 1929-2000, finding evidence of a negative impact of a rise in the top income share on output growth.

^10. As can be seen in Table 1, more recent panel data studies (which will be reviewed in the next subsection) also present very diverse results.

^11. Within a distinct framework, and following the structuralist perspective represented by seminal contributions such as that of Pasinetti (1981), there is a research line that stresses the role of the macroeconomic structure as an important determinant of income inequality and its relationship with economic growth. This research line is motivated by the need to integrate Marxian ideas of income distribution with the Keynesian concern regarding the lack of an appropriate treatment on the demand side.

^12. Note that the relationship between income inequality, human capital inequality and economic growth may be more complex than it seems at first sight. In fact, as noted by Castelló (2010), in most developed countries over the last decades, the stock of human capital has increased significantly and has been more equally distributed, which may have positively influenced growth. However, more human capital has not led to higher incomes, since wages have stagnated and their share in global income has declined substantially, signalling greater inequality in the functional distribution of income. Functional distribution of income is likely to influence growth in a different way from size distribution: within the post-Keynesian tradition, a decrease in wage shares may have either a negative or a positive aggregate impact, depending on whether a wage-led or a profit-led demand regime prevails in the economy (for more details, see Bhaduri & Marglin, 1990). Thus, in developed countries we have a situation characterised by a decline in inequality of human capital distribution accompanied by a persistent increase in inequality of income functional distribution, both affecting growth in different ways. This may in part explain the diversity of results found by Castelló (2010) for developed economies.

^13. According to the authors, heterogeneous panel cointegration methods were employed because they are better designed to deal with some problems plaguing previous studies of the inequality-growth nexus: omitted variables, country heterogeneity, endogeneity and averaging data over time.

^14. As the authors show, the capital share in GDP is strongly and positively correlated with the income Gini coefficient, so it can be seen as a proxy of income inequality.

^15. Note that the relationship between income inequality and savings has also been explored within a distinct framework, namely that of gender inequality. In particular, Seguino and Floro (2003) find that an increase in women’s share of wage income leads to a higher aggregate saving rate, the reason being that the propensity to save is higher for women than for men.