Abiad, Oomes and Ueda (2008) showed that financial liberalisation positively relates to better capital allocation efficiency. In turn, a more productive allocation of capital avoids the concentration of financial resources on rich or well-established firms only (Koudalo & Wu 2022). A theoretical contribution regarding the financial liberalisation and income inequality link was provided by Bumann and Lensink (2016), which includes agents with varying investment abilities (investors and savers) and the banking sector. Financial liberalisation lowers the wedge between interest rates on deposits and loans, thereby improving banking sector efficiency. Improved banking sector efficiency that reduces borrowing costs leads to an increase in aggregate loan demand, which requires an increase in the deposit rate to restore financial market equilibrium. The income of savers improves with an increase in the deposit rate and subsequently income distribution. Conversely, for countries with low financial depth, the interest elasticity of the demand for loans is low, indicating that a banking sector efficiency increase with the related borrowing cost decrease will have a minimal impact on loan demand. In this instance, equilibrium in the financial market requires a decrease in the deposit rate, which reduces savers’ income, negatively affects income inequality.

Similarly, the relationship between FDI and income inequality presents open and unresolved questions. Foreign direct investment inflows can spur country productivity increases as imported technologies spread through the economy decreasing inequality (Avdjiev & Spasova 2022). On the contrary, several reasons exist why FDI could increase income inequality. Jaumotte, Lall and Papageorgiou (2013) noted that financial globalisation, specifically FDI, aggravated a rising inequality trend. A possible explanation is that FDI is often concentrated in higher technology sectors tending to benefit those who already have relatively higher education and skills. Another reason is that multinational enterprises often repatriate profits back to the origin country (Kaulihowa & Adjasi 2018).

Avdjiev and Spasova (2022) summarised the channels through which each component of financial openness may influence inequality (Table 1).

Theoretically, the access to credit channel could operate for all the different financial openness components. The easing of funding conditions with more external finance should enhance access to credit for smaller firms and low-income households (Avdjiev & Spasova 2022). The case is likely stronger for debt flows (mostly cross-border bank loans), considering the bank-based financial systems in developing countries. Relatedly, more financial openness could decrease inequality through the funding conditions channel. Portfolio equity flows may increase inequality through the capital gains channel, as equity holdings tend to be concentrated among the wealthy (Avdjiev & Spasova 2022). If portfolio equity flows are used for tax avoidance and illicit flows, the wealthy disproportionately benefit (Eichengreen et al. 2021).

Foreign direct investment could be associated with increasing inequality through the skilled premium channel. With FDI often concentrated in higher technology sectors, individuals with relatively higher qualifications and skills are more likely to benefit (Jaumotte et al. 2013). The same process driving the skilled premium channel accounts for the creation of the technology diffusion channel, albeit this channel takes longer to emerge and influences inequality in a different direction (Avdjiev & Spasova 2022). Foreign direct investment inflows can allow better technologies to spread through recipient economies, employing more people in high-skilled industries, thereby reducing inequality.

A negative link between the different components of financial openness and inequality is plausible through the foreign exchange rate channel. In the presence of a currency mismatch, local currency appreciation translates into increased creditworthiness for borrowers, increased investment and boosted economic activity (Hofmann, Shim & Shin 2016). Conversely, through the special interest group channel, different components of financial openness may increase inequality. Small elites usually benefit if established interests capture financial liberalisation reforms (Claessens & Perotti 2007).

Empirical work on the relationship between financial openness and income inequality is relatively rare, revealing mixed findings (Koudalo & Wu 2022). Using both de facto and de jure financial openness indicators for a panel of 51 countries, Jaumotte et al. (2013) found that financial globalisation is related to an increase in inequality. Zhang and Ben Naceur (2019) established that financial liberalisation worsened income inequality. Erauskin and Turnovsky (2022) showed that financial globalisation increased overall income inequality. Their findings also suggest that reduced foreign investment costs have a more significant impact on inequality than reduced borrowing costs. Using data from 73 countries from 2000 to 2016 and employing a panel quantile regression approach, Kebede and Tawiah (2023) showed that the impact of financial globalisation on income inequality varied across different de facto and de jure dimensions of financial globalisation. Overall, their results revealed that financial globalisation unfavourably affects income inequality.

Conversely, Li and Yu (2014) found that financial liberalisation in general leads to lower income inequality in Asian countries. The effect was more profound in countries exhibiting greater human capital development. Delis, Hasan and Kazakis (2014) provided cross-country evidence that banking system liberalisation decreases income inequality. Additionally, external capital flow liberalisations and privatisations had a similar effect on inequality. The empirical findings of Bumann and Lensink (2016) showed that financial liberalisation lowers income inequality, conditional on the level of financial depth in the country.

For an African context, Koudalo and Wu (2022) examined the link between financial liberalisation and income inequality using a panel of 51 countries from 1995 to 2018. It was found that income inequality increased with the level of financial liberalisation. A possible explanation is that increasing financial liberalisation encourages banks to distribute financial resources more discriminately to rich clientele while dismissing the poor from access to financial services, thereby widening the income gap. Kaulihowa and Adjasi (2018) examined the impact of FDI on income inequality in 16 African countries. The authors concluded that while FDI improved the distributional aspect of income, a diminishing effect is present with a further increase in FDI, which worsened inequality.

In one of the few studies that investigated the link between multiple de facto financial openness measures (gross external liabilities and their various components) and inequality, Avdjiev and Spasova (2022) found substantial heterogeneity among the different financial openness components. For instance, as opposed to FDI and portfolio debt, the relationship between portfolio equity and inequality was insignificant for most periods. Moreover, an increase in other investment liabilities, which consisted mostly of cross-border bank loans, was associated with a decrease in inequality as opposed to other external liabilities’ components.

The dependent variable (INEQ_it) refers to income inequality. Following other researchers in the field of income inequality, we used top income shares as a measure (Alvaredo et al. 2017). A limitation of using the Gini coefficient is that impacts on various parts of the income distribution are obscured (Erauskin & Turnovsky 2022). The findings of Erauskin and Turnovsky (2022) suggested that most of the impact of financial openness is experienced at the extremes of income distribution.

Financial openness (FO_it) and its various components (i) FDI (FDI_it), portfolio equity (PE_it) and debt (DEBT_it⁾ are the main explanatory variables of interest. The following control variables, as standard determinants of income inequality, are included: GDP per capita and its squared term, trade, inflation, financial development and population. The variables and data sources are described as follows:

INEQ_it = Income inequality is measured by the top 10% and top 1% income share, using data obtained from the World Inequality Database.

FO_it = Stock of foreign liabilities as a percentage of GDP of a country, with data obtained from the External Wealth of Nations Mark II database (updated) of Lane and Milesi-Ferretti (2007).

FDI_it = Stock of foreign direct investment liabilities as a percentage of GDP of a country, with data obtained from the External Wealth of Nations Mark II database (updated) of Lane and Milesi-Ferretti (2007).

PE_it = Stock of portfolio equity liabilities as a percentage of GDP of a country, with data obtained from the External Wealth of Nations Mark II database (updated) of Lane and Milesi-Ferretti (2007).

DEBT_it = Stock of the sum of the stocks of portfolio debt liabilities and other investment liabilities as a percentage of GDP of a country, with data obtained from the External Wealth of Nations Mark II database (updated) of Lane and Milesi-Ferretti (2007).

GDPpc_it = GDP per capita (log) (constant 2015 US$) of a country obtained from the World Bank’s World Development Indicators. GDP per capita and its squared term were included to control the inverted U-shaped relationship between economic growth and inequality.

Trade_it = Trade of a country is the sum of exports and imports of goods and services measured as a share of gross domestic product, with the variable obtained from the World Bank’s World Development Indicators. In both empirical and theoretical literature, the impact of trade openness on income inequality is contentious (Bergh & Nilsson 2010). Eliminating trade barriers may increase the demand for unskilled labour, increasing their income (Koudalo & Wu 2022). However, Gourdon, Maystre and De Melo (2008) showed that inequality increases are positively associated with trade openness in countries with an uneducated labour force.

Inflation_it = Consumer price index (2010=100) of a country obtained from the World Bank’s World Development Indicators. BulÌr (2001) found that lower inflation rates improved income inequality.

Financial Development_it = The Financial Development Index of the IMF ranks countries on the depth, access and efficiency of financial institutions and financial markets. Current theories offer differing expectations about the effect of financial development on income inequality (Altunbaş & Thornton 2020). Demirgüç-Kunt and Levine (2009) noted that finance can operate at extensive and intensive margins.

Population_it = Total population (log) of a country obtained from the World Bank’s World Development Indicators. An increase in population size might be detrimental to the poor if their share of labour income is fixed in the national income (Koudalo & Wu 2022). However, with larger populations, higher productivity can ensue, potentially increasing wages for the poor.

The relationships between overall de facto financial openness, its various components and income inequality are specified using four empirical models. Different empirical models allow for the possibility of heterogeneity among the links between various components of financial openness and income inequality (Equations 1–4).

In panel data analysis, when the cross-sectional (N) and time (T) dimensions are large, it becomes necessary to investigate the time series properties of the panel data and possible cross-sectional dependence. Given the increasingly globalised world, countries may be correlated because of common shocks, regional interconnectedness and spillover effects. For weak cross-sectional dependence cases, the correlation between countries reduces as the number of countries increases. Thus, Pesaran (2015) noted that weak cross-sectional dependence errors do not pose any estimation and inferential problems. However, strong cross-sectional dependence is an issue and should be accounted for in the estimation of panel data. Pesaran (2015) developed a framework to test for weak cross-sectional dependence (CD) in panel, and we employed the procedure to investigate whether the CD is weak or whether there is evidence of strong CD. The CD test (Equation 5) is given as follows:

where ρ_it denotes the correlation coefficient of μ_it and μ_jt given by ρ^ij=T−1∑i=1Tξit ξit and ξ_it are the rescaled residuals defined as: ξit=eit(T−1eiei)12. Under the above setting, the hypothesis is tested as follows:

H₀: γ = 0 (weak CD) versus H₁: γ ≠ 0 (strong CD exist)

Furthermore, the study employed the second-generation panel unit root test developed by Pesaran (2007) that addresses the problem of CD. The technique augments the standard Dickey-Fuller (DF) regression with cross-sectional averages and the lagged levels and differences of the individual series to correct for CD. This allows the standard panel unit root test to rely on the simple averages of the individual cross-sectional augmented Dickey-Fuller (CADF) statistic. The new asymptotic results obtained from the individual CADF and their simple averages are known as the cross-sectional augmented IPS (CIPS) test. The individual CADF_i and the associated CIPS=T−1∑i=1NCADFi statistics are examined as N → ∞ followed by T → ∞ and jointly as N and T approach infinity, such that N/T → k, where k is a fixed, finite, non-zero positive constant (Pesaran 2007). The CADF is simple, intuitive and valid when the N and T dimensions are of the same magnitude. Monte Carlo simulations confirm its strong size and power even for small N and T.

Assume y_it is the observed i^th cross-sectional unit at time t and suppose y_it evolved according to a simple dynamic linear heterogeneous panel data model (Equation 6) given as:

where the initial value, y_i0, has a given density function with a finite mean and variance, and the error term μ_it has a single-factor structure given as Equation 7:

where f_t is the unobserved common factor, and ε_it denotes the individual-specific error terms, and combining equations 3 and 4 yields Equation 8:

where α_i = (1 – Ø_i)μ_i, β_i = –(1 – Ø_i) and Δy_it = y_it – y_i,t–1. The panel unit root test of interest Q = 1 can be expressed as Equation 9:

Letting γ¯=N−1∑j=1Nγj and that γ ≠ 0 for a fixed N as N → ∞. The common factor f_t can be approximated by the cross-sectional mean of y_it, namely y¯t=N−1∑j=1Nyit and its lagged values, y¯t−1,y¯t−2,… ….,y¯t−N. Following Pesaran (2007), a simple case where ε_it is serially uncorrelated, the cross-sectional mean and lagged values are sufficient to filter the effects of unobserved common factors, f_t. The test of the unit root hypothesis, equation 6, is based on the t-ratio of the Ordinary Least Squares (OLS) estimate of β_i in Equation 10 (augmented DF regression):

Finally, the study employed the four-panel structural-based tests of the null hypothesis of no-cointegration developed by Westerlund (2007). Unlike the residual-based cointegration approach, the structural-based technique does not impose any common factor and is designed to test the null hypothesis of no error correction. The four tests are simple and easy to implement. Two tests evaluate the alternative hypothesis that the entire panel is cointegrated, while the other two assess the alternative hypothesis that at least one individual panel is cointegrated (Westerlund 2007). Thus, we implemented the above null hypothesis to evaluate the possibility of a long-run relationship between income inequality and financial openness in Africa.

Next, we employed the advanced quantile via moment (MM-QR) condition developed by Machado and Santos Silva (2019). Studies using mean regressions do not account for the heterogeneity of the relationship between financial openness and income inequality. The MM-QR estimation procedure can reveal disregarded heterogeneous covariance effects in panel data models and allow for models with endogenous explanatory variables. This is important, especially in our case, where the impact of financial openness can vary across the distribution. The MM-QR estimation procedure addresses outliers of the dependent variable by computing changes across its distribution through conditional medians measured in quantile differences. Furthermore, by estimating the location and scale coefficients, MM-QR shows how covariates affect the position and variability of the dependent variable.

Ethical clearance to conduct this study was obtained from the University of Stellenbosch and Stellenbosch Business School Research Ethics Committee (No: 32434).