An Empirical Analysis of the Impact of Public Debt Service Costs on Social Expenditure in South Africa

Abstract

This study investigates the impact of public debt service costs on social expenditure in South Africa, focusing on three categories of social expenditure, namely, government education expenditure, government health expenditure and government social protection expenditure. It addresses the challenge of financing debt service costs while maintaining social expenditure. The study employed annual time series data from 1994 to 2024 and the autoregressive distributed lag (ARDL) technique to examine the effect of public debt service costs on social expenditure. The results reveal that debt service exhibits a negative and statistically significant impact across all three categories of social expenditure under consideration in the long run and short run in South Africa. Moreover, the results reveal that public debt has a negative relationship with all three categories of social expenditure. The exchange rate and revenue were found to have a positive relationship with all three categories of social expenditure under consideration. Urban population was found to have a positive relationship with government education expenditure and social protection expenditure. These findings underscore the need to focus on reducing the fiscal pressure stemming from increasing debt service costs, while upholding social expenditure. It is recommended that policymakers focus on debt stabilisation and reduction, thereby easing the crowding-out of social expenditure.

1. Introduction

The level of public debt has risen sharply in recent times. According to Tiftik et al. (2025), global debt grew by approximately $7 trillion in 2024; however, this growth was smaller compared to the growth in 2023. Moreover, the global debt-to-GDP ratio has risen by over 1.5% since 2020, reaching approximately 328% of GDP. Most developing and emerging economies have faced increasing debt since the global financial crisis of 2008 due to weak financial positions (Kose et al. 2020). In addition, the COVID-19 pandemic also intensified public debt pressures for developing and emerging economies. On average, developing and emerging economies experienced a rise in their public debt-to-GDP ratio, whereas advanced economies recorded a decline. In 2023, average public debt declined by 3% of GDP to 103% of GDP in advanced economies; this record was slightly below the pre-COVID-19-pandemic conditions (International Monetary Fund 2024). In contrast to advanced economies, public debt in emerging economies increased by 2% of GDP to 57% of GDP in 2023. Similarly, in developing countries, public debt reached a new record of 50% of GDP in 2023, exceeding pre-COVID-19-pandemic levels (International Monetary Fund 2024).

Public debt in South Africa has continued to increase throughout the years. The statistics from the South African Reserve Bank reveal that the total public debt has risen above the pre-COVID-19-pandemic levels. In other words, the total public debt increased from 56.1% of GDP in 2019 to 68.9% in 2020, 68.8% in 2021, 70.7% in 2022, 73.1% in 2023 and 77.1% in 2024 (South African Reserve Bank 2025a). In developing and emerging countries like South Africa, the government uses debt as one of its instruments to finance development and growth. However, high and unsustainable levels of public debt may hinder development and growth (Dabrowski 2014; Ani et al. 2024). Increasing public debt in South Africa has sparked concern among economic participants about the direction of the future debt trajectory. High public debt or borrowing without adequate planning can lead to high debt service, which has several negative consequences for the economy (Joy and Panda 2020). Mbewe (2024) contends that as economic issues arise, governments rely more on borrowing. As a result, public debt continues to rise, which constrains the government’s ability to allocate expenditures to other sectors, such as the social sector.

Social expenditure can be described as the financial resource allocated by governments to address social needs. It comprises all public spending on social protection, education and health services (Karaalp-Orhan et al. 2024). Accumulated public debt and the associated debt service cost constrain the government’s budget, potentially crowding out expenditure on social programmes. This trade-off is emphasised in the debt overhang theory and debt crowding-out hypothesis.

South Africa is confronted with rising levels of debt; over the past decade, the country has had one of the largest increases in public debt-to-GDP. This build-up of debt has resulted in a rapid rise in debt service costs, which absorb over 20% of government revenue (Republic of South Africa National Treasury 2023; Republic of South Africa National Treasury 2024). This constrains the government’s capacity to channel funds to other essential sectors. Social sectors such as health, education and other critical social services have been adversely affected by rising debt service costs. With high levels of debt service costs, governments are pushed to cut spending on essential services. According to the Republic of South Africa National Treasury (2023), for every 5 rands collected in revenues, the government pays 1 rand to lenders instead of funding social sectors such as health, education and other critical social services. The literature has shown that high debt and debt service costs reduce government expenditure on social sectors due to funds’ reallocation to service debts (Gurowa et al. 2023; Miningou 2023).

South African fiscal aspects are under pressure, as public debt levels increase, and debt service costs take a growing share of the national budget. Over the past decade, the cost of servicing debt has grown at a yearly average rate of approximately 13.1%, reaching R385.6 billion in the financial year 2024/25 (South African Reserve Bank 2025b). This exhibits an increased shift in funds to finance debt, thereby crowding out expenditure on key priority areas such as economic development initiatives, social services and infrastructure. This raises concerns about the impact of debt service costs on social expenditure, which is aimed at alleviating poverty and improving the standard of living of the people. Several studies have investigated the nexus between debt service costs and government expenditure in developing countries (Gurowa et al. 2023; Mbewe 2024). Despite growing debate around the nexus between debt service costs and government expenditure, empirical evidence on the influence of debt service costs on social expenditure in South Africa is scanty. A recent study has investigated the nexus between debt service cost and private capital formation in South Africa (Zhou 2021). To date, little research has been carried out to assess whether growing debt service costs are crowding out social expenditure. Given this background, the aim of this study is to empirically investigate the impact of debt service costs on social expenditure in South Africa. Given the South African fiscal aspects, this study is critical to policymakers and academics/researchers looking to formulate and develop policies that can improve social expenditure, thereby alleviating poverty and improving the standard of living of people.

The rest of this paper is organised as follows: Section 2 examines theoretical and empirical literature; Section 3 provides the methodology; Section 4 presents the empirical results; and Section 5 offers the conclusion to the study and policy recommendations.

2. Literature Review

This section is divided into two subsections, namely, the theoretical background and the literature review. The theoretical background section reviews the theory/s that underpin(s) the relationship between public debt service costs and expenditures, while the literature review section reviews existing empirical evidence on public debt and expenditures.

2.1. Theoretical Background

There are several theories developed to explain the relationship between debt and expenditure. However, this study reviews the debt overhang theory and debt crowding-out hypothesis. Debt overhang theory implies that, to cut down the accumulated debt, governments must increase the tax rate and, consequently, this discourages investment. Accumulated debt negatively affects government investment, thereby diverting government expenditure towards debt servicing. Loko et al. (2003) highlighted that to pay debt service, the government often reallocates resources meant for social sectors such as health, education and social protection. Therefore, debt relief leads to increased government investment in the social sector. With regard to the crowding-out hypothesis, high debt service costs increase the budget deficit, thereby cutting down public savings/revenue. This, in turn, affects investments and economic growth, since the reallocation of public savings/revenue to fund debt reduces social sector expenditure (Rahim et al. 2023). It can be said that the debt overhang theory and debt crowding-out hypothesis assume a trade-off between debt service costs and government expenditure.

2.2. Empirical Literature Review

The nexus between public debt service costs and expenditures has been widely investigated, with a particular focus on different components of public expenditure. Using a panel approach, Mbewe (2024) explored the link between public debt service and education expenditure using the pooled mean group (PMG) estimator for the period from 1995 to 2022. The study revealed that high public debt service costs have a negative impact on education expenditure in the long run. This aligns with the debt overhang theory, with the view that high debt servicing costs reduce government investment. A study conducted in Nigeria also shows that public debt service costs exert adverse effects on social and community sector expenditure (Gurowa et al. 2023). Similar findings were reported by Shabbir and Yasin (2015) for Asian countries. The crowding-out effect of public debt servicing on government expenditure has been documented in many studies. In Nigeria, for example, debt servicing was found to crowd out capital expenditure and recurrent expenditure (Iliyasu and Gambo 2022). While studies have found that debt service costs crowd out education expenditure, social expenditure and total government expenditure, the magnitude of the effect varies across institutional settings. The magnitude was found to be high in the study by Shabbir and Yasin (2015).

Tashevska et al. (2020) employed a panel regression to examine whether social protection expenditure crowded out expenditures on other purposes in the European Union for the period from 1995 to 2018. The findings show that there was a crowding-out of education and infrastructure expenditure. Furthermore, increasing public debt and deficit financing were found to have a negative impact on education, infrastructure and important public services.

In contrast to studies on public debt, Nkala and Tsegaye (2017) investigated the link between household debt and consumption spending in South Africa using the vector error correction model (VECM) for the period from 1994 to 2013. The study found that household debt positively influenced consumption spending, suggesting that household debt increases the household income available to consume. However, debt servicing was found to have a negative impact on consumption spending, confirming the crowding-out effect. This study provides important insights into the relationship between household and consumption spending.

Said and Sani (2020) explored the nexus between health expenditure and the public debt burden in Sub-Saharan nations using the generalised method of moments (GMM) approach for the period from 2000 to 2014. The GMM findings show that debt burden has a negative impact on health expenditure in Sub-Saharan nations. On the contrary, panel evidence from 42 African nations has shown that public debt servicing can have both negative and positive effects on health expenditure depending on whether the nation is classified as low-income, low-middle-income or upper-middle-income (Fosu et al. 2025).

A recent study by Gbenga (2025) investigated the link between debt servicing and government expenditure in Nigeria using the ordinary least squares (OLS) approach for the period from 2000 to 2023. The findings of the study indicate that, in the long run, debt servicing has a significant positive impact on government expenditure; however, external debt has a negative impact on government expenditure. The findings of this study oppose the prior findings of Gurowa et al. (2023) and Iliyasu and Gambo (2022). This shows that there are varying results on the impact of debt servicing on government expenditure. These differences may stem from variations in fiscal capacity, macroeconomic conditions, and institutional quality across countries.

Kapindula and Kaliba (2022) examined the influence of external debt service on infrastructure expenditure in Zambia using the ARDL model for the period from 1970 to 2014. The results exhibit that external debt service has a negative influence on infrastructure expenditure in the long run. This aligns with the debt overhang theory, with the view that high debt limits government investments. This indicates that, regardless of whether debt service is external or domestic, it affects different categories of expenditure.

Using a fully modified ordinary least squares approach for BRICS nations, Joy and Panda (2020) found that debt servicing has a negative impact on gross domestic saving and gross capital formation. This demonstrates that payment for debt service has a detrimental effect on gross capital saving and gross capital formation. Therefore, this affects the government’s ability to spend through investment. Picarelli et al. (2019) reported that public debt reduces public investment in European Union countries. Similarly, Zhou (2021) found that debt service costs have a negative impact on private capital formation in South Africa. Moreover, domestic public debt was found to crowd out private capital formation, while external debt crowds in private capital formation.

Using the bootstrap autoregressive distributed lag (ARDL) approach to investigate the influence of public debt on public spending in Nigeria, Abu et al. (2022) found that public debt has a nonlinear effect on public spending and that early phases of increasing public debt increase public spending. Nonetheless, as public debt increases beyond a certain level, public spending diminishes at a later stage.

Despite the contribution of the above empirical studies examining the relationship between debt service costs and social expenditure, a notable gap persists in the literature, especially within the South African context. Existing studies in South Africa focus on the relationship between public debt service and investment, as well as household debt and consumption spending. For instance, Zhou (2021) investigated the impact of public debt and debt servicing on private capital formation. Nkala and Tsegaye (2017) focused on the nexus between household debt and consumption spending. These studies do not consider the possible effect of debt service costs on social expenditure. Studies that have examined these effects on social expenditure have generally focused on either education expenditure or health expenditure (Mbewe 2024; Said and Sani 2020). Therefore, the current study distinguishes itself by considering different categories of social expenditure. Moreover, conventional linear methods, such as the fully modified ordinary least squares or vector error correction model (VECM), employed by Zhou (2021) and Nkala and Tsegaye (2017), typically suffer from weaknesses such as strict integration requirements, small sample size limitations, and endogeneity issues. Therefore, employing these techniques may produce biassed estimates. Given these limitations, this study seeks to contribute to the literature by investigating the impact of public debt service costs on social expenditure in South Africa using the ARDL approach. It is argued in the literature that the ARDL approach produces unbiased estimates and valid t-values, regardless of the endogeneity of some regressors (Harris and Sollis 2003).

3. Methods

3.1. Model Specification

This study follows and modifies the empirical model by Gbenga (2025), who investigated the nexus between debt servicing and public expenditure in Nigeria. In the study, public expenditure was modelled as a function of debt servicing, external debt and exchange rate. The model is mathematically expressed as follows:

$$GEX=f(DSV,EXD,EXC)$$

(1)

where GEX, DSV, EXD and EXC represent government public expenditure, debt servicing, external debt and the exchange rate, respectively. The current study seeks to examine the impact of debt service costs on social expenditure; therefore, Equation (1) is modified. In the interest of the current study, Equation (1) is modified as follows: government public expenditure is replaced by social expenditure, and external debt is replaced by public debt. In addition to debt service costs, social expenditure can be influenced by other economic variables, and, as a result, the government revenue and urban population are added to the model to enhance the empirical analysis. The choice of the variables in this study is grounded on the empirical literature and theoretical framework that emphasise factors affecting government expenditure. For instance, debt service costs and public debt are used based on the debt overhang theory and the debt crowding-out hypothesis. Moreover, the use of the exchange rate is supported by the empirical literature (Gbenga 2025; Kapindula and Kaliba 2022). Moreover, the exchange rate movements affect government expenditure through the depreciation or appreciation of local currency. For instance, a weaker local currency makes imports expensive, thus increasing government and vice versa. It was reported that government revenue positively affects government expenditure (Iliyasu and Gambo 2022; Shabbir and Yasin 2015). Therefore, the current study added revenue to the model. As more people stay in urban areas or a country becomes more urbanised, there is a demand for government expenditure (Fosu et al. 2025). Therefore, the current study incorporates the urban population into the model. In this study, social expenditure is captured through education expenditure, social protection expenditure and health expenditure. This is attributable to the fact that these categories constitute approximately 20%, 15%, and 12% of total government expenditure, respectively (Statistics South Africa 2025). The modified model is mathematically expressed as follows:

$$SE=f(DSV,PDBT,EXC,REV,URBP)$$

(2)

where SE, PDBT, EXC, REV and URBP represent social expenditure, public debt, exchange rate, revenue and urban population. Equation (2) is transformed from its mathematical expression into an econometrics model:

$$LnSE=a+{\beta }_{1}Ln{DSV}_t+{\beta }_2{LnPDBDT}_t+{\beta }_3{LnEXC}_t+{\beta }_4{LnREV}_t+{\beta }_5{LnURBP}_t+{\epsilon }_t$$

(3)

where all variables are stated as previously and taking a logarithm to alleviate the problem of heteroskedasticity, $a$ is the constant variable of the regression, ${\beta }_1$ to ${\beta }_5$ are parameters to be estimated and ${\epsilon }_t$ is the error term.

3.2. Data

Annual time series data from 1994 to 2024 is employed in this study. The sample period was chosen based on South Africa’s major institutional, political and fiscal shift after the beginning of democracy in 1994. The fiscal policy regime in the chosen period is characterised by expanded social service delivery, inclusive governance and reconstruction of government expenditure to address poverty and inequality. However, the study is constrained by a relatively small sample of 31 observations. While small samples may reduce statistical estimation power, the technique employed in this study is well-suited for small sample sizes (

Table 1. Data measurement and sources.

	Variables	Measurement	Sources
Social expenditure	Government education expenditure (GEE)	R millions	South African Reserve Bank
	Government health expenditure (GHE)
	Government social protection expenditure (GSPE)
Explanatory	Debt service (DSV)	Total debt service (% of gross national income (GNI))	World Development Indicators (World Bank)
Controls	Urban population (URBP)	Total number of urban populations
	Exchange rate (EXC)	Official exchange rate (local currency per US$, period average)
	Revenue (REV)	National government revenue as % of GDP	South African Reserve Bank
	Public debt (PDBT)	Total national government debt as % of GDP

Source: Author’s own.

).

3.3. ARDL Technique

This study adopts the autoregressive distributed lag (ARDL) technique to examine the short-run and long-run effects of public debt service costs on social expenditure in South Africa. The adoption of the ARDL technique offers several advantages. For instance, the technique accommodates variables that are integrated of order I(0) and I(1). Moreover, unlike the VECM technique, which requires a large sample size, it can handle the analysis of a small sample (Pesaran et al. 2001). This makes it an appropriate technique for the present study, given the relatively short time span from 1994 to 2024. In addition, the ARDL technique can separate between dependent and independent variables, thereby eliminating the issue that may arise because of the presence of autocorrelation and endogeneity (Rehman and Afzal 2003). The ARDL model yields the following specification, symbolising the standard OLS model in Equation (3).

$$\begin{array}{c}{LnSE}_t=C_0+{\displaystyle \sum _{i=1}^p}{\beta }_1∆{LnSE}_{t-1}+{\displaystyle \sum _{i=1}^p}{\beta }_2∆{LnDSV}_{t-1}+{\displaystyle \sum _{i=1}^p}{\beta }_3∆{LnPDBT}_{t-1}+{\displaystyle \sum _{i=1}^p}{\beta }_4∆{LnEXC}_{t-1}+{\displaystyle \sum _{i=1}^p}{\beta }_5∆{LnREV}_{t-1}\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{\displaystyle \sum _{i=1}^p}{\beta }_6{∆LnUP}_{t-1}+b_1{LnSE}_{t-1}+b_2{LnDSV}_{t-1}+b_3{LnPDBT}_{t-1}+b_4{LnEXC}_{t-1}+b_5{LnREV}_{t-1}\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+b_6{LnUP}_{t-1}+{\epsilon }_t\hfill \end{array}$$

(4)

All explanatory and explained variables are described in the previous equations. $C_0$ is constant, p represents the lag length for the unrestricted error correction model (UECM), $∆$ is the first differencing operator, ${\beta }_1$ to ${\beta }_6$ represent the short-run coefficients and $b_1$ to $b_1$ represent the long-run coefficients.

Before the estimation of the ARDL model, several tests were conducted. The first test conducted is a descriptive statistics test to determine the underlying features of the dataset. The descriptive statistics provide insight into the dispersion, central tendencies and distribution properties of the individual variable considered in this study. In addition, descriptive statistics use the quantitative measurements such as the mean, kurtosis, skewness and standard deviation to summarise the data.

The second test conducted is the unit root test to identify the order of integration of the variables. To comply with the properties of the ADRL model, it is important to conduct a unit root to ensure the suitability of the ARDL model. The ARDL model requires variables to be integrated to the order of zero, I(0), or order of one, I(1), but not order of two, I(2). This study uses the augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) unit root tests for robustness. The null hypotheses of these state that the variable has a unit root and the alternative hypothesis states that the variable is stationary. After identifying whether the variables are integrated to the order of zero or order of one, the bound cointegration test is employed within the ARDL framework. The estimation of the model should be conducted based on the number of optimal lags suggested by the information criteria. In this study, the optimal lag length was selected using the Akaike information criterion (AIC) automatic selection.

After the estimation of the ARDL model, several stability and diagnostic tests were performed to analyse the dependability and goodness of fit of the estimated model. The residual diagnostic tests performed in the study include the heteroscedasticity test to identify whether the residuals show a constant variance, the Breusch–Godfrey LM test to assess serial correlation and the normality test to analyse whether the residuals are normally distributed. Therefore, failure to satisfy the assumption of homoscedasticity, no serial correlation and normality may lead to incorrect conclusions about the estimated model coefficient. The Ramsey RESET test was conducted to assess the robustness of the model specification.

4. Result Analysis and Interpretation

The purpose of this study is to investigate the impact of public debt service costs on social expenditure using the ARDL technique. This section provides empirical results and analysis. Studying the inherent features of data is essential in econometric analysis to validate the chosen model. Therefore,

Table 2. Descriptive statistics.

	GEE	GHE	GSPE	DSV	PDBT	EXC	REV	URBP
Mean	190,029.5	109,889.5	138,241.7	3.7183	43.6774	9.7063	21.9225	31,978,102
Median	143,733	82,045	106,910	2.9303	42.7	8.2612	21.8	32,205,804
Maximum	485,792	275,677	369,601	8.3698	77.1	18.4502	24.9	40,767,329
Minimum	29,756	14,475	15,781	1.4137	24	3.5507	19.4	23,359,070
Standard deviation	144,656.7	88,486.74	115,161.9	2.0809	14.7628	4.3954	1.6117	5,393,090
Skewness	0.6173	0.5714	0.7471	0.8614	0.8654	0.5359	0.0709	−0.0292
Kurtosis	2.0332	1.9089	2.3086	2.5021	2.8594	2.1171	1.8022	1.7596

Source: Author’s own estimation obtained from EViews 14.

exhibits the descriptive statistics of the variables under consideration.

Table 2 shows that government expenditure (GEE) in South Africa during the period under consideration has recorded a mean value of 190,029.5 and a standard deviation of 144,656.7, demonstrating high variability and that the data is widely spread out from the mean. The gap between the mean (190,029.5) and median (143,733) implies a positively skewed distribution. A positive skewness (0.6173) shows sporadic events where government education expenditure was higher than average. Similarly, government health expenditure and government social protection expenditure display high standard deviation, demonstrating that the data is widely spread out from the mean. Debt service (DSV) displays a mean of 3.7183 and standard deviation of 2.0809, demonstrating that the data deviates from the mean. Moreover, it has a kurtosis of 2.5021, indicating that the distribution is platykurtic and the data is lighter-tailed than the normal distribution.

Public debt (PDBT) has a mean of 43.6774 and a standard deviation of 14.7628, suggesting the dispersion of the data. A positive skewness (0.8654) shows occasional periods where public debt was greater than average. This indicates the period of high levels of indebtedness. Since the early 2000s, public debt has increased rapidly due to weak economic growth, the financial failure of government-owned enterprises and structural expenditure burdens. The exchange rate (EXC) shows a mean of 9.7063 with a standard deviation of 4.3954. Revenue (REV) exhibits a mean of 21.9225 and a standard deviation of 1.6117, indicating the variability of the data. The urban population (URBP) recorded a mean value of 31,978,102 and skewness of −0.0292, indicating that the distribution is skewed to the left and has a long left tail.

This study employs the augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) unit root tests. The ADF test was pioneered by Dickey and Fuller (1979) and is extensively employed to examine the presence of a unit root in time series. The ADF test is the extension of the standard Dickey–Fuller test, which incorporates the additional terms to account for potential trends and serial correlation in time series data (Guo 2023). The PP test was pioneered by Perron and Phillips (1988) to assess the presence of a unit root in time series. The PP test can be employed in time series containing heterogeneity and serial correlation. Therefore,

Table 3. Unit root test results.

		At Level		1st Difference		Order of Integration
Variables	Tests	t-Stat	p-Value	t-Stat	p-Value
GEE	ADF	−0.7287	0.9614	−5.405	0.0007 ***	I(1)
	PP	−0.7247	0.9617	−5.4027	0.0007 ***	I(1)
GHE	ADF	−1.7953	0.6799	−4.4295	0.0079 ***	I(1)
	PP	−2.0007	0.5774	−3.6922	0.0391 **	I(1)
GSPE	ADF	1.6209	1.0000	−5.3617	0.0009 ***	I(1)
	PP	−1.5472	0.7897	−13.0925	0.0000 ***	I(1)
DSV	ADF	−3.2469	0.0947 *	N/A	N/A	I(0)
	PP	−3.3901	0.0717 *	N/A	N/A	I(0)
PDBT	ADF	−0.3608	0.9846	−4.7163	0.0039 ***	I(1)
	PP	−0.4273	0.9816	−4.6897	0.0041 ***	I(1)
EXC	ADF	−1.8060	0.6765	−3.4551	0.0667 *	I(1)
	PP	−1.8810	0.6392	−4.9635	0.0021 ***	I(1)
REV	ADF	−4.1925	0.0139 **	N/A	N/A	I(0)
	PP	−4.9671	0.0022 ***	N/A	N/A	I(0)
URBP	ADF	−2.9372	0.4246	−4.5666	0.0057 ***	I(1)
	PP	−2.4913	0.3297	−5.7166	0.0003 ***	I(1)

Note: ***, ** and * represent the level of significance at 1%, 5% and 10%. Source: Author’s own estimation obtained from EViews 14.

presents the unit root test results.

Table 3 shows that the variables under consideration reveal a mixed order of integration. The ADF and PP test results show that government education expenditure (GEE), government health expenditure, government social protection expenditure, public debt (PDBT), urban population (URBP) and exchange rate (EXC) become stationary after first differencing and are, therefore, categorised integrated of order one, I(1). However, debt service cost (DSV) and revenue (REV) are stationary at levels, suggesting that they are integrated to the order of zero, I(0). The unit root test results fulfil a main prerequisite for employing an ARDL modelling framework. As a result, the ARDL bound test is conducted to establish the long-run relationship between the variables under consideration.

Table 4. The bound cointegration test.

Model	f-Stat	Significance Level	I(0) Critical Limit	I(1) Critical Limit
LnGEE = LnDSV + LnPDBT + LnEXC + LnREV + LnURBP	10.7489 **	1%	2.820	4.210
		5%	2.140	3.340
LnGHE = LnDSV + LnPDBT + LnEXC + LnREV + LnURBP	3.9248 **	10%	1.810	2.930
LnGSPE = LnDSV + LnPDBT + LnEXC + LnREV + LnURBP	11.1073 ***

Note: *** and ** represent a level of significance at 1% and 5%. Source: Author’s own estimation obtained from EViews 14.

shows that the bound test’s f-stat for the government education expenditure model is 10.7489, which is greater than the upper bound critical limit at the 1% level, suggesting a cointegrated long-run relationship between government education expenditure and the explanatory variables, including debt service cost, public debt, exchange rate, revenue and urban population. Moreover, the bound test’s f-statistic for the government health expenditure model is 3.9248, which is greater than the upper bound critical limit at the 5% level of significance, suggesting a cointegrated long-run relationship between government health expenditure and the explanatory variables, including debt service cost, public debt, exchange rate, revenue and urban population. Similarly, the bound test’s f-statistic for government special protection expenditure suggests that there is a cointegrated long-run relationship between government social protection expenditure and the explanatory variables, including debt service, public debt, exchange rate, revenue and urban population. The confirmation of a cointegrated long-run relationship suggests that the long-run and short-run effects can be established using the ARDL model. Therefore, the results of the long-run and short-run ARDL model are presented in

Table 5. ARDL results.

Variables	LnGEE	LnGHE	LnGSPE
LnDSV	−0.2757 (0.0079) ***	−0.2519 (0.0852) *	−0.8803 (0.0351) **
LnPDBT	−0.0095 (0.8631)	−0.3384 (0.1380)	−1.1404 (0.0577) *
LnEXC	1.3828 (0.0151) **	0.6197 (0.1753)	0.8139 (0.0418) **
LnREV	2.4427 (0.0120) **	1.1695 (0.0817) *	3.3741 (0.0383) **
LnURBP	−0.2347 (0.0315) **	0.1279 (0.4362)	3.1664 (0.0275) **
f-stat	7.4667 (0.0611) *	8.4669 (0.0514) *	16.6953 (0.0579) *

Note: ***, ** and * represent the level of significance at 1%, 5% and 10%. The value inside the parentheses is the corresponding probability value for the t-statistic. Source: Author’s own estimation obtained from EViews 14.

and

Table 6. Short-run ARDL model results.

Variables	LnGEE	LnGHE	LnGSPE
DLn(DSV)	−0.0725 (0.0004) ***	−0.3029 (0.0003) ***	−0.2059 (0.0000) ***
DLn(PDBT)	0.1407 (0.0077) ***	1.5090 (0.0000) ***	0.2059 (0.0000) ***
DLn(EXC)	0.8639 (0.0000) ***	0.1806 (0.0016) ***	0.3315 (0.0011) ***
DLn(REV)	1.0653 (0.0001) ***	1.1494 (0.0001) ***	1.4255 (0.0000) ***
DLn(URBP)	15.4262 (0.0000) ***	1.0296 (0.4650)	−2.9363 (0.0000) ***
ECM	−0.5611 (0.0000) ***	−0.4625 (0.0000) ***	−0.5351 (0.0000) ***

Note: *** represents the level of significance at 1%. The value inside the parentheses is the corresponding probability value for the t-statistic. Source: Author’s own estimation obtained from EViews 14.

.

The long-run results from the regression analysis are presented in Table 5. The regression estimates corresponding to each measure of social expenditure analysed in this study are reported below.

• Government education expenditure

The results presented in Table 5 reveal that debt service cost displays a negative relationship with government education expenditure in the long run. This implies that a 1% increase in debt service cost would result in an approximately 0.27% decrease in government education expenditure in South Africa, ceteris paribus. This significant negative impact of debt service costs on government education expenditure indicates that high debt service costs limit fiscal funds available for education. In practical terms, this suggests that when government directs a larger share of fiscal budget to fund debt service, it limits the government’s ability to fund development programmes such as social sector expenditure. Public debt is found to have a negative and insignificant impact on government education expenditure in the long run. The negative coefficient suggests that a 1% increase in public debt would lead to a 0.0095% decrease in government education expenditure in the long run, ceteris paribus. However, the effect is statistically insignificant within the sample period. Interestingly, the exchange rate is found to exert a significant positive influence on government education expenditure in the long run. This implies that a 1% increase in the exchange rate would lead to an approximately 1.38% increase in government education expenditure in the long run, ceteris paribus. This could be attributed to efforts to offset the inflationary effects of currency depreciation on the social sector, thereby shielding vulnerable citizens from the rising cost of living. Government revenue has a positive and statistically significant impact on government education expenditure in the long run. This implies that as government revenue collection increases, government education expenditure also rises. This finding may be interpreted within the South African context, where government expenditure exceeding revenue compels the government to raise tax rates to generate additional revenue to finance rising expenditure. Table 5 further reveals that the urban population exerts a negative and statistically significant impact on government education expenditure in the long run. This counterintuitive finding may be interpreted within the South African context, where, in urban areas, there is a mixture of government and private sector participation. Therefore, there is a shared provision of social services, such as health and education, resulting in the government shifting part of the burden to the private sector, which reduces public social expenditure despite a rising urban population.

• Government health expenditure

The results presented in Table 5 reveal that debt service cost displays a negative relationship with government health expenditure in the long run. This implies that a 1% increase in debt service cost would result in an approximately 0.25% decrease in government health expenditure in South Africa, ceteris paribus. The economic implication of this finding is that the government allocates more funds toward servicing debt, resulting in a larger share of the national budget being absorbed by debt obligations. Consequently, this reduces the fiscal space available for health expenditure. Therefore, higher debt servicing crowds out government spending on health. Public debt is found to have a negative and insignificant impact on government health expenditure in the long run. Moreover, the exchange rate and urban population have a positive and insignificant impact on government health expenditure in the long run. Government revenue is found to exert a positive and significant impact on government health expenditure in the long run. This implies that a 1% increase in government revenue would result in an approximately 1.16% increase in government health expenditure in the long run, ceteris paribus. The significant negative impact of debt service costs and public debt on government social protection expenditure indicates that high debt service costs limit fiscal funds available for social protection.

• Government social protection expenditure

The results presented in Table 5 reveal that debt service costs and public debt display a negative relationship with government social protection expenditure in the long run. An increase in debt service cost and public debt by 1% will lead to about a 0.88% and 1.14% decrease in government social protection expenditure in the long run, respectively. This implies that growing indebtedness and rising interest obligations limit government social protection expenditure in the long run. Economically, increasing debt service costs and public debt may limit fiscal space, thereby constraining government capacity to increase social protection expenditure. The exchange rate, government revenue and urban population have a positive and statistically significant relationship with government social protection expenditure in the long run. Regarding the urban population, the results suggest that a percentage increase in the number of people living in urban cities increases social protection expenditure by 3.16% in the long run, ceteris paribus. The statistical significance of the urban population may reflect the government’s deliberate efforts to channel additional resources toward social protection programmes aimed at preserving social stability and fostering economic inclusion.

Overall, the effects of debt service cost on social expenditure in the long run are negative and statistically significant across all the categories of social expenditure under consideration. These findings align with the debt overhang theory and crowding-out hypothesis on the trade-off between debt service and government expenditure. Moreover, the existing empirical literature has established a negative influence of debt service on social expenditure (Gurowa et al. 2023; Mbewe 2024; Shabbir and Yasin 2015). In addition, the negative relationship between the public debt and social expenditure categories aligns with the findings of Said and Sani (2020) and Tashevska et al. (2020). This underscores the importance of practical debt management and sustainable fiscal policy to prevent the crowding-out of social expenditure. The positive relationship between the government revenue and social expenditure categories corroborates the ideas of Peacock and Wiseman (1961) regarding the theory of public expenditure, which highlights that government expenditure and revenue (tax) are positively related because of the displacement effect. Furthermore, this relationship is consistent with the empirical literature (Iliyasu and Gambo 2022; Shabbir and Yasin 2015). The f-statistic for all models is statistically significant, suggesting that the chosen explanatory variables provide meaningful statistical insight into the variation in South African social expenditure.

Table 6 reveals that in the short run, debt service is negative and statistically significant with all three categories of social expenditure in South Africa. However, public debt, the exchange rate, and government revenue have a significant positive impact on all three categories of social expenditure. Urban population has a positive relationship with government education expenditure and government health expenditure. However, it has a negative relationship with government social expenditure. The findings highlight that urban population influences social expenditure differently over time. In the short run, the government could prioritise increasing education expenditure because of increasing demand for schools and education infrastructure while social protection expenditure is reduced. In the long run, social protection expenditure is increased due to welfare demand, whereas education expenditure declines as the growth of private schools in urban areas reduces the pressure on public education spending. We now turn to the error correction term (ECT), which measures the speed of adjustment. The observed ECT coefficients for the three models are −0.5611, −0.4625, and −0.5351 and are statistically significant. These negative coefficients indicate that any short-run deviation from the determined long-run equilibrium will spontaneously initiate a corrective mechanism, driving the model back to its equilibrium state. The extent of the government education expenditure model’s coefficient, −0.5611, indicates the speed of adjustment. Particularly, it suggests that roughly 56.11% of any disequilibrium in the model observed in the previous period is rectified in the current period. Moreover, this adjustment would take approximately one year and seven months (1/0.5611 = 1.78), since the speed of adjustment is moderate. Across all three models, the estimated speed of adjustment is moderate. To guarantee the robustness of the model, several diagnostic tests are conducted and the results are presented in

Table 7. Residual and stability diagnostic tests.

Tests	LnGEE	LnGHE	LnGSPE
Normality (Jarque–Bera)	0.3147 (0.8543)	0.0696 (0.9657)	1.0335 (0.5964)
Serial correlation (LM)	2.6328 (0.2681)	4.7277 (0.0941)	0.4373 (0.8036)
Heteroskedasticity (Breusch–Pagan–Godfrey)	25.5745 (0.3214)	24.6771 (0.3671)	23.8901 (0.4099)
Ramsey RESET	6.4435 (0.1264)	0.8368 (0.4569)	1.4008 (0.4466)

Source: Author’s own estimation obtained from EViews 14.

.

The diagnostic test results presented in Table 7 reveal that all models are correctly specified, as we fail to reject the null hypothesis of the Ramsey RESET test. Moreover, all models pass the assumption of homoscedasticity, normality and no autocorrelation, suggesting that the estimated residuals are free of serial correlation and heteroscedasticity, and they are also normally distributed.

5. Conclusions and Policy Implications

This study set out to investigate the impact of public debt service costs on social expenditure in South Africa. The rationale behind this investigation lies in the country’s rising debt service costs. Therefore, this study looked into the nexus between debt service costs and social expenditure, using three categories of social expenditure, namely, government education expenditure, government health expenditure and government social protection expenditure. This investigation was carried out using an autoregressive distributed lag model and annual time series data from 1994 to 2024. The variables used in the study are as follows: government education expenditure, government health expenditure, government social protection expenditure, debt service, public debt, exchange rate, revenue and urban population.

The results of the ARDL model reveal that debt service exhibits a negative and statistically significant impact across all three categories of social expenditure under consideration in the long run and short run. This suggests the crowding-out effect of debt service on social expenditure in South Africa. The rising debt service costs decrease social expenditure in South Africa. This finding aligns with the debt overhang theory and crowding-out hypothesis on the trade-off between debt service and government expenditure. Moreover, the results reveal that public debt has a negative relationship with government education expenditure, government health expenditure and government social protection expenditure; however, the relationship is statistically significant with government social expenditure. The exchange rate and revenue were found to have a positive relationship with all three categories of social expenditure under consideration. Surprisingly, urban population was found to exhibit a negative effect on government education expenditure in the long run. However, it was found to have a positive relationship with government health expenditure and government social protection expenditure.

The findings of the study have policy implications. The data reveals that government expenditure on debt service is crowding out the funds available for social expenditure in South Africa. Therefore, this underscores the need to focus on reducing the fiscal pressure stemming from increasing debt service costs, while upholding social expenditure. It is recommended that policymakers should focus on debt stabilisation and reduction, thereby easing the crowding-out of social expenditure. Moreover, policymakers should ensure that restrictions such as expenditure floors for social sections are introduced to protect social expenditure. While the proposed strategies are theoretically sound, to be successfully implemented, they need to strike a balance between fiscal sustainability and political reality in South Africa. A gradual tactic of fiscal modification complemented by effective public financial management could be a more sustainable strategy for protecting social expenditure while dealing with rising debt service costs. The data further reveals that revenue generation has a beneficial effect on social expenditure in South Africa. Therefore, policymakers should implement programmes that encourage private investment and support small enterprises, thereby generating sustainable government revenue to finance social expenditure. The policy recommendations of this study are important for fiscal policy management in South African and other countries with similar economic challenges.

The study produced insights into the relationship between public debt service costs and social expenditure in South Africa; however, it has some drawbacks. The current study has relied on the impact of public debt service cost on social expenditure in South Africa. Future studies could benefit by considering a comparative analysis between economies with similar characteristics.