The impact of redistributive policies on inequality in OECD countries

Abstract

Due to behavioural effects triggered by redistributional interventions, it is still an open question whether government policies are able to effectively reduce income inequality. We contribute to this research question by using different country-level data sources to study inequality trends in OECD countries since 1980. We first investigate the development of inequality over time before analysing the question of whether governments can effectively reduce inequality. Different identification strategies, using fixed effects and instrumental variables models, provide some evidence that governments are capable of reducing income inequality despite countervailing behavioural responses. The effect is stronger for social expenditure policies than for progressive taxation.

Keywords:

• inequality
• redistribution
• social expenditure
• progressive taxation

JEL Classification:

• D31
• D60
• H20

Acknowledgements

We would like to thank Mark Taylor (the editor), two anonymous referees as well as Corrado Giulietti, Martin Guzi, Judith Niehues, Sebastian Siegloch and Eric Sommer for helpful comments and suggestions. Peichl is grateful for financial support from the Deutsche Forschungsgemeinschaft DFG. The usual disclaimer applies.

Notes

¹ The OECD has started the New Approaches to Economic Challenges (NAEC) initiative with the aim to help countries not only deliver strong growth, but also to support a fairer distribution, among others. The key elements of this exercise include revisiting the fundamental assumptions about the functioning of the economy and to better address the synergies, complementarities and trade-offs between policies (see OECD, 2012c for an overview). Our article contributes to the aims of this initiative by providing evidence of the inequality reducing potential of government policies when taking into account the equity-efficiency trade-off.

² See Niehues (2010) for a more thorough discussion of this argument.

³ We, however, refrain from merging different data sources as inequality data are usually not comparable across data sources.

⁴ This is certainly partly due to the inherent problems of endogeneity and reverse causality. Issues of data quality, as discussed in Section II, are another reason.

⁵ In their survey of inequality data, Atkinson and Brandolini (2001) provide many examples of inconsistencies both across and within data sets.

⁶ We utilize the Quality of Government (QoG) Social Policy Dataset provided by Samanni et al. (2010). It combines and merges different country-level data sources, among which are all inequality data sources we employ in our analyses.

⁷ The initial Deininger and Squire data are among the most widely used data sources in the literature on inequality. The data set only includes measures of the Gini coefficients that meet three conditions: data are (i) based on household surveys, (ii) cover a sufficient share of the population and (iii) considers a comprehensive coverage of different income sources. However, it is yet criticized for its various inconsistencies, most strikingly in the use of different income concepts (e.g., Atkinson and Brandolini (2001)). The original data only run until 1996, which is why we rely on the UNU-WIDER extension.

⁸ Income versus expenditure, gross versus net of taxes, household versus personal unit of analysis.

⁹ Following the literature, we sort countries into the following welfare regimes: Liberal (Canada, Japan, Switzerland, US, South Korea, New Zealand, Mexico), Social-democratic (Denmark, Finland, Iceland, Norway, Sweden), Conservative (Austria, Belgium, France, Germany, Luxembourg, Netherlands), Radical (Australia, Ireland, UK), Southern (Greece, Italy, Portugal, Spain, Turkey) and Eastern (Czech Republic, Hungary, South Korea, Poland, Slovakia).

¹⁰ Also see Acemoglu and Robinson (2002) for theoretical background.

¹¹ Roine et al. (2009) additionally control for measures of financial asset prices and profit income which could have an impact on inequality. Unfortunately, such information is not available to us as the data used by Roine et al. (2009) is only available for a very limited set of countries.

¹² The country fixed effects also account for possible systematic and time invariant differences in the measurement of inequality.

¹³ We choose the Gini coefficient not only because it is the most widely used measure with the biggest data coverage but also because it takes into account the whole income distribution (albeit being most sensitive to changes in the middle). The Luxembourg income study (LIS) data also contains information about percentile ratios. Employing those in a robustness check (not shown) as left-hand side variables broadly confirms the results found with the Gini coefficient. Interestingly, but not surprisingly, social expenditure (tax progressivity) is more important in reducing inequality in measures which are more sensitive at the bottom (top) of the distribution.

¹⁴ Note that a long literature on the (optimal) progressivity of the income tax system exists (see Piketty and Saez (2013) for a recent survey). However, there is very little evidence on the causal effect of (progressive) taxes on inequality – as discussed in Section I.

¹⁵ This approach builds on the first-difference GMM estimator originally proposed by Arellano and Bond (1991).

¹⁶ Angrist and Pischke (2009) note that insufficient first-stage results in exactly identified models do not do any harm except causing second-stage SEs to be large. This is why we abstain from (i) discussing the relevance of the instruments in detail and (ii) only focus on sufficiently precise estimates in our interpretations of the results.

¹⁷ This is inherent to many other data sources, in which cardinal measurement is critical to assume. See, for example, the discussion on happiness research in Frey and Stutzer (2002).

¹⁸ Another possible line for future research is along the lines of the works by Jäntti and Jenkins (2010) and García et al. (2013). Although both papers are based on the inequality effects of macroeconomic factors in general, the approaches suggested have the potential to expand this type of analysis to measure the effects of policies on inequality.