Do Rising Top Incomes Spur Economic Growth? Evidence From OECD Countries Based on a Novel Identification Strategy

• Published date: 01 March 2020
• Author: Helmut Herwartz,Yabibal M. Walle
• Date: 01 March 2020
• DOI: http://doi.org/10.1111/roiw.12399

© 2018 Internationa l Association for Res earch in Incom e and Wealth

126

DO RISING TOP INCOMES SPUR ECONOMIC GROWTH? EVIDENCE

FROM OECD COUNTRIES BASED ON A NOVEL IDENTIFICATION

STR ATEG Y

by Helmut Herwartz and yabibal m. walle*

Georg-Augu st-University of Goetti ngen, Germa ny

We investigate the causal relationship between the growth rate of top income shares and economic

growth in 12 OECD economies for the period 1950–2010. To analyze patterns of short- and long-run

causality, we build upon recent advances in structural-vector autoregressive modeling of non-Gaussian

systems. This framework allows us to discriminate between rival transmission channels by means of

dependence tests, since independent shocks are unique for a particular causation pattern. We consider

the share of income accruing to the top 1 percent (



1), to the next 9 percent (



9), and to the top decile

(



10). While structural models display considerable heterogeneity across countries, mean group and

pooled results strongly favor a specific transmission pattern. In particular,



1 has a long-run positive

impact on economic development. This result, which is also confirmed by identified impulse-response

functions, is particularly evident for the post-1980 period.

JEL Codes: C32, D31, O47

Keywords: economic growth, top income, income inequality, structural-vector autoregressive model,

contemporaneous causality

1. introduction

A growing empirical literature has recently constructed long time series data

on top income shares for several OECD and a few non-OECD countries. Top

income shares are often found to be highly correlated with broader inequality mea-

sures such as the Gini coefficient (see, e.g., Atkinson et al., 2011; Burkhauser et al.,

2012; Leigh, 2007). Taking advantage of this high correlation, researchers have

used top income shares data to study the inequality–growth nexus in cases in which

data for broader inequality measures are missing or not of sufficient quality.

However, the top income shares–growth relationship is not just a proxy for the

inequality–growth nexus; it is also an interesting policy issue in its own right. For

instance, while top income earners represent a very small share of the population,

they, however, receive a substantial share of national income (Atkinson et al.,

Note: We thank Stephan Klasen, Robert Schwager, the editor, Prasada Rao, and two anonymous

referees for valuable comments.

*Correspondence to: Yabibal M. Walle, Chair of Econometrics, Georg-August-University of

Goettingen, Humboldtallee 3, D-37073 Goettingen, Germany (ywalle@uni-goettingen.de).

Review of Inc ome and Wealth

Series 66 , Number 1, March 2020

DOI : 10.1111 /roi w.123 99

bs_bs_banner

Review of Income and Wealth, Series 66, Number 1, March 2020

127

© 2018 Internationa l Association for Res earch in Incom e and Wealth

2011).1 Changes in top income shares could thus exert a sizeable impact on several

macroeconomic aggregates; in particular, national income, welfare, and inequality.

Furthermore, top income shares represent (a specific kind of) inequality at the top

quantiles of income distributions. Consequently, distinct transmission channels in

the inequality–growth relationship might be at work when considering the top

income–growth link vis-à-vis, say, the Gini coefficient–growth link.2

Similar to the highly debated role of income inequality in economic growth,

theoretical predictions on the impact of rising top incomes on aggregate economic

growth are not clear a priori. The conventional textbook approach views inequal-

ity (including top income inequality) as good for incentives and, hence, as growth

promoting (Aghion et al., 1999; Mankiw, 2013). However, inequality may hamper

economic growth by diminishing national savings, reducing the number of individ-

uals who have access to credit, undermining social and political stability, and exac-

erbating rent-seeking activities (Galor and Zeira, 1993; Perotti, 1996; Solow et al.;

2014; Todaro and Smith, 2011). In line with the conflicting theoretical predictions,

existing empirical studies on the relationship between top income shares and eco-

nomic growth have documented inconclusive results (see, e.g., Andrews et al., 2011;

Herzer and Vollmer, 2013; Roine et al., 2009).

In this paper, we take advantage of recent contributions to identification in

structural-vector autoregressive (SVAR) models. As shown by Lanne et al., (2017),

Moneta et al., (2013), and Gouriéroux and Monfort (2014), the traditional identi-

fication problem of distinguishing between rival causation patterns—for example,

Cholesky factors (Sims, 1980) or long-run relations (Blanchard and Quah, 1989)—

can be resolved in a data-driven manner in non-Gaussian systems. Specifically,

the detection of independent orthogonalized shocks in non-Gaussian systems pro-

vides external information which allows the testing of otherwise just-identifying

structural assumptions. Taking advantage of non-Gaussianity of growth rates of

income shares and per capita income, we assess the level of dependence between

orthogonalized shocks that are determined under the presumptions of distinct

causation patterns. While alternative profiles of short-run causality refer to poten-

tial links among growth rates, long-run causality profiles represent relationships

among variables in levels. Assuming that the “true” structural shocks are indepen-

dent and non-Gaussian, independence diagnostics allow us to rank overall four

alternative structural hypotheses in their scope to filter out independent shocks

from the data. To diagnose the actual level of dependence of alternatively com-

posed samples of structural shocks, we rely on a recent test of the null hypothesis

of independence of random variables (Bakirov et al., 2006). This test has been

shown to be consistent against any form of dependence. Noting that our identifica-

tion strategy rests on the maximum p-value out of four alternatives, the structural

model selection builds upon the principles of Hodges–Lehmann estimation.

Dictated by data availability, we study the link between top income shares and

economic growth in 12 OECD economies for the post-1950 period. Noting that

2For instance, the top 1 and 10 percent of the U.S. population received about 17.45 and 46.35 per-

cent of the aggregate national income, respectively, in 2010 (Alvaredo etal., 2016).

1For an empirical support to this view, see, among others, Voitchovsky (2005), who reports that

inequality at the top end of the income distribution is growth promoting, while inequality among the

poor has a negative relationship with economic growth.

Review of Income and Wealth, Series 66, Number 1, March 2020

128

© 2018 Internationa l Association for Res earch in Incom e and Wealth

the top decile group is highly heterogeneous (Atkinson et al., 2011; Roine et al.,

2009), we consider three top income inequality measures: the share of income

accruing to the top 1 percent (henceforth,



1), to the next 9 percent (



9), and to

the top decile (



10

=

1 +



9). We find that causality directions linking growth

of per capita income and top income shares display considerable heterogeneity

across economies. However, both mean group averages of identified impulse-

response functions and inferential results from pooled samples strongly favor a

long-run positive impact of



1 on economic activity against other postulated

causal relationships. Conditional pooling reveals that the positive role of



1 in

spurring macroeconomic performance is particularly strong during the post-1980

period—a period in which



1 has been on the rise in most of the economies con-

sidered. Unlike the link between



1 and per capita income, however, the structural

relations between



9 (



10) and per capita income are heterogeneous in direction

and generally more in line with an a priori view that economic growth impacts on

the growth of top income shares in the short run. In fact, long-run causality from

top income inequality to economic activity is significantly rejected in pooled sam-

ples of residuals from



10 or



9 growth rates.

Our result that



1—but not



9—exerts a long-run impact on per capita

income is consistent with both the “superstar” and financial deregulation theo-

ries on the rise of top incomes in recent decades. First, according to the “super-

star” hypothesis, the recent increase in top income shares could be attributed to

globalization and advances in information and communications technology that

have increased the relative productivity of highly talented individuals (Kaplan and

Rauh, 2013; Rosen, 1981). These “superstars” more likely belong to the top percen-

tile than the next top nine percentiles. Moreover, the positive and significant role

of



1 on economic performance is not obtained in the pre-1980 period but, rather,

in the post-1980 period, where the “superstar” hypothesis is more likely to hold.

Therefore, the “superstars” might have been an important driving force behind our

result that



1—but not



9—has a long-run impact on economic activity. Second,

our result is also in line with the hypothesis that the financial deregulation of the

past four decades, through its role in driving up wages in the financial sector, is

partly responsible for the recent rise in top income inequality (Boustanifar et al.,

2018; Tanndal and Waldenström, 2018). Given that these high-wage earners likely

belong to the top percentile earners, the result that



1—but not



9—has a long-

run impact on economic activity could also be reflecting the positive role of finan-

cial deregulation in economic development (see, e.g., Levine, 2005).

If we investigate whether



1 benefits income groups other than the top 1

percent, we find that it indeed drives up the per capita income of the next 9 per-

cent. However,



1 does not exert a statistically significant impact on the per capita

income of the bottom 90 percent. Hence, according to our results, the bottom 90

percent have, on average, neither benefited from “trickle-down” effects nor experi-

enced decreasing group-wise per capita income despite the decline in their share of

aggregate income.

Section 2 provides a brief literature review. Section 3 describes the data, while

Section 4 sketches our methodological approach. The empirical results are pre-

sented and analyzed in Section 5. Section 6 concludes. The first four appendices

(in the online Supporting Information) provide an explicit representation of the