The role of financial development in the relationship between income inequality and economic growth: an empirical approach using cross-country panel data

Abstract

The study complements the existing literature on the role of credit constraints in the interplay between income inequality and economic growth. The question “what type of financial development matters for inequality-growth relationship” is answered empirically by adopting a multi-dimensional index of financial development. The analysis covers 35 OECD member countries and 34 non-OECD economies starting from the year 1980 with varying coverage across countries. The results of the panel estimation techniques suggest that in the non-OECD countries, income inequality is positively associated with subsequent growth of per capita GDP under sufficiently developed financial markets. If the markets are poorly developed, the partial correlation between inequality and growth is statistically insignificant. For OECD countries, the association seems to be non-existent although weak evidence for growth-dampening inequality is found if both the level of inequality is high and the financial markets are highly developed. The results imply that promoting the development of financial markets – rather than institutions – may alleviate the adverse effects of income inequality on economic growth in under-developed countries.

Appendices

Appendices

A Data selection for the WIID

As informatively summarized by Jenkins (2015), Atkinson and Brandolini (2001) state that non-comparability in secondary data sets may arise because of differences in the definitions of income, in the data sources or in the processing of the income data in the original source. Differences both within countries in time and across countries may emerge. Many of the differences are associated with predictable patterns on inequality if their nature is not drastically heterogeneous over time and across countries. Unfortunately, the assumption of homogeneity is unlikely to hold for the WIID despite major improvements on the earlier databases and thus the practical implications need to assessed by comparing the WIID series with other sources of at least as good a quality.

In the WIID, each observation is labeled as one of possible income, consumption or expenditure concepts as strongly recommended by the seminal evaluative studies. Following the assertive conclusion of Jenkins (2015), I explicitly report the data selection algorithm inspired by Jäntti et al. (2018) in “Appendix A”. After separating the net income observations from the rest, two issues remain for empirical work: the observations are of varying quality and there are often multiple observations for each country-year pair. Some of the multiple observations are due to multiple surveys but predominantly the measurements come from the same survey and it is just the computation (and the statisticians in charge) that change. Helpfully, the WIID team has introduced a variable called a quality score, which ranks the observations from 3 to 13. By ranking the observations based on this score, presented in “Appendix A”, and picking the highest, I can use the observations of best possible quality to form the final country panel and get rid of many of the duplicate observations. In case of observations tied on the quality score for a given country-year pair, a simple average is taken to obtain unique observations. I believe that this data selection procedure may be helpful for future researchers who need to merge the WIID into some other cross-country panel.

Many recent studies, of which some have received much attention (Ostry et al. 2014), have used the Standardized World Income Inequality Database (Solt 2016, SWIID) as their source for data on the Gini coefficients. The SWIID is based on the WIID, supplemented by other sources and all observations come from its imputation model. In his conclusions, Jenkins (2015) states that costs associated with the use of the WIID are present for the SWIID too. Additionally, he urges to set questions about the imputation model against the benefits of coverage and draws a conclusion that the WIID should be used instead of the SWIID given that the use of the WIID is accompanied by a tractable data selection algorithm.

Data selection in practice. The observation is defined as net income if the WIID4 variable resource_detailed, previously labeled as welfare definition, is one of the following: “Earnings, net”, “Income, net”, “Monetary income, net”, “Monetary income, net (excluding property income)” or “Taxable income, net”; as consumption income if resource_detailed is “Consumption”; and as market income if resource_detailed is one of the following: “Earnings, gross”, “Factor income”, “Income, gross”, “Market income”, “Monetary income, gross”, “Taxable income, gross” or “Taxable income, gross (including deductions)”.

Based on the variable quality_score running from 3 to 13, I rank the observations and pick the highest to use the observations of best possible quality to form the final country panel and get rid of many of the duplicate observations. In case of observations tied on the quality score for a given country-year pair, a simple average is taken to obtain unique observations.

The quality score is defined in the following way by the WIID team (UNU-WIDER 2018): We award points to the observations based on their attributes in the following way (maximum is 13 points). Gini coefficient is available (1). Resource concept: Consumption, Income (net), Income (gross), Monetary income (gross), Monetary income (net) (5), Income, Monetary income, Market income (3), Factor income, Primary income, Taxable income, Earnings (1). Equivalence scale: Per capita or equivalized (3), No adjustment (2). Area coverage: All, Urban, Rural (1). Population coverage: All (1). Distributional share information: All of d1-q5 are available (2), All of q1-q5 are available (at least one of d1-d10 is missing) (1).

Comparative analysis between different data sources. The OECD Income Distribution Database (OECD 2019)¹⁰ provides data on the net income Gini coefficients for its member states. All series correspond to same OECD income definitions and thus they should be more comparable across time and across the OECD countries than the WIID ones. Therefore, the OECD database offers a point of reference to evaluate whether the WIID series differ from series of likely higher quality in a subsample of OECD countries. For non-OECD countries, a comparative exercise would be a cumbersome task since the reference series would have to be gathered from various sources listed under footnote$^{3}$ and improvements on comparability relative to the WIID would be difficult to establish.

Following Atkinson and Brandolini (2001), the WIID, the OECD data and the Standardized World Income Inequality Database (Solt 2016, SWIID)¹¹ are used to form cross-country inequality rankings and to compare within-country inequality trends. The WIID series correspond to the result of the above-described data selection algorithm and thus the observations are averages over the periods 2000–2004 and 2010–2014. Due to gaps in the WIID and OECD data, the comparison is restricted to 14 countries. Other choices of reference years would reduce the ranking sample even further. A single major glitch clearly emerges: the 2010 WIID ranking places Denmark as fifth while all others rank the country in the top two. As can be seen in Fig.  6, the WIID very likely overestimates the level of inequality in Denmark in 1995 and 2000. Otherwise, the rankings are fairly stable although changes occur especially within the Nordic countries who share low levels of income inequality and whose values for the Ginis are close to one another irrespective of the data source. All pairwise correlations between the rankings in a point in time are well above 0.95. (Tables 6 ,7,8,9)

Table 6 Inequality rankings in a subset of 14 OECD countries. Countries are ranked based on the value of the Gini coefficient. The smallest Gini is marked by 1, the largest by 14

Based on graphical analysis, of which Fig.  6 is an example, the WIID in general matches the data provided by the OECD modestly well although some of the time variation is undoubtedly due to differences between surveys or calculations of the income distributions. Canada is an example of a case where all three alternatives paint a similar picture, the SWIID probably underestimates the extent of inequality in Mexico while the WIID series show lower values than the OECD ones for the US (Figs. 7, 8) (Tables 10,11,12,13)

B Full fixed effects panel regression tables

See Tables 7, 8, 9, 10, 11, 12, 13.

Table 7 Parsimonious linear panel growth regression. Fixed effects panel estimation results with only convergence term and the Gini coefficient as explanatory variables, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows

Table 8 Financial development (aggregate index). Fixed effects panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. Columns (2), (4) and (6) correspond to equation (1), where Svir is replaced with FD. Columns (1), (3) and (5) correspond to specifications without an interaction term

Table 9 Development of financial institutions. Fixed effects panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. Columns (2), (4) and (6) correspond to equation (1), where Svir is replaced with FI. Columns (1), (3) and (5) correspond to specifications without an interaction term

Table 10 Development of financial markets. Fixed effects panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. Columns (2), (4) and (6) correspond to equation (1), where Svir is replaced with FM. Columns (1), (3) and (5) correspond to specifications without an interaction term

Table 11 Depth of financial markets. Fixed effects panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. Columns (2), (4) and (6) correspond to equation (1), where Svir is replaced with FMD. Columns (1), (3) and (5) correspond to specifications without an interaction term

Table 12 Access to financial markets. Fixed effects panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. Columns (2), (4) and (6) correspond to equation (1), where Svir is replaced with FMA. Columns (1), (3) and (5) correspond to specifications without an interaction term

Table 13 Efficiency of financial markets. Fixed effects panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. Columns (2), (4) and (6) correspond to equation (1), where Svir is replaced with FME. Columns (1), (3) and (5) correspond to specifications without an interaction term

C Association between the top income shares and growth of per capita GDP

Fig. 6

The disposable income Gini coefficients in Canada, Denmark, Mexico and the United States

Fig. 7

The distributions of top disposable income shares in OECD and non-OECD countries

Fig. 8

Estimated association (95 % level confidence interval) between the top 20 % income share and per capita growth conditional on different measures of financial development, non-OECD countries

D Controlling for endogeneity: system GMM

Fig. 9

Estimated association (95 % level confidence interval) between the top 10 % income share and per capita growth conditional on different measures of financial development, non-OECD countries

$$\begin{aligned}\frac{1}{4} (lnY_{i,t+4}-lnY_{i,t}) &= \gamma lnY_{i,t-1} +\beta _{1}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{top25}\right) + \beta _{2}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{bottom75}\right) \nonumber \\&\quad + \beta _{3}\left( \frac{1}{5}\sum _{j=0}^{4}FM_{i,t-5+j}\right) + \beta _{4}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{top25} \times OECD\right) \nonumber \\&\quad + \beta _{5}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{bottom75} \times OECD\right) \nonumber \\&\quad + \beta _{6}\left( \frac{1}{5}\sum _{j=0}^{4}FM_{i,t-5+j} \times OECD\right) + \beta _{7}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{top25} \times \frac{1}{5}\sum _{j=0}^{4}FM_{i,t-5+j}\right) \nonumber \\&\quad + \beta _{8}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{top75} \times \frac{1}{5}\sum _{j=0}^{4}FM_{i,t-5+j}\right) \nonumber \\&\quad + \beta _{9}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{top25} \times \frac{1}{5}\sum _{j=0}^{4}FM_{i,t-5+j} \times OECD\right) \nonumber \\&\quad + \beta _{10}\left( \frac{1}{5}\sum _{j=0}^{4}Gini_{i,t-5+j}^{top25} \times \frac{1}{5}\sum _{j=0}^{4}FM_{i,t-5+j} \times OECD\right) + \alpha _i + \eta _t + \varepsilon _{i,t}, \end{aligned}$$

(4)

where FM stands for the development of financial markets while otherwise the notation follows equation (1). Since the sGMM estimator is for numerical issues ill-suited for subsample analysis due to the limited data coverage of this study, additional interactions are introduced.

Fig. 10

System GMM estimates (95 % level confidence interval) for the Gini coefficient at the top 25 % and at the bottom 75 % percent on per capita growth conditional on development of financial markets (FM)

Table 14 System GMM estimates for the association between the Gini coefficient and economic growth conditional on financial market development and the level of inequality. System GMM panel estimation, dependent variable: growth of per capita GDP inside non-overlapping five-year growth windows. The model is given in equation (4)