The Impact of Inequality and Redistribution on Growth

• Published date: 01 June 2019
• Author: Makram El‐Shagi,Liang Shao
• Date: 01 June 2019
• DOI: http://doi.org/10.1111/roiw.12342

THE IMPACT OFINEQUALITY AND REDISTRIBUTION ONGROWTH

by Makram El-Shagi*AND Liang Shao

School of Economics, Henan University, China

In this paper, we reassess the impact of inequality on growth. The majority of previous papers have

employed (system) GMM estimation. However, recent simulation studies indicatethat the problems of

GMM when using non-stationary data such as GDP have been grossly underestimated in applied

research. Concerning predetermined regressors such as inequality, GMM is outperformed by a simple

least-squares dummy variable estimator. Additionally, newdata have recentlybecome available that

not only double the sample size compared to most previous studies, but also address the substantial

measurement issues thathave plaguedpastresearch.Usingthesenewdataandan LSDVestimator, we

provide an analysis that both accountsfor the conditions where inequality is beneficial or detrimental

to growth and distinguishes between market-driven inequality and redistribution. We show that there

are situations where market inequality affects growth positively while redistribution is simultaneously

beneficial.

JEL Codes: O4, I3

Keywords: inequality, growth, redistribution, education

1. Introduction

For decades, the question of whether and how inequality affects growth has

been the subject of open debate in academia. Both the theoretical and empirical

literature are inconclusive as to whether the effect of inequality on growth is pre-

dominantly positive or negative. In this paper, we revisit this question using a new

dataset that has been carefully treated to minimize measurement error and incon-

sistency, and weperform robust inference using fixed effects, thus addressing a

major criticism ofearlier studies that have usedawithin-coun try framework.

The early theoretical literature propagated the idea that inequality could fos-

ter growth because the rich have a higher propensity to save (Kuznets, 1955;

Kaldor, 1957). Particularlywhen assuming imperfect capital markets, this can be

necessary to increase investment and thus economic growth. In the early 1990s, a

political economy–based literature challenged this earlier finding, partlydriven by

cross-country evidence supporting a negative relation between inequality and

growth. Persson and Tabellini (1994) argue that inequality promotes institutions

that prevent the proper protection of property rights. In a similar vein, Alesina

and Rodrik (1994) propose a median voter model where inequality drives taxation

Note: The authors are highly indebted to Peter Egger, the participants of the 2nd HenU/

INFER Workshop on Applied Macroeconomics, and two anonymous reviewers for their valuable

comments.

*Correspondence to: Makram El-Shagi, School of Economics, Henan University, China (mak-

ram.elshagi@gmail.com).

1

DOI: 10.1111/roiw.12342

bs_bs_banner

239

Review of Inc ome and Wealth

Series 65, Numb er 2, June 2019

V

C2017 International Association for Research in Income and Wealth

in a way that reducesgrowth. Alesina andPerotti (1996) strengthenthis general

idea for the case of non-democratic settings, suggesting that rising inequality

drives political instability, which then induces higher social costs that in turn

reduce growth. However, this interpretation was quickly challenged again, for

example, by Forbes (2000), who argues that the results of those models are far

more conditional on the setting than is often portrayed, and by Li and Zou

(1998), who demonstrate that a generalization of both Alesina and Rodrik (1994)

and Persson and Tabellini (1994) reverses the results. Galor and Moav (2004) cre-

ate a unified theory, building a development model where inequality is helpful in

the early stage of development, when capital must be accumulated as the primary

engine to drive growth, but is detrimental lateron,when human capital accumula-

tion is relatively more critical for growth. In general, this third generation of theo-

retical and empirical papers emphasizestheambiguity of theinequality–growth

nexus, thereby mirroring the huge variation in empirical findings.

Empirical evidence is similarly controversial: it ranges from finding negative

to positive relationships, to finding no relationship at all, or suggesting diverse

types of non-linearities. Ever since the seminal papers by Deininger and Squire

(1996, 1998), who collected a wide array of distributional data that allowed panel

inference, most studies have employed a dynamic panel setting with country-

specific effects. Nevertheless, the empirical literature of recent years is subject to a

few substantial problems.

First, the bulk of theliterature inrecent yearshas reliedon systemGMM.

At first glance, this seems like an obvious choice. Due to the importance of eco-

nomic convergence, which necessitates the use of lagged GDP as regressor, growth

regressions are typically dynamic panels with short Tand large N, lending them

to GMM estimation. Because system GMM is more efficient than first-difference

GMM, particularly with persistent variables, it seems perfectly suited. What is

often ignored, however, are the fairly severe mean stationarity assumptions

required for system GMM. Recent studies show that system GMM is greatly

biased when estimating the impact of predetermined regressors in dynamic panels

and is greatly outperformed by a simple least-squares dummy variable (LSDV)

approach (Moral-Benito, 2013).

Second, employing panel data, most studies have some type of fixed-effects

setup (usually estimated via GMM, as noted above) and thereby estimate the

within-country effect of changing inequality (e.g. Li and Zou, 1998; Forbes,

2000). However, this has been strongly criticized by several authors, most notably

Barro (2000). These authors argue that the degree of measurement error, mostly

caused by varying definitions of inequality, is so large that within-country varia-

tion is often driven by chance rather than change. Therefore, they suggest using

random effects models, thus exploiting the cross-country variation more strongly.

While we agree with the Barro critique, we do have some objections to this solu-

tion. The assumptions behind the random effects specification—in particular, the

lack of correlation between thefixed effects and regressors—seemhardly credible

in this context. Yet, a new dataset recently compiled by Solt(2014) and Solt

(2016) makes careful adjustments to account for structural breaks in measure-

ment, thereby alleviating these concerns.

2

240

Review of Income and Wealth, Series 65, Number 2, June 2019

V

C2017 International Association for Research in Income and Wealth