A reassessment of inequality and growth in the United States

Abstract

In a recent interesting paper, Frank (2009) investigated the long-run inequality–growth nexus with a large panel of annual data for the 48 states in the United States over the 1945 to 2004 post-war period. By implementing the Pooled Mean Group (PMG) estimators, Frank (2009) concluded that there is a significant and positive relationship between inequality and growth. However, we find that Frank (2009) was actually measuring the effect of inequality on economic development (proxied by the logarithm of the real state income per capita) rather than on economic growth (defined by the first difference of the logarithm of the real state income per capita). To this purpose, we suggest a more adequate specification to reassess the relationship using the same data set and estimation technique. The fresh empirical results indicate that the inequality–growth connection continues to hold in a positive and significant manner, and the findings are robust to alternative lag order structures and income inequality measures.

Keywords:

• inequality
• growth
• pooled mean group estimator

JEL Classification:

• D31
• O40

Acknowledgements

The authors are grateful to M. Hashem Pesaran and Mark W. Frank for making available the code and data, respectively, used in this article.  Any remaining errors are our own responsibilities.

Notes

¹See section 5.1 of Pesaran et al. (1999, p. 627) for a similar description.

²For simplicity, we do not include the state-invariant time effect as Frank does. Moreover, we use δ _ij (used in Pesaran et al., 1999) in replace of β _ij (used in Frank, 2009).

³Similar specification has been used in Loayza and Ranciere (2006) to analyse the effect of financial development on economic growth. See their equations 1(1) and 2(2) but note that their dependent variable y denotes the per capita income growth rate rather than the log of per capita income as in Frank (2009).

⁴In fact, as argued in Pesaran et al. (1999), the approach does not require pretests of unit root and can be applied to stationary as well as nonstationary variables.

⁵In column 3 of table 3 (Frank, 2009), the coefficient on ‘gini’ is not significantly different from 0 either (judged by the estimate 0.061 and its SE 0.071). There is a typo regarding the significance. Moreover, Frank (2009) found that the coefficient on ‘top9099’ is positive but not significant in column 2 of table 3. In contrast, our results show that the ‘top9099’ variable does have a significantly positive coefficient.