Inequality and growth in the United States: is there asymmetric response at the state level?

ABSTRACT

A previous study that tried to assess the impact of economic growth on income inequality in the U.S. used state-level data and an ARDL panel model to conclude that economic growth worsens income inequality in the U.S. In this article, we use the same data set but an ARDL time-series model applied to each state in the U.S. to show that the above conclusion is only valid in 20 states. Additionally, we use a nonlinear ARDL approach to show that the effects are asymmetric in the short run as well as in the long run. Significant long-run asymmetric effects reveal that in 28 states both an increase and a decrease in real output have worsened income distribution.

KEYWORDS:

• Income distribution
• economic growth
• asymmetry
• state-level data
• United States

JEL CLASSIFICATION:

• O16

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Indeed, lagged error term from (1) is equal to linear combination of lagged-level variables if we solve (1) for ε_t and lag the solution by one period.

² Note that Pesaran, Shin, and Smith (2001, 300) critical values are for large samples. For small samples like ours we rely upon critical values tabulated by Narayan (2005).

³ At the asymptotic level Banerjee et al.’s critical values are almost the same as upper-bound critical values of Pesaran et al.

⁴ Another advantage of this approach is that it performs better in small samples such as ours as demonstrated by Panopoulou and Pittis (2004). Furthermore, the approach also accounts for endogeniety and feedback between variables. As Pesaran, Shin, and Smith (2001, 299) write ‘our approach is quiet general in the sense that we can use a flexible choice for the dynamic lag structure in … as well as allowing for short-run feedbacks.’

⁵ This recommendation is based on dependency between POS and NEG variables.

⁶ For some other applications of this approach see Apergis and Miller (2006), Dell’Anno and Halicioglu (2010), Delatte and Lopez-Villavicencio (2012), Verheyen (2013), Hajilee and Al-Nasser (2014), and Bahmani-Oskooee and Fariditavana (2016).

⁷ Note that by 52 states we mean 50 states, Washington D.C. and Puerto Rico.

⁸ Note that the size of coefficient estimate attached to ECM_t−1 measures the speed with which variables are adjusting towards their long-run equilibrium values. Since data are annual, only 7% of adjustment takes place in one year.

⁹ Other diagnostics associated with linear and nonlinear models are similar to those of the U.S. models reviewed before in that residuals are autocorrelation free in almost all models and each optimum model is correctly specified. Estimates are mostly stable and nonlinear models enjoy a better fit.