On the impact of financial development on income distribution: time-series evidence

Abstract

Financial market development is said to have equalizing or unequalizing effects on income distribution. Previous research used cross-sectional and panel data and provided mixed results. Suspecting that they suffer from aggregation bias, we adhere to time-series data and error-correction modeling technique and address the issue one more time in each of the 17 countries for which we have time-series data. In 10 counties, short-run effects of financial market development on income distribution were found to be equalizing. In five countries, the effects were unequalizing. However, the equalizing effects lasted into the long run only in three countries of Denmark, Kenya and Turkey.

Keywords:

• income inequality
• financial markets
• bounds testing
• time series

JEL Classification:

• O10
• O16
• O50

Acknowledgements

Valuable comments of an anonymous referee are greatly appreciated. Any remaining error, however, is our own responsibility.

Notes

¹ Empirical tests of the inverted-U hypothesis have been mixed at best. While Ram (1991), Anand and Kanbur (1993), Fields (1994) and Deininger and Squire (1998) failed to support the hypothesis, Ahluwalia (1976), Papanek and Kyn (1986), Campano and Salvatore (1988) and Eusufzai (1997) supported the hypothesis.

² For some other financial market-related issues, see Ahmed and Wahid (2011), Apergis et al. (2011), Aizenman and Kendall (2012), Leventis et al. (2012), Bose et al. (2012), Dao (2013), Spyrou (2013), Abraham and Harrington (2013) and Montes and Machado (2013).

³ For more on inflation and income distribution, see Easterly and Fischer (2001) and Clarke et al. (2006).

⁴ For review of the literature, see Demirguc-Kunt (2009) and for support of this view see Greenwood and Jovanovic (1990).

⁵ This equality of lagged error term and linear combination of lagged level variables could easily be derived by lagging all variables in Equation 1 by one period and solving it for lagged error term.

⁶ For details of normalization, see Bahmani-Oskooee and Tanku (2008) and footnote 16. Note that in order to calculate the t-ratio for a normalized coefficient, we need the standard error of the ratio of two coefficients. This is done using nonlinear least-square technique and the Delta method (see MFIT4.0 manual by Pesaran and Pesaran (1997).

⁷ For some other applications of this approach, see Bahmani-Oskooee and Hegerty (2007), Halicioglu (2007), Narayan et al. (2007), Tang (2007), Mohammadi et al. (2008), Wong and Tang (2008), De Vita and Kyaw (2008), Payne (2008), Bahmani-Oskooee and Gelan (2009), Chen and Chen (2012) and Wong (2013).

⁸ The time period studied for each country with each financial indicator is different due to unavailability of data. See University of Texas Inequality Project, available at http://utip.gov.utexas.edu/data.html.

⁹ Note that if the size of credit is much higher than the size of deposits, it may suggest potential instability of the financial sector.

¹⁰ Detailed data sources appear in the Appendix.

¹¹ The same analysis could be applied to short-run effects of other variables. In each country, there is at least one model in which other variables carry at least one significant short-run coefficient. Furthermore, Kusnets’s inverted-U hypothesis regarding the impact of economic growth on income distribution receives support in the cases of Belgium, Finland, Greece, India, Israel and Malta since there is at least one model in each of these countries in which initially positive coefficients followed by negative ones.

¹² The exception is Norway where there two measures of financial development carry positive coefficients and two measures carry negative coefficients.

¹³ The exception in this group is India in which the coefficient is positive in one model and negative in another.

¹⁴ Again one exception is Canada in which there is one model in which the trade variable is significantly positive and another model in which it is significantly negative.

¹⁵ The exceptions are Australia where government spending has no effect and Norway where there is evidence of both effects by two different measures.

¹⁶ More precisely, ECM is generated using the following:

¹⁷ The exceptions are the first model in Table 3 (Malta) and two models in Table 6 (Turkey).

¹⁸ For a graphical presentation of the CUSUM and CUSUMSQ tests, see Bahmani-Oskooee et al. (2005).