What drives inequality and poverty in the EU? Exploring the impact of macroeconomic and institutional factors

Abstract

Employing panel data techniques, we investigate the macroeconomic and institutional determinants of inequality and poverty in the EU over the period 1994–2008. We pay particular attention to the effects of macroeconomic environment, social protection and labour market institutions. The empirical analysis shows that the social transfers in cash, and principally the transfers that do not include pensions, exert a prominent impact on inequality and poverty. Also significant is the effect of the GDP per capita. The impact of employment on inequality and poverty is not empirically sound. The same holds for the labour market institutions; an exception is the union density, which appears conducive to a less dispersed personal income distribution. Importantly, the results support the view that the social protection system acts as a catalyst in determining the effectiveness of social spending and the distributive role of economic growth and employment.

Keywords:

• inequality
• poverty
• macroeconomic environment
• social protection
• welfare regimes
• labour market institutions

JEL Classifications:

• D31
• I32
• I38
• C33

Acknowledgements

The authors are grateful to Anastasia Kostaki and two anonymous referees for valuable comments. Parts of earlier versions of this paper were presented at the First International Conference in Political Economy, September 2010, Rethymno, the workshop on the European year for combating poverty and social exclusion, Hellenic Statistical Authority, December 2010, Athens, and the workshop ‘Social policy and social cohesion in Greece under conditions of economic crisis', Bank of Greece, May 2011, Athens. We would like to thank the participants for useful comments. The usual disclaimers apply.

Notes

1. It should be pointed out that, over the last few decades, the vast majority of empirical works on the determinants of inequality and poverty have largely concentrated on micro-factors, linked to household or individual characteristics. The impact of macroeconomic and institutional factors has been the subject of a much lower number of empirical studies. This reflects the dominant view that inequality and poverty can be mainly attributed to differences in individual characteristics and choices, and can be better explained using as a basis micro- rather than macro-level analyses.

2. For a detailed description of the ‘trickle-down effect’ see, inter alia, Arndt (1983) and Kakwani, Prakash, and Son (2000).

3. See also Eurostat’s national strategic and joint reports on social protection and social inclusion (http://ec.europa.eu/social/main.jsp?langId=en&catId=750), which place particular emphasis on the role of economic growth in poverty alleviation.

4. Note that in these studies poverty is defined using the relative approach. For the difference between the relative and absolute poverty see, for example, Foster (1998).

5. Note, however, that Caminada and Goudswaard (2009) find that the link between social spending and poverty may crucially depend on how social expenditures are defined.

6. The selection of countries and time period is solely based on the availability of comparable data for inequality and poverty by Eurostat. Luxembourg has been excluded due to its particularly low population compared with the rest of the EU-15 countries, and its extremely high GDP per capita, which has led us to classify it as an outlier.

7. The Gini coefficient is the most widely used summary measure of inequality. It is more sensitive to transfers at the middle of the income distribution.

8. In particular, it holds that where n is the number of people (or households) of the population under investigation, y is the disposable income and is the mean disposable income.

9. We hold that, compared with the unemployment rate, the employment rate is a better indicator to capture the impact of employment performance on inequality and poverty. The principal reason is that the employment rate is not influenced by fluctuations in the labour force, as is the case with the rate of unemployment. The latter is likely to decrease due to a decline in the labour force, emanating from a rise in the number of discouraged unemployed or the number of household members (largely women) that cannot participate in the labour market because of child or elderly care. Such a decline in the unemployment rate is not expected to be conducive to lower inequality and poverty, as it has no direct impact on people’s incomes. Note, however, that the results of our analysis are only marginally affected when the unemployment rate is used instead of the employment one.

10. Note that the conservative-corporatist group also comprises Luxembourg.

11. This distinction relies on the work of Esping-Andersen (1990) and on the discussion regarding the features of the welfare systems in Southern Europe (see Leibfried 1992; Ferrera 1996; Petmesidou 1996; Matsaganis 1999).

12. The union coverage is arguably a more accurate measure for capturing the effects of unions on incomes: in many countries workers benefit from collective bargaining even if they do not belong to a union. However, the data on union coverage provided by organisations or researchers do not cover all countries of our sample as well as the entire time span of our analysis. As a result, in the empirical investigation of this paper we inevitably use the union density as a proxy for the effects of collective bargaining.

13. For a detailed description of the way that this index is constructed see OECD (1999).

14. For a more precise definition of the gross replacement rates and a detailed analysis of the difference between the gross and the net replacement rates see Martin (1996), OECD (2004) and Howell and Rehm (2009). In our empirical investigation the gross replacement rates are used instead of the net replacement rates, since the latter are not available before 2001. Note also that the gross replacement rates are available only every second year. Thus, the data have been interpolated to get annual figures.

15. Note also that the Gini coefficient and the poverty rate are highly correlated (r=0.90; n=187). This comes with no surprise since both indices capture aspects of the inequality in the distribution of income: the Gini coefficient captures the total inequality in the distribution of income, while the poverty rate, defined according to the 60% of the median income, focuses on the inequality at the bottom half of the income distribution.

16. It is beyond the scope of this paper to investigate the interactions between the welfare regimes and the labour market institutions. This, however, would be an interesting issue for future research.

17. See also Drukker (2003).

18. We used a modified Wald statistic (see Baum 2001).

19. See, for example, Reed and Ye (2011). The FGLS methodology relies on Parks (1967) and the PCSE estimator has been proposed by Beck and Katz (1995).

20. Α useful extension of our econometric investigation would be to estimate dynamic versions of the regressions under examination using the Generalised Method of Moments (GMM) technique developed by Arellano and Bond (1991), Arellano and Bover (1995) and Blundell and Bond (1998). This would be particularly useful so long as both inequality and poverty are characterised by hysteresis effects. Nonetheless, as Roodman (2009a), Roodman (2009b) has pinpointed, the GMM estimator is appropriate only for panels with a small number of time periods and a large number of units. In our panel this does not hold, as it comprises 15 years and 14 countries. Consequently, when our regressions are estimated using GMM the number of instruments is inadequately high and thereby the p values of Hansen J statistic are close or equal to 1. This implies that no reliable conclusions can be inferred from our dataset by applying the GMM methodology.

21. Time dummies have been used to account for period-specific effects. They have been included in our regressions when they turned out to be jointly significant, using a Wald test.

22. Note that the R ²s remain high even when the time dummies are not included in the regressions.

23. The quantitative effects mentioned below are derived from the estimation results using both the FGLS and the PCSE methodology and refer only to statistically significant coefficients.

24. Note that the weak relationship between employment and poverty is broadly in line with the fact that a high proportion of poor individuals in the EU belongs to households whose head is classified as employed (see Dafermos and Papatheodorou 2012).