AN EMPIRICAL RETROSPECT OF THE CAUSAL EFFECT OF GOVERNMENT EDUCATION SPENDING ON GROWTH IN A NEOCLASSICAL GROWTH MODEL

There has been consensus on the notion that education results in economic prosperity and growth in many countries. This has resulted in a strong focus on education policy, with large investments and a lot of public debates concerning the subject. Various schools of thought have made differing suggestions about how government spending impacts economic growth over the years. The Keynesian view is that there is a positive relationship between government spending and economic growth, where the causal effect runs from government spending to economic growth. Conversely, the Neo-classical school asserts that the relationship between the two variables is negative. The topic, therefore, remains a debatable issue. The present study, therefore, tests the causal effect of education expenditure on economic growth in South Africa for the period 1994 to 2021, with the aid of the autoregressive distributed lag approach. Consistent with Keynesian theory, the study results confirm the positive impact of government spending on economic growth. A Granger causal relationship exists between government education expenditure and economic growth, indicating that over time, education expenditure positively impacts economic growth through human capital. This implies that investing (spending) in education is critical in promoting economic growth, especially in the long term.


Introduction
Across the globe, education is considered one of the key catalysts for economic growth. By investing in education, countries can elevate their human resources, thereby boosting economic growth (Suwandaru et al., 2021). Likewise, education receives the greatest share of the South African government spending (5% of GDP), with 21% of non-interest allocations set aside for basic and higher education (National Treasury, 2021). The allocations to the Ministry of Education were increased at an average annual rate of 3.3% from R28.5 billion in 2021/22 to R31.4 billion in 2024/25 (National Treasury, 2022. Total spending on education increased by R80 billion over five years, from R169 billion in 2009/10 to R249 billion in 2013/14 (Odhiambo, 2020).
After decreasing by 0.22% in 2016, public spending on education as a percentage of GDP increased by 15.03% in 2020 (World Bank, 2021). In 2019-2020, South Africa spent more than 20% of its budget on primary and tertiary education, and total education expenditure exceeded 6% of the gross domestic product (United Nations International Children's Emergency Fund, 2019). South Africa's public spending on education as a percentage of GDP increased from 6.5% in 2019 to 6.8% in 2020, an increase of 5.09%.
In his 2021 budget speech, Finance Minister Tito Mboweni announced that the government would cut spending on education and cultural functions over the next three years. However, despite the announcement by Finance Minister Tito Mboweni, the allocation to the education sector is increasing year by year (National Treasury, 2021). The 2021/2022 budget allocated approximately R408.2 billion for education. Government spending as a percentage of GDP was reported at 6. 84% and 18.42% in 2020, respectively (World Bank, 2021. Education is one of the most important factors contributing to countries" progress, welfare and level of economic and social development (Savrul and Tunc, 2021). It is a crucial factor for sustainable economic growth. Therefore, public expenditures on education are of great interest to both researchers and policymakers (Ziberi et al., 2022). As much as education is important for a country's economic growth, public spending on education is also important for improving education. Public spending on education is therefore expected to affect the country's economic growth.
There has been an increase in interest in Economics literature in studies examining the link between education spending and economic growth (Mercan & Sezer, 2014;Gheraia et al., 2021). Although there is a great deal of literature on this topic, previous studies have revealed mixed results, especially in countries with different environments and cultures. (Taasim, 2020). Furthermore, despite extensive research on this topic in other countries, little has been done in South Africa. Considering this background, this study aims to investigate how government spending on education affects economic growth in South Africa. Furthermore, to determine whether education spending and economic growth are associated in the long term, evaluate short-term dynamics among the variables studied, and examine causal relationships among the variables.

Literature review
The relationship between government spending and economic growth has attracted the attention of economists, policymakers, and politicians for many years, but the topic is still a debatable issue. As argued by Shkodra, et al., (2022), ""There exists a large body of literature on the impact of government spending on a country's economic growth. However, even though the topic has been investigated extensively, the results are generally contradictory."" The question is whether the impact of government size on economic growth is positive, negative, or negligible. As argued by Alqadi and Ismail (2019), "government spending and economic growth remain contentious issues among economists". Different schools of thought have come to different conclusions, where the majority confirm a positive impact of government spending on economic growth (Kimaro et al., 2017;Leshoro, 2017;Lee et al., 2019;Nyasha & Odhiambo, 2019;Olaoye, et al.,(2020). However, others have found a negative impact (Lupu et al., 2018;Onifade, et al., 2020).). There seems to be no study that reveals that government spending has no significant impact on economic growth.
Keynes views government spending as a ladder to economic growth, which encourages short and long-run economic growth (Ahuja & Pandit, 2020; Kgomo & Ratombo, 2022). According to Keynesian theory, government spending has a positive effect on economic growth. The Keynesian theory postulates that the more a country spends, the higher its economic growth will be as a result of expansionary fiscal policy (Riza and Wiriyanata, 2021). The assumption is that when government spending increases, production will follow, stimulating aggregate demand and thereby increasing GDP (Đukić, 2021).
As stated by Ahuja and Pandit (2020), in the Keynesian framework, it is government spending that regulates the rate of economic progression. This perspective overstates the significance of government expenditure and affirms the positive impact of public expenditure on GDP growth. Consistent with this theory, are the studies by Milhana & Nufile, (2019); Ahuja & Pandit (2020); Nuru (2021); Gheraia et al., (2021);Shkodra, et al., (2022).
Inconsistent with the Keynesian school of thought, neo-classical theorists suggest that the relationship between the two is negative. The neo-classical school are of the view that the expansion of government spending leads to the competition of (crowding out) the private sector by increasing domestic interest rates and increasing tax rates with distortionary effects on the allocation of resources. Advocates of this view are amongst others, Kouton (2018); Karaçor et al., (2017;Onifade et al., 2020;Nyasha & Odhiambo, 2019).
According to the Ricardian School of thought (the Ricardian Equivalence Hypothesis), the effect of government spending, whether financed by government debt or tax revenues, on economic growth is neutral. In other words, this relationship between government spending and economic growth does not exist. The main reason behind this neutral effect of government spending on economic growth, according to supporters of the Ricardian view, is consumer expectations about future tax increases. If consumers expect future tax increases, they will increase their savings by reducing current consumption, which in turn neutralizes the government spending multiplier mechanism (Alqadi & Ismail, 2019).
Wagner"s law, disagreeing with the Ricardian Equivalence hypothesis that the effect of government spending on economic growth is neutral, postulates that there is a correlation between government expenditure and economic growth, but economic growth causes government spending. Thus, Wagner"s Law assumes that economic growth is the cause of the increase in government spending.
Supporters of the Barro view also believe that there is a nonlinear impact of government spending on economic growth. According to this theory, expansion in government spending has a positive effect on economic growth up to a certain threshold, and then the impact will be negative beyond that threshold (Alqadi & Ismail, 2019;Maneejuk, & Yamaka, 2021;Villela, and Paredes, 2022). The study by Yakubu & Gunu (2022) is one of the studies that show that government educational expenditure has an insignificant effect on economic growth in both the long-run and short-term, implying that the nexus is nonlinear or neutral.
The relationship between government spending and economic growth has attracted the attention of economists, policymakers, and politicians for many years, but the topic is still a debatable issue. As argued by Shkodra, et al., (2022), ""There exists a large body of literature on the impact of government spending on a country's economic growth. However, even though the topic has been investigated extensively, the results are generally contradictory. Despite the theoretical grounds pointing to a positive relationship between government spending and economic growth, the extant research on this nexus is inclusive (Ahuja & Pandit, 2020).
The impact of government spending on economic growth depends on what the government spends money on and how well the institutional mechanism decides to manage expenditure (Shkodra et al., 2022). Education is regarded as one of the primary drivers of economic progress all around the world. Countries can improve their people resources by investing in education, which can accelerate economic growth (Suwandaru et al., 2021). There is no question that education is one of South Africa's top domestic priorities and the biggest long-term challenge it faces today. Besides, despite the vast literature on developing economies, there seems to be a dearth in the literature on the nexus between government expenditure on education and growth in South Africa. The study by Luthuli (2017) focuses on the impact of education expenditure on education attainment and not economic growth. Conversely, the study by Nkomo (2016) assessed the impact of education and health public spending on economic growth, not solely education spending. This necessitates the need to conduct a study on South Africa.

Theoretical framework
The neo-classical production function forms the basis of the model employed in this study. The model by Mallick et al. (2016) is one of the earlier theories which employs a modified version of the Cobb-Douglas production function to examine the link between government expenditure on education and economic growth. According to Mallick et al. (2016) the GDP function is expressed as follows: Where represents the total economic growth and is a proxy for public (government) expenditure on education. The government expenditure on education (a measure of education quantity) represents human capital formation which can make a skilled labour force.
It should be noted that various studies measured education quantity using multiple proxies. represents labour force, which is measured by the total, labour force; is capital, which is proxied by Gross Fixed Capital Formation.

The Empirical Model
The present study follows Malick et al. (2016) and Amaghionyeodiwe (2018) in employing the modified Cobb-Douglas function. Inconsistent with the two studies, the study substitute labour with government tax and public debt. The model is therefore specified as follows: The equation can be rewritten as follows: represents GDP (econoic) growth, is education expenditure. is gross fixed capital formation, which means capital.
represents the government tax incentives, and is the proxy for national debt.

A Priori Assumption
and 0, implying that GDP positively correlates to Education expenditure, GFCF and government tax incentives, while it is negatively related to the national debt.

Introduction
This study utilizes the autoregressive distributed lag (ARDL) bounds test approach to determine the long-run nexus and short-run dynamics between government expenditure on education and economic growth. Before the ARDL tests were conducted, stationarity tests were undertaken to check the presence of unit roots in the series. This study uses the Augmented Dickey-Fuller (ADF) and Phillips-Perron tests.
The Granger Causality test is used to determine the significant causal relationship between the variables. The null hypothesis presents that one of the variables in question does not causally affect the other variable in the linear analysis. If both variables do Granger cause (affect) one another; then this is bidirectional causality. However, if it is only one variable that Granger causes (affects) the other, then this is considered unidirectional causality.
Diagnostic tests are conducted diagnostic tests to confirm that there are no problems with residuals. This is to check whether the model is proficient or not. This study will test for heteroskedasticity and autocorrelation. Ramsey"s RESET (regression specification error test) will be conducted to check for misspecification of the functional form. To test for Structural Breaks, Cumulative Sum of Residuals (CUSUM) test and Cumulative Sum of Squares (CUSUMQ) test, as well as the Chow breakpoint tests are used.

Data Issues
This study employed annual time series data for the period spanning from 1994 to 2021, sourced from the South African Reserve Bank, the International Monetary Fund, and the World Development Bank Indicators. EViews software was used for analyzing the data as it is a good tool for time series analysis.

Stationarity tests Results
The Unit Root test o ensured that variables are integrated in the same order. It is an important phenomenon for a series to be tested for stationarity since this can influence its behaviour (Ruiters and Charteris, 2020). For this reason, variables were tested for unit root to avoid spurious results and to ensure that no second difference variables exist in our model, as this would violate the ARDL estimator. An augmented Dickey-Fuller test (ADF) was used to test the null hypothesis that a unit root exists in a time series sample. The assumption used in the test for stationarity is that the null hypothesis states that there is a unit root at whatever level of confidence. As such, Table  5.1 presents ADF results at levels after 1st difference and 2nd difference under the assumption of intercept (constant) only.
The findings of the Augmented Dickey-Fuller (ADF) test are shown in Table 5.1 below. The ADF test was conducted under the null hypothesis (H 0 ) that the series has a unit root (non-stationary) versus the alternative hypothesis (H 1 ) that the series is stationary. The ADF test statistics were compared with critical values at the 5% significance level. Accordingly, if the calculated ADF statistic is greater than the critical value at 5%, the null hypothesis that the series has a unit root is rejected and conclude that the series has no root test; therefore, it is stationary, and vice versa.
The results presented in table 5.1 below reveal that all the variables are non-stationary at a level for Augmented Dickey-Fuller (ADF), as the calculated t-statistics in absolute terms are less than the critical values at the 5% level of significance, respectively. However, all variables become stationary after 1 st differencing under Augmented Dickey-Fuller (ADF) test, and as a result, the Auto regressive distributive lag (ARDL) model was employed. The results also reveal that GDP is stationary at a level under the Philip-Perron unit root test, while all other variables become stationary after the 1 st differencing under the Philip-Perron test.

Order selection criterion
Identifying the long-run structure and formulating a long-run analysis required a lag order selection with the ARDL method. Table 5.2 presents the requisite lag order selection criterion conforming to the selected method applicable to this discipline. According to the results presented in Table 5.2, by considering the lowest value with an asterisk (*), it is evident that the Akaike Information Criterion value of -19.44198* is less than -17.73147* of the Schwarz Information Criterion and -18.955517* of Hannan-Quinn Criterion. Therefore, this value indicates the best optimal lag for the model as lag 2. Therefore, a chosen criterion should minimise the asterisk figure to determine the best optimal lag. The rule states that if the calculated F-statistic is lower than the critical value for the lower bound I (0), we fail to reject the null hypothesis that there is no long-run relationship and conclude that there is no cointegration. However, if the F-statistic is greater than the critical value for the upper bound I (1). In that case the null hypothesis is rejected there is no long-run relationship between the dependent variable and its explanatory variables and conclude that there is cointegration. The F-statistic value (8,454950) is greater than the I (1) critical value bound (4.01). Consequently, the null hypothesis is rejected, there is no equilibrating (long-run) relationship and conclude that there is long-run relationship between the dependent variable and its explanatory variables under review.
In interpreting the ARDL long-run results, the signs of the coefficients are reversed, and they explain short-run causal effects. As the p-values of education expenditure (0.0030) are less than 0.05 at the 5% level of significance and gross fixed capital formation (0,0989) is less than 0.10 at the 10% level of significance, it is therefore, concluded that education expenditure and gross fixed capital formation has a short-run causal effect on the gross domestic product (economic growth). However, there is no causal effect from the labour force and poverty to economic growth in the short run. Therefore, it can be concluded that, in the short-run, education expenditure and fixed capital formation have a positive impact on economic growth. As expected, the ARDL error correction regression results reveal that the error correction term (ECT), represented as CointEq (-1), has a correct negative sign with an associated coefficient estimate of −0.036102. This implies that about 4% (3.6%) of any movements into disequilibrium are corrected within a year. Furthermore, the p-value of 0.0000, which implies perfect significance, also supports a highly significant long-run causal relationship between the regressand and its regressors.

.3 Granger Causality Test
The causality test aims to check how the variables react to each other and the direction of causality between them (Waseem, 2015). Table 5.6 provides a Causality test between the dependent variable and independent variables. The Granger Causality results reveal that education expenditure has a causal effect on economic growth (GDP) as depicted by the p-value of 0.0222, which is less than 0.05 at the 5% significance level. Thus, the null hypothesis that education does not granger cause economic growth is rejected, against the alternative that it does granger cause education. However, GDP does not lead to education expenditure. In the case of GDP and labour, causality runs from GDP to labour, not the other way round. This is depicted by the p-value (0.8539), which is insignificant implying that the labour force does not granger cause GDP, but economic (GDP) growth leads labour force. Regarding gross fixed capital formation and poverty, neither gross fixed capital formation nor poverty granger causes GDP. The F-statistic p-value of 0.1005, which is greater than 0.05, implies failure to reject the null hypothesis that there is no serial correlation of any order up to p. It is therefore concluded that there is no serial autocorrelation. The Breusch-Pagan LM test provided a formal test for heteroscedasticity, which tested for the violation of assumption 5, which indicated that the error term should have a constant variance. The above table indicates that the null hypothesis of no evidence of heteroscedasticity cannot be rejected because Obs*R-squared is greater than the Chi-Square values. Thus, it can be noted that there is significant evidence of homoscedasticity. Since the pvalue of the F-statistic (0.1303) is greater than 0.05 at the 5% level of significance, we, therefore, fail to reject the null hypothesis of homoscedasticity (no heteroscedasticity). Thus, it is concluded that the residuals are homoscedastic at the 5% significance level.

 Heteroscedasticity Test
 Residual Normality Test The Jarque-Bera test is used to ascertain the normality of residuals within a model. Regarding the results, the probability value of the Jarque-Bera is 0.358386 and is non-significant at the 5% levels of significance. Therefore, we fail to reject the null hypothesis of the normal distribution and conclude that the residuals are normally distributed. The p-value for our F-statistic is 0.5261, which is greater than 0,05 at the 5% significance level. We, therefore, fail to reject the null hypothesis that the model does not suffer from omitted variables and conclude that the model is correctly specified.

 Cumulative Sum of Residuals (CUSUM) test and Cumulative Sum of Squares (CUSUMQ) test
Finally, the CUSUM and CUSUMSQ plots to check the stability of the long-run parameters and the shortrun movements for the ARDL-Error Correction Model are given in Figures 5.2 and 5.3, respectively. If the plots of the CUSUM and CUSUMSQ statistics stay within the critical bounds of a five per cent level of significance, the null hypothesis that all coefficients in the given regression are stable cannot be rejected. Examination of plots in Figures 5.2 and 5.3 shows that CUSUM and CUSUMSQ statistics are well within the 5% critical bounds, implying that short-run and long-run coefficients in the ARDL-Error Correction Model are stable or the residual variance is stable.  The rule of chow breakpoint states that if the value of the F-statistic is greater than 0.05 at the 5% level of significance level, we fail to reject the null hypothesis of no break. In table 5.12, it is evident that the F-statistic (0.0988) is insignificant at the 5% level of significance. Therefore we fail to reject the null hypothesis, meaning there is no structural break at the chosen period. This is supported by the results of the CUSUMQ test, which reveal that the model coefficients are stable, or the residual variance is stable.

Conclusion and Policy Implications
Consistent with Amaghionyeodiwe (2019), the study findings reveal that government spending on education and economic growth in South Africa is positively and significantly related. Long-term Granger causality exists between government expenditure on education and economic growth, indicating that in the long run, government educational expenditure, through its impact on human capital, significantly and positively influences economic growth. This demonstrates that any investment (spending) on education is critical in significantly promoting economic growth, especially in the long term. The results are logical and agree with the Keynesian theory, which postulates that government spending has a positive effect on economic growth. Keynesian.
These results are, however, inconsistent with some previous empirical results. For example, Vijesandran and Vinayagathasan (2014) found a negative long-run association between education and the economic in Sri Lanka. Kouton (2018) also found a negative link between the two variables in Côte d"Ivoire.
The implication is that, as the government invests more funds in education, this tends to boost human capital, which is translated into economic growth in the long run. Therefore, the policy suggestion is that government education expenditure should increase. However, this expenditure must be of quality so it may result in more inclusive growth. In other words, the rate of pupils enrolled in primary education should be high so that not only a high economic trajectory will be improved, but also in more inclusive growth. As argued by Kouton (2018: p14), ""what is also important is the efficiency with which education expenditure is translated in education outcomes through better ratios of education".
Finally, policies on education expenditure should be reviewed and updated, which will be advantageous for wealth creation. This would mean that the role of the government would no longer just be to invest massively in education but to set up the economic environment to increase the benefits of education for economic development.