This study adopted a simplified version of endogenous growth theory. Endogenous growth theory emphasizes the significant influence of endogenous variables such as macroeconomic policy and policy volatility on long-term growth. Specifically, macroeconomic policy is expected to have a substantial effect on economic growth, but the impact of these macroeconomic factors on heterogenous growth components (gender objectives) remains underexplored and less understood. Using an endogenous framework, this study explores outliers in the growth and endogenous factors relationship to reach conclusions capable of redefining policies and research on the subject. We seek quantitative answers in the links between macroeconomic policy and gender links for some factors. On the one hand, higher policy volatility could lead to conservative household measures, due to precautionary motives, resulting in lower labor force participation. On the other, it increases risk-adjusted returns, which increases investment as well as economic growth, thus leading to increased female labor participation (Bloom, Canning, Fink, & Finlay, 2009). It is therefore clearly expedient to unravel the correlates, magnitude, and policy implications of the macroeconomic policy-gender relationships using the endogenous growth framework since whatever can lead to changes in gender imbalances are most likely to influence growth objectives.

In accounting for the dynamics of the relationship between macroeconomic policy and gender imbalances in SSA, this study follows Stratigaki (2005). The empirical strategy is to estimate a series of baseline fixed-effect estimators by assuming that all explanatory variables are strictly exogenous. Second, we estimate dynamic panel data GMM estimators and impose (and test) the common factor restrictions to account for the potential endogeneity of macroeconomic policy and gender imbalances within the Sub-Saharan countries. We estimated the dynamic system GMM for some factors. Rather than follow a conventional full-sample distribution, GMM uses assumptions about specific moments of random variables and that makes it more robust than maximum probability at the expense of some efficiency (Adekunle, Onanuga, Akinola, & Ogunbanjo, 2020). The dynamic system GMM enables the most flexible identification and is highly consistent and asymptotically normal when dealing with large sample properties (Hansen, 1982) particularly with a large cross-section and a time series dimension with T < 25. It can also effectively integrate moment conditions when the estimator is over-identified. Our macroeconomic policy-gender model takes the functional form:

where GENDER_it is gender equality in country i over period t; MACRO_POLit is macroeconomic policies in country i over period t, t is the time-series dimension of the scope that the study covered (1993 through 2017), and i is the domain that contains the cross-sectional characteristics of the data (the Sub-Saharan Africa countries under investigation).

If the assumption of strict exogeneity on macroeconomic policy for gender equality is violated, our baseline fixed-effects estimator is potentially inconsistent. Therefore, to obtain asymptotically consistent parameter estimates, we estimate single equation dynamic GMM estimators by using a common factor representation (Blundell & Bond, 2000).

The dynamic panel regression model to capture the relationship between macroeconomic policy and gender imbalances is specified as follows:

where, GENDER_it gives gender equality which assesses the extent to which country i has installed institutions and programs to enforce laws and policies that promote equal access for men and women in education, health, the economy, and protection under the law; ρ gives the value of the dependent variable when explanatory variables are zero; MACRO_POLit denotes macroeconomic policies for country i over period t; X_jit is the other regressors included in the model as control variables for country i over period t; j is the number of included control variables, and ω, δ, and θ are the parameter estimates measuring the impact of explanatory variables on the dependent variables.

A country-specific fixed effect is assumed for the disturbance term as follows:

where ε_it represents the error term. It entails e_i, which represents country-specific fixed effects that are time-invariant, while, μ_it is assumed to be independent with zero (0) mean and constant variance σμ2 both over time and across countries that is, uit≈n0,σμ2.

To adjust for the violation of the orthogonal assumption in the dynamic model in (2), we differenced the equations as

However, estimating the ordinary least square on the first differenced dynamic panel model still violates the strict exogeneity assumption since the transformed error term Δμ_it still correlates with GENDER_it-1 as both contain μ_it-1. The possibility of EGENDERit-h∆μit=0∀h≥2, t=3,⋯⋯T makes it possible to use the lagged variables as instruments to adjust the explanatory variables to be orthogonally consistent as in Anderson and Hsiao (1982), Blundell and Bond (2000), Blundell, Bond, and Windmeijer (2000).

In setting a clear line of thought on the relationship between macroeconomic policy and gender in SSA, we estimated cross-country panel data for forty-eight (48) SSA countries based on regional classifications over a period of 25 years, from 1993 through 2017. This scope, governed by data availability, allows the researchers to establish the influence of heterogenous macroeconomic policies for the realization of gender objectives across SSA countries since the IMF and World Bank austerity measures were abolished in 1992. The structural adjustment programme (SAP) era witnessed significant state-imposed austerity measures which limited consumables through a shortfall in money supply and a mindset of prioritizing exports over imports to achieve a favorable balance of payments, although never materialized. It will be interesting to see how well gender objectives have fared since the policies were abolished and alternative measures proposed in many African nations by various political regimes.

Gender Equality GENDER was measured using data from the gender parity index rating for SSA countries in consonance with Shannon, Im, Katzelnick, and Franco (2013). The gender parity index for the gross enrolment ratio in primary and secondary education is the ratio of girls to boys enrolled at primary and secondary levels in public and private schools. Macroeconomic policies considered in this study are fiscal policies and monetary policies series. The monetary policies component measures used are nominal exchange rate as in Oseni, Adekunle, and Alabi (2019) and inflation as in Mboweni (1999). Fiscal policies measures used were net government expenditure (net general government expenditure includes all current government expenditures and revenues for purchases of goods and services and income received) as in Oseni, Akinbode, Babalola and Adegboyega (2020) and taxes on income, profits, and capital gains as used in the work of Casale (2012). We controlled for population growth as in Kibirige (1997) and technological change as in Asongu (2013). These controls were relevant to avert problems of omitted variable biases and because of their high relevance in explaining differences in the realization of gender objectives in SSA. The data used in this study are secondary data spanning the period from 1993 through 2017 and derived from the UNESCO Institute for Statistics, the IMF database, and World Bank Development Indicators for various issues up to 2017.

In accounting for the dynamics of macroeconomic policies and gender imbalances in SSA, the study made use of a four-prong econometric procedure. First is the pre-estimation assessment using a descriptive statistics method to help show, describe, and summarize the data in a meaningful way and also to know if the data are normally distributed through their averages and Jarque-Bera values (Gujarati & Dawn, 2009). Second is panel unit root testing to ensure the variables under investigation are covariance-stationary. The tools used here for detecting non-stationarity of the data are the panel unit-root tests developed by Levin, Lin, and Chu (2002), and Im, Pesaran, and Shin (2003). The more traditional unit-root tests, such as the Dickey-Fuller, Augmented Dickey-Fuller (ADF), Phillips-Peron, and KPSS tests, may also be applied to serve the same purpose. However, those univariate/single-equation methods are well known for their low power in small samples. By contrast, the panel unit-root tests can be more potent than the conventional tests since they combine the information from the time-series dimension with that from the cross-sectional dimension.

Since the pioneering work of Levin et al. (2002), several panel unit-root tests have become available. Here we use the tests developed by Levin et al. (2002) and Im et al. (2003). As in the literature, the tests are based on estimating the model:

where y_it denotes the variable y observed for the ith of N entities in the tth of T periods, and Δ is the difference operator. The LLC test involves the null hypothesis H₀ : ρ_i = 0∀i against the alternative H_A : ρ_i = ρ < 0∀i. The Breitung test does not employ a bias adjustment, which results in a substantially higher power than that of the LLC test. The IPS test involves the same null hypothesis as the last test, but its alternative hypothesis allows for non-stationarity for some individuals. The idea of IPS is to compute the average of the individual ADF test statistics.

Once stationarity of the variables has been verified, the dynamic system GMM was used to account for the structural dynamics of the model.

As presented in Table 3 below, one of the first panel unit root tests formulated by Levin et al. (2002) suggests the following hypotheses for testing stationarity in panel data. Under the null hypothesis, the LLC test shows that each time series contains a unit root, i.e., H₀ : ρ_i = 0∀i, and for the alternative hypothesis, each time series is stationary, i.e., H_A : ρ_i = ρ < 0∀i. Like other unit root tests in the literature, the LLC assumes that the individual processes in each cross-section are independent. The LLC test is mainly based on the estimation of the following equation;

where i=1, 2 … N, t =1, 2 …T

This test might potentially be treated as a pooled Dickey-Fuller or augmented Dickey-Fuller test with different time lags across the units of the panel.

The IPS test, formulated by Im et al. (2003), is an extension of LLC test incorporating heterogeneity in the dataset under the alternative hypothesis. Here, IPS test estimation is also based on Eq. (5). The null hypothesis is stated as H₀ : ρ_i = 0∀i against the alternative hypothesis of H_A : ρ_i < 0 where i=1, 2, 3, …, N₁; ρ_i = 0, i = N₁ + 1, N₁ + 2, …, N.

In the IPS test, it is presumed that all series are non-stationary under the null hypothesis and a fraction of the series is stationary under the alternative hypothesis. This is how it differs from the LLC test, in which all series are supposed to be stationary under the alternative hypothesis. The outcomes of LLC and the IPS test are shown in Table 3. All tests confirmed that variables are non-stationary at levels and stationary after first differences. It is thereby inferred that variables are first difference stationary.

The outcome of the dynamic two-step system GMM is contained in Table 4. The coefficient of the lagged dependent variable is positive and statistically significant at a 5% level. This conforms with theoretical predictions that the previously attained level of gender equality influences prevailing gender equality. The trend in the gender discourse particularly in terms of education, female representation in governance and management boards across major organizations is essential in scaling new heights in equal gender representation and pay (Adekunle, Tella, & Adelowokan, 2020). Thus, a percentage increase lagged dependent variable will result in 0.476% increase in the attainment of gender equality in SSA. Further evidence from this study revealed that the two fiscal policy measures have contrasting effects on gender attainment in SSA, with an increase in government expenditure having a positive effect and taxes on income, profits, and capital gains having a negative influence on gender equality in SSA (both at a 5% level of significance). Based on this, we aver that an appropriate fiscal policy interaction that takes into consideration the socio-economic status of SSA residents will most likely produce the most gains in the pursuit of an egalitarian society that is free from gender imbalances. Explicitly, we have established that a percentage increase in government expenditure and taxes on income, profits, and capital gains would respectively lead to a 0.159% increase and a 0.324% decrease in the gender disparity index in SSA. These results align with the findings of Anyanwu (2016) in their analysis of gender and youth employment in Africa.

In other findings, monetary policy indices have a mixed effect on the gender disparity index in SSA. With negative and positive relations with the gender disparity index at a 5% significance threshold, we establish that a 1% percentage increase in the exchange rate and inflation will lead respectively to a 0.562% decrease and a 0.493% increase in the gender disparity index in SSA. These results are similar to those found in Diouf and Pépin (2017); those in Cho (2016) could be due to undesirable exchange rate fluctuations, which caused commodity prices to increase beyond an acceptable threshold. The uneven exchange rate and commodity price factors have consequences for a large pool of SSA women who are predominantly engaged in jobs located in the informal economy. A persistent rise in general price levels could be interpreted as additional cash flow in an economy largely dominated by women, and in turn, as the labor participation rate increases, female income increases and female welfare also increases. However, the worsening exchange rate hurts these growth trajectories in SSA, which is overly import-dependent, causing a decline in purchasing power parity and leading to hardship.

This study found that population growth worsens gender parity gains, an inverse relationship confirmed in our study. At a 5% level of significance, a 1% percentage increase in population will result in a 0.683% decrease in gender parity in SSA. This result affirms that even under the most favorable conditions, gender parity objectives may be difficult to realize when the population continues to grow as is currently the case in SSA. Population growth makes shared resources increasing difficult to redistribute evenly, keeps pressure on female quota representations, and could even leave some women at a disadvantage in navigating their socio-economic desires. However, technological change could lead to faster realization of gender objectives. With a 1% percentage increase in technological change leading to a 0.243% increase in gender parity in SSA (5% level of significance), this study found similar results to Murage, Pittchar, Midega, Onyango, and Khan (2015). The positive relationship between technological change and gender equality could be due to recent advances in globalization and smart technology that has seen economies transform from industry-based to service-centric. The path of green evolution could also explain this relationship since physical interactions are beginning to play less of a role in modern economic development. Even with a public health crisis of COVID-19 dimensions, the role of technological change has spread beyond IT firms to other sectors that are necessarily not driven by IT capacities. All these structural transformations have exposed age-long dichotomies as regards gender emancipation and the role of macroeconomic policies in gender balances.

We tested for the validity of the instrument used in the system GMM technique. Compared to the OLS model system GMM does not assume normality, and it allows for heteroscedasticity in the data. Irrespective of the kind of model, dynamic panels are known for heteroskedasticity problems in the data set, which can be controlled (Kittler et al., 2000). The system GMM approach assumes linearity and that the error terms not autocorrelated, justifying the need to test for the validity of the instruments through the examination of the first order and second-order autocorrelation in the disturbance term. In tandem with Arellano and Bond (1991) the GMM estimator requires the presence of first-order serial correlation and not second-order serial correlation in the residual term. Since the null hypothesis inference assumes no first-order and second-order serial correlation, we reject the null hypothesis in the first-order serial correlation and accept the null hypothesis for the second-order serial correlation test in order to obtain appropriate diagnostics. The result above confirms the existence of first-order serial correlation since the null hypothesis of first-order serial correlation was rejected (z = -2.54; p < 0.05) at a 5% significance level and no second-order serial correlation since the null hypothesis of no second order serial correlation is accepted because calculated z is not statistically significant at 5% (z = -0.84; p > 0.05). This, therefore, supports the validity of our model specification. The Hansen J-statistics test the null hypothesis of correct specification and valid overidentified restrictions, i.e. the validity of instruments as demonstrated by Oguzie, Onuoha, and Onuchukwu (2005). They argued further that Hansen J-Statistics is the most commonly used diagnostics test in GMM estimation for assessment of a model’s appropriateness. The results of the Hansen J-Statistics test of overidentifying restrictions do not reject the null hypothesis at any conventional level of significance (p > 0.05; i.e p)= 0.851, thus, confirming the model has valid instrumentation. The F-statistics value for all the variables are jointly significant at a 5% level of significance.

In order to check the validity of the system GMM results, the study also employed pooled OLS and fixed effects in line with Blundell, Bond, and Windmeijer (2000). They suggested additional detections of dynamic panel validity by checking if the estimated coefficient of the lagged dependent variables lies between the values obtained from pooled Ordinary Least Square (POLS) and Fixed Effect (FE) estimators. Our results established that, as seen in Table 5, the coefficient of the lagged dependent variables of the system GMM results lies between the values obtained from the POLS and FE estimators (FE = 0.232 < GMM=0.4765 < POLS = 0.636).