Labour Productivity, Wages, and Social Welfare: Implications for South Africa’s Budget Deficit and Fiscal Policy

Abstract

This study investigates the relationship between labour productivity, wages, and social welfare expenditure (SWE) in South Africa, with implications for fiscal sustainability and budget deficits. A theoretical model linking government expenditure, taxation, and labour market dynamics is developed and empirically tested using data from 1994 to 2022. Results from state and private labour market regressions reveal significant evidence of wage–productivity decoupling in the state labour market, where wages are influenced more by institutional factors than productivity growth. Conversely, private sector wages show a positive association with productivity, inflation, and working capital balances. The budget deficit model demonstrates strong alignment with empirical trends, though it underestimates the impact of economic shocks such as COVID-19. Findings suggest that increases in productivity alone will not reduce social welfare dependency in South Africa, given structural inequality, weak labour absorption, and low skills development. Policy implications highlight the need for targeted investment, industrial expansion, and education reform to mitigate rising welfare expenditure and ensure fiscal sustainability.

1. Introduction

Declining global labour productivity in recent decades, coupled with declining economic growth, has coincided with increasing conflict between capital and labour, increasing income inequality, and the rise of political populism (Erber et al. 2017; Alvaredo et al. 2018; Jackson 2019). In the pursuit of higher economic growth, and by extension increased labour productivity, it is assumed that social progress without economic growth is impossible (Jackson 2019). In pursuing economic growth, modern economies have adopted debt-driven monetary regimes despite no consensus for a causal relationship going from debt to economic growth (Panizza and Presbitero 2014; Erber et al. 2017). Governments have increasingly adopted larger budget deficits in recent decades, leading to increased public debt despite the negative effect of increasing tax and interest expenditure affecting savings rates (Teles and Mussolini 2014). This is due to labour productivity not providing the required economic growth to sustain increasing debt levels and debt-service costs (Erber et al. 2017).

In South Africa, gross domestic product (GDP) increased on average by 2.3% annually between 2001 and 2021 (World Bank 2023). Net public debt over the same period increased on average by 5.4% (National Treasury 2023). The dichotomous growth rates of GDP and net public debt have led to South Africa’s debt-to-GDP ratio increasing from 44% in 2001 to 74% in 2021. High debt-to-GDP ratios have negative impacts on economic activity and long-term interest rates (Baum et al. 2013; Alcidi and Gros 2019). Between 2001 and 2021, the low rate of economic growth in South Africa was attributed to low entrepreneurial activity, low-quality educational outcomes, persistently high unemployment, onerous labour legislation, high levels of corruption, and low fiscal sustainability due to the deteriorating debt-to-GDP ratio (Mahadea and Simson 2010; Van Scheers 2016; Meyer 2017; Mhlaba and Phiri 2019; Burger and Calitz 2021). Collectively, these factors contribute to South Africa being one of the most economically unequal countries in the world (Francis and Webster 2019).

Despite government’s policy focus on supporting the redistribution of income and the promotion of social welfare, increasing public expenditure alone does not necessarily achieve the desired outcomes of reduced poverty and inequality (Rajkumar and Swaroop 2008). This is evident when considering that social development expenditure has become the largest expenditure item in the 2022 budget at 16.9% of total expenditure. Debt-service cost also increased over the past two decades to become the second largest expenditure item at 14% (National Treasury 2023). These expenditure outcomes have been achieved despite a persistently high unemployment rate, increased public debt and very low economic growth (Rodrik 2008; Alexander 2010; National Treasury 2023).

The primary objective of this study is to develop a theoretical model to describe the relationship between labour productivity, market wages, and social welfare expenditure (SWE) in a national budget, with a specific focus on how these different factors contribute to either a budget deficit or surplus. The model framework will be subjected to empirical analysis for the period 1994 to 2022 in South Africa. Secondly, the study will also aim to investigate the implications of the primary objective on fiscal and tax policies adopted over the period. The findings of this study will make a significant contribution to the growing literature on labour productivity, social development objectives, and forward-looking fiscal economic policy tools in developing economies such as South Africa.

2. Literature Review

Over the past few decades, global economic trends have been characterised by weaker economic growth and declining labour productivity (Alvaredo et al. 2018; Erber et al. 2017; Jackson 2019). Economic theory assumes that social progress requires economic growth, which is largely driven in positive association with changes in labour productivity (Jackson 2019). However, the adoption of debt-driven monetary policy led to increasing public debt levels and budget deficits. Across this period, labour productivity has been unable to increase at the required rate to maintain stable or decreasing debt-service costs (Erber et al. 2017; Teles and Mussolini 2014). In the case of South Africa, low economic growth over the past two decades is directly linked to structural issues, such as persistently high unemployment, poor educational outcomes, onerous labour legislation, and low fiscal sustainability due to a deteriorating debt-to-GDP ratio (Burger and Calitz 2021; Mahadea and Simson 2010; Meyer 2017). The compounding effect of these structural factors has led to low rates of change in income inequality outcomes over time, leading to social development expenditure becoming one of the largest budget items, yet literature notes that increasing public expenditure alone does not necessarily translate to reduced poverty and inequality (Rajkumar and Swaroop 2008). These outcomes highlight the difficulty of achieving fiscal sustainability whilst addressing deep-seated structural inequalities.

Classical economic theory posits a positive association between labour productivity and wages (Nikulin 2015). However, a significant body of contemporary research finds evidence for a decoupling effect, particularly in developed economies, where labour productivity gains do not necessarily translate into proportional wage growth (Gil-Alana and Skare 2017). The literature indicates that several factors drive this decoupling effect.

Macroeconomic policy factors, such as minimum wage policies, inflation, interest rates, economic growth, and unemployment rates, affect labour productivity and wage decoupling (Compagnucci et al. 2021; Gil-Alana and Skare 2017; Nasir et al. 2022).

The structure of the labour market and the level of technology present in the labour market influence and reinforce the wage–productivity relationship through factors such as capital investment, education levels, labour market organisation, and the nature of employment (Fatuła 2018; Islam et al. 2015; Katovich and Maia 2018; Škare and Škare 2017). Specific to the South African context, the high proportion of low-skilled labour, high unemployment, and the size of the informal labour market create an environment characterised by structural barriers, contributing to this decoupling effect (Gradín 2021a; Klein 2012; Posel and Casale 2019)

The link between the labour market and fiscal policy is a critical area of investigation since state labour employment levels are often driven by factors distinct from private sector profitability. Such factors include demand for state services, the poverty rate, and the extent of unionisation (Herrera and Munoz 2019; Maluleke 2016). Furthermore, the observation of Wagner’s law—that government expenditure increases as national income increases—is confirmed in the South African economy, linking the size of the state’s budget directly to national income (Akitoby et al. 2006; Odhiambo 2015).

The efficacy of SWE in reducing poverty and inequality remains a focal point in fiscal studies (Sidek 2021). While SWE is intended to redistribute income, its success is limited if the labour market cannot absorb the working-age population into productive, high-wage employment. Ultimately, government revenue (derived from personal income tax, corporate tax, and VAT) and expenditure (including SWE and non-SWE) collectively determine the budget deficit or surplus (Ojong et al. 2013).

The existing literature establishes the independent importance of both the global and local drivers of wage–productivity decoupling and the relationship between social welfare spending and overall fiscal outcomes. The main weakness identified in the existing literature is the lack of a unified, comprehensive theoretical model that explicitly integrates the dynamics of labour productivity, market wages (distinguishing between public and private sectors), and social welfare expenditure (SWE) to directly predict or infer the budget deficit or surplus within a developing economy context like South Africa.

This study thus presents novel and significant contributions through the development of an integrated theoretical and empirical framework, where a novel theoretical model linking government expenditure and taxation with segmented labour market dynamics (public versus private) to specifically model the budget differential is presented. Secondly, through the approach of a segmented labour market model, this study provides a dual-sector decoupling effect, which offers a crucial, nuanced perspective missing in much of the wider decoupling literature. This study also makes a significant contribution by providing forward-looking fiscal economic policy tools for developing economies.

3. Materials and Methods

3.1. Social Welfare Expenditure Model

Government expenditure in a year is given by Equation (1)

$$B_E = B_{-S} + B_S$$

(1)

where $B_E$ is total government expenditure, $B_{-S}$ is non-social welfare expenditure, and $B_S$ is social welfare expenditure, which refers to social income grants only in this model (National Treasury 2023). Since government balances expenditure against income generated through various taxes (National Treasury 2023), income generated is given by Equation (2)

$$T_R= PIT+CIT+VAT+OTH$$

(2)

where $T_R$ is total revenue, $PIT$ is personal income tax, $CIT$ is corporate income tax, $VAT$ is value-added tax, and $OTH$ are other sources of tax revenue, such as customs, excise duties, and other general levies. The budget deficit or surplus can be defined accordingly as the difference between income and expenditure, given by Equation (3)

$$B_{\Delta }=B_E - T_R = (B_{-S} + B_S)-\left(PIT+CIT+VAT+OTH\right)$$

(3)

where $B_{\Delta }$ is the budget differential. When $B_{\Delta }>0$ this constitutes a budget deficit due to an income shortfall or excess expenditure. When $B_{\Delta }<0$ this constitutes a budget surplus due to an income surplus or shortfall in expenditure (Ojong et al. 2013). Equation (3) can be expanded by the following Equations (4)–(9) related to the variables $B_{-S}$, $B_S$, $PIT$, $CIT$, and $VAT$. The variable $B_{-S}$ is defined by Equation (4)

$$B_{-S}=B_{-S}^{*} \ast N_L \Rightarrow B_{-S}^{*}={\displaystyle \frac{B_{-S}}{N_L}}$$

(4)

where $B_{-S}^{*}$ is the per capita (labour force) government expenditure related to non-social welfare, and $N_L$ is the labour force population, defined as the population of working age that can participate in labour. The variable $B_S$ is defined by Equation (5)

$$B_S=N_S\ast W_S$$

(5)

where $N_S$ is the number of individuals receiving a social welfare income grant and $W_S$ is the average annual social welfare income grant. We assume that the number of individuals receiving a social welfare income grant $N_S$ excludes disability, military veterans, and old-age grant recipients. This is due to such individuals largely not forming part of the labour force population (Statistics South Africa 2022). $N_S$ thus approximates the number of unemployed individuals within the labour force, given by Equation (6)

$$N_S\approx N_{UE} = N_L-N_E$$

(6)

where $N_{UE}$ is the number of unemployed individuals, $N_E$ is the number of employed individuals, and $N_L$ is the population of working-age individuals who can participate in labour. This equation defines the total number of individuals receiving social welfare grants ($N_S$) as being approximately equal to the number of unemployed individuals ($N_{UE}$) within the labour force. It simplifies the social expenditure challenge by equating those receiving grants with the population who can work but are currently jobless ($N_L-N_E$). $PIT$ is defined by Equation (7)

$$PIT = N_E\ast W_A\ast T_{R,P}$$

(7)

where $W_A$ is the average annual gross labour wage, and $T_{R,P}$ is the average tax rate on personal income. $CIT$ is defined by Equation (8)

$$CIT = C_P\ast T_{R,C}$$

(8)

where $C_P$ is total corporate profits and $T_{R,C}$ is the corporate tax rate. $VAT$ is defined by Equation (9)

$$VAT=G_C\ast T_{R,V}$$

(9)

where $G_C$ is total household consumption, $T_{R,V}$ is the value-added tax rate, and $k$ is a factor representing the proportion of consumption eligible for taxation. Equation (3) can thus be substituted with Equations (4)–(9) to yield Equation (10)

$$B_{\Delta }=\left[\left(B_{-S}^{*}\ast N_L\right)+\left(N_S\ast W_S\right)\right]-\left[\left(N_E\ast W_A\ast T_{R,P}\right)+\left(C_P\ast T_{R,C}\right)+\left(G_C\ast T_{R,V}\right)+OTH\right]$$

(10)

We can rearrange Equation (10) when $B_{\Delta }=0$, given by Equation (11)

$$\left(B_{-S}^{*}\ast N_L\right)+\left(N_S\ast W_S\right)=\left(N_E\ast W_A\ast T_{R,P}\right)+\left(C_P\ast T_{R,C}\right)+\left(G_C\ast T_{R,V}\right)+OTH$$

(11)

We then multiply both sides of Equation (11) by ${\displaystyle \frac{1}{N_L}}$ and substitute Equations (2) and (4) to yield Equation (12)

$$\left[\left({\displaystyle \frac{B_{-S}}{N_L}}\right)+\left({\displaystyle \frac{N_S}{N_L}}\ast W_S\right)\right]=\left[\left({\displaystyle \frac{N_E}{N_L}}\ast W_A\ast T_{R,P}\right)+\left({\displaystyle \frac{G_C}{N_L}}\ast T_{R,V}\right)\right]+\left[{\displaystyle \frac{\left(C_P\ast T_{R,C}\right)+OTH}{N_L}}\right]$$

(12)

which represents variables related to social welfare expenditure on the left-hand side of the equation, and variables related to government revenue on the right-hand side of the equation.

3.2. Wage and Labour Productivity Model

The labour market is segmented into two distinct labour sectors, namely the state labour market sector and the private labour market sector (Burdett 2012). Whilst the private labour market sector can further be segmented into formal and informal labour sectors, this segmentation is not incorporated into the model; in the South African labour market, the employment and wage contribution of the informal labour market to the total labour market is disproportionately small (Davies and Thurlow 2010; Shahen et al. 2020).

Since there are two distinct labour market sectors, individuals opt for preferred employment between the two sectors based on the overall net economic benefits in each sector (Lokshin and Jovanovic 2003). In South Africa, the state labour market is associated with a wage premium relative to the private labour market (Bhorat et al. 2016). Individuals consequently accept private sector employment only if the state labour market has reached saturation or the private sector provides a higher net economic benefit, described by Equation (13)

$$i ϵ N_E\to \left\{\begin{array}{c} N_{E,P} if W_{A,P}>W_{A, S}\\ N_{E,S} if W_{A,P}<W_{A, S}\end{array}\right.$$

(13)

subject to Equation (14)

$$N_{E,P}+ N_{E,S}= N_E ~ f\left(GDP, K, T\right)$$

(14)

where $i$ represents an individual in the employed labour market, $N_{E,P}$ is the number of employment opportunities in the private labour market sector, $N_{E,S}$ is the number of employment opportunities in the state labour market sector, $W_{A,P}$ is the average annual gross wage in the private labour market sector, and $W_{A,S}$ is the average annual gross wage in the state labour market sector. The function $f$ relates the number of employment opportunities in the economy as a function of gross domestic product ($GDP$), gross capital ($K$), and the degree of technological innovation in the economy ($T$).

3.2.1. State Labour Market Model

The number of employment opportunities that exist in the state labour market sector is dependent largely on the overall demand for state services (Herrera and Munoz 2019). The demand for state services is driven by the overall poverty rate, state wage rate, unionisation in the state labour market, urbanisation, national income, and trade openness factors (Maluleke 2016). If population size determines the number of state employees required to render public service, under the assumption that other factors exist in a sufficient state, Equation (15) defines the number of state employees with respect to the proportion of the population that requires state services

$$N_{E,S}=m_1 lo{g}_{\alpha }\left(N_P^{*}+k_1\right)+k_2$$

(15)

where $N_P^{*}$ is the population dependent on state services, $\alpha$ is a coefficient, and $m_1$, $k_1$ and $k_2$ are constants. Equation (15) assumes a log equation form, since the proportional increase in state labour is a non-negative, decreasing slope as population size increases. This is attributed to constraints on higher proportional employment given national income and expenditure constraints. This equation models the number of state employees ($N_{E,S}$) as a logarithmic function of the population dependent on state services ($N_P^{*}$). The logarithmic form ensures that while the number of state jobs increases as the dependent population grows, the proportional rate of increase gradually slows down. This diminishing return reflects the financial and fiscal constraints, such as limits on national income and expenditure, that prevent the state from indefinitely expanding its payroll at the same rate as the demand for services, as well as taking productive efficiency gains with employment scaling into consideration.

In the South African context, most of the population is dependent on state services; thus, Equation (16)

$$N_P^{*} \approx N_P$$

(16)

which approximates $N_P^{*}$ as the full population $N_P$. This approximation is likely to overstate the population dependent on state services, given that universal unemployment benefits do not exist. The approximation is substantiated by the large number of grant recipients relative to the number of unemployed individuals. Substituting Equation (16) into Equation (15) and taking the first derivative of Equation (15) yields Equation (17)

$${\displaystyle \frac{\Delta N_{E,S}}{m_1}} = m_2\left[{\displaystyle \raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$ 0.1+[N_P \ln \left(\alpha \right)+k_1 \ln \left(\alpha \right)]$}\right.}\right]+k_3$$

(17)

where $\Delta N_{E,S}$ is the cumulative change in state employment levels and $m_2$ and $k_3$ are constants. The constant 0.1 is added in the denominator to normalise the function. The GDP per capita relates the size of the economy to the population, given by Equation (18)

$$N_P = {\displaystyle \raisebox{1ex}{$GDP$}\!\left/ \!\raisebox{-1ex}{$GD{P}_{N_P}$}\right.}$$

(18)

where $GD{P}_{N_P}$ is GDP per capita. Substituting Equation (18) into Equation (17) and rearranging yields Equation (19)

$$GDP= m_3\left[{\displaystyle \frac{1}{GD{P}_{N_P}}}\left({\displaystyle \frac{m_2}{\left[\frac{\Delta N_{E,S}}{m_1}-k_3\right]}}-\left(k_1+0.1\right)\ln \left(\alpha \right)\right)\right]+k_4$$

(19)

Wagner’s law, the observation that state expenditure increases as national income increases, is observed in both the short and long run in the South African economy (Akitoby et al. 2006; Odhiambo 2015). A linear transformation is applied to Equation (19), yielding Equation (20)

$${\displaystyle \frac{R}{GDP}} = m_3\left[{\displaystyle \frac{1}{GD{P}_{N_P}}}\left({\displaystyle \frac{m_2}{\left[\frac{\Delta N_{E,S}}{m_1}-k_3\right]}}-\left(k_1+0.1\right)\ln \left(\alpha \right)\right)\right]+k_4$$

(20)

where $R = 1E+15$ is a scaling factor. Since national income and expenditure are inextricably linked to the size of the economy (Magazzino 2012), state expenditure per capita is proportional to GDP per capita as given by Equation (21)

$$B_{E, N_P} = {\displaystyle \raisebox{1ex}{$B_E$}\!\left/ \!\raisebox{-1ex}{$ N_P$}\right.} = m_4 GD{P}_{N_P}+k_5$$

(21)

where $B_{E, N_P}$ is the budget expenditure per capita, $m_4$ is a coefficient, and $k_5$ is a constant. The sum of all state employee wages is a proportion of the state’s budget expenditure as per Equation (22)

$$N_{E,S}\ast W_{A,S}=W_{E,S}= m_5 B_E+k_6$$

(22)

where $W_{E,S}$ is the annual state labour wage, $m_5$ is a coefficient, and $k_6$ is a constant. Rearranging and substituting Equations (21) and (22) into Equation (20) yields Equation (23)

$$W_{E,S}=m_5 N_P\left[{\displaystyle \frac{m_3 m_4 N_{E,S}^{*}}{\left(GD{P}^{*}-k_4\right)}}+k_5\right]+k_6$$

(23)

where

$$N_{E,S}^{*}={\displaystyle \frac{m_2}{\left[\frac{\Delta N_{E,S}}{m_1}-k_3\right]}}-\left(k_1+0.1\right)\ln \left(\alpha \right)$$

and

$$GD{P}^{*}={\displaystyle \frac{R}{GDP}}$$

which relates the annual state labour wage as a function of the size of the population, the size of the economy, and the change in the size of the existing state labour market.

3.2.2. Private Labour Market Model

The private labour market sector utilises labour as a component with respect to the fundamental output of firms, which is to generate profitability and return in accordance with shareholder expectations, given the level of capital risk (Jones and Felps 2013). Factors, such as the size of the economy, economic growth, skills availability in the labour market, the level of unemployment, the degree of capital available in the market, inflation, and the interest rate, dictate how firms elect to balance labour versus technological integration in firm production (Arvanitis and Loukis 2009; Iamsiraroj 2016). Firms also need to balance labour input, and by extension wages, against market prices for produced goods to ensure firm profitability (Carlsson and Westermark 2011).

Conversely, individuals in the labour market possess a minimum wage threshold that must be met in the employment market to catalyse labour market participation to support basic standard of living requirements as opposed to the non-economic benefits related to remaining unemployed (Carlsson and Westermark 2011; Mann 2020). We assume there exists an interdependence between current wages at time $t=i$ and future wages at time $t=i+1$ through the market price mechanism for produced goods. This arises due to the firm’s market price for produced goods at time $t=i$ affecting firm profits at $t=i$, which affects labour demand at $t=i+1$ due to liquidity constraints (Carlsson and Westermark 2011).

Firms manage profitability and liquidity through the efficient management and funding of their working capital cycle via both equity and debt (Anderson and Carverhill 2012). The equation for determining working capital is given by Equation (24)

$$K_W=C_A-C_L=\sum C_{A,i}-\sum C_{L,i}={{\displaystyle \sum }}_{i=1}^j(C_{A,i}-C_{L,i})$$

(24)

where $K_W$ is the balance of working capital, $C_A$ is the balance of current assets with components $i$, and $C_L$ is the balance of current liabilities with components $i$. The balance of current assets comprises cash and cash equivalents, accounts receivable, inventory, marketable securities, prepaid liabilities, supplies, and other liquid assets (Dong and Su 2010). The balance of current liabilities comprises accounts payable, accrued salaries and related expenses, current payables on long-term debt, short-term debt, tax payable, and other accrued expenses (Durrah et al. 2016).

Firms’ focus on efficient working capital management requires a degree of ease in obtaining external liquidity as and when required over the course of the economic cycle (Enqvist et al. 2014). This factor determines the level of employment the firm can support at given levels of profitability, especially for small or labour-intensive firms (Dao and Liu 2017). This mechanism relies on the intuition that, particularly in credit-constrained emerging markets, higher cumulative working capital provides firms with the necessary internal liquidity to finance immediate operational costs, such as scaling up private employment, without having to rely on expensive or unavailable external financing. This channel is crucial for understanding job expansion capacity when traditional bank lending is tight or inaccessible. Firms manage the number of employment opportunities in equilibrium with market-clearing wages and working capital constraints, given by the linear Equation (25)

$$N_{E,P}=M_1{{\displaystyle \sum }}_{t=1}^u{\left[{{\displaystyle \sum }}_{i=1}^j{{\displaystyle \sum }}_{i=1}^j(C_{A,i}-C_{L,i})\right]}_t+z_1$$

(25)

where $N_{E,P}$ is the number of employment opportunities in the private labour market sector, ${{\displaystyle \sum }}_{i=1}^j(C_{A,i}-C_{L,i-1}^{*})$ is the summation of working capital balances for firms $i$ in each year, ${{\displaystyle \sum }}_{t=1}^u{\left[\dots \right]}_t$ is the cumulative summation of working capital balances for each year, and $M_1$ and $z_1$ are constants. This equation states that the number of private sector employment opportunities ($N_{E,P}$) is linearly related to the cumulative balance of working capital for all firms. The intuition is that sustained, high cumulative working capital across the economy is a proxy for the total liquidity and financial health of the private sector, which provides the necessary financial buffer to support and expand its workforce.

Assuming market-clearing wages in the private labour market, the average wage number in the labour market is tied directly to the initial average wage and a proxy for inflation over time, with consideration given to working capital constraints, given by Equation (26)

$$W_{E,P}=N_{E,P} {\overline{W}}_{E,P}\iff {\overline{W}}_{E,P,t}=M_2\left[{\overline{W}}_{t=0} \left(1+{{\displaystyle \sum }}_{i=1}^j{I}_t\right)\left(\sqrt{(1+\Delta GD{P}_{t-1}}\right)\right]+z_2$$

(26)

where $W_{E,P}$ is the annual private labour market wage, ${\overline{W}}_{E,P}$ is the average annual private labour market wage, $t$ is the specific period in years, ${\overline{W}}_{t=0}$ is the average annual private labour market wage at time $t=0$, ${{\displaystyle \sum }}_{i=1}^j{I}_t$ is the cumulative inflation rate, $\Delta GD{P}_{t-1}$ is the annual growth rate in GDP at time $t-1$, and $M_2$ and $z_2$ are constants. This equation models the average annual private labour market wage ($W_{E,P}$) as a function of the initial wage, cumulative inflation (${{\displaystyle \sum }}_{i=1}^j{I}_t$), and past GDP growth ($\Delta GD{P}_{t-1}$). The inclusion of the square root function on the GDP growth rate ($\Delta GD{P}_{t-1}$) is an empirical choice to reflect the non-linear relationship where wage increases experience diminishing returns at very high economic growth rates. This structural form captures the real-world observation that wages tend to be sticky and do not increase proportionally with extremely high national income growth, with much of the benefit often accruing to profits rather than labour remuneration. Rearranging Equation (26) and substituting into Equation (25) yields Equation (27)

$$W_{E,P}=M_3\left[M_1\left[\left(C_t^{\gamma }\right)\left(M_2+z_2\right)\right]+z_1\left[\left(M_2{I}_t^{*}\right)+z_2\right]\right]+z_3$$

(27)

where

$$C_t^{\gamma }={{\displaystyle \sum }}_{t=1}^u{\left[{{\displaystyle \sum }}_{i=1}^j(C_{A,i}-C_{L,i})\right]}_t$$

(28)

and

$$I_t^{*}={\overline{W}}_{t=0} \left(1+{{\displaystyle \sum }}_{i=1}^j{I}_t\right)\left(\sqrt{(1+\Delta GD{P}_{t-1}}\right)$$

(29)

Equation (27) describes the relationship between the annual private labour market wage $W_{E,P}$ and factors related to working capital balances, the initial wage level in the private labour market at time $t=0$, and inflation.

A random selection of the largest firms by market capitalisation on the Johannesburg Stock Exchange across several different industries is used as proxy to estimate the annual private labour market wage across the model period. Working capital data is restricted to the period 2000 to 2021 due to limited data availability for some firms. Pre-2000 working capital balances are extrapolated from the available data and zero-based in 1993. The annual private labour market wage is estimated from South African Revenue Service tax data, assuming an average personal income tax rate of 20%. The income tax rate remained relatively stable over the period under analysis; however, during changes in income tax rates, measurement error may accumulate over the time period due to non-linear effects.

3.3. Budget Deficit Model

Labour productivity is defined as the amount of goods and services produced per unit of time (Attar et al. 2012), and is given by Equation (30)

$$LP= {\displaystyle \frac{GDP}{N_{E }\ast T_W}}$$

(30)

where $GDP$ is gross domestic product and $T_W$ is the amount of labour hours assuming a 40 h workweek. Wages are assumed to follow a bilateral causality with labour productivity, since strong unionisation and labour regulation factors can swing causality in the direction whereby employees are more able to dictate and set market wages, especially in the South African state labour market (Budeli 2012; Tang 2012; Botha 2015). In general, unions are strong contributors to income inequality reduction through their capability to achieve wage growth for their members (Barth et al. 2020). In the South African context, union wage premia are extremely high, and unions have contributed to a marginal increase in income inequality as they are concentrated at the top of the income distribution and in the state labour market (Ntlhola et al. 2019; Kerr and Wittenberg 2021). Skilled workers also command wage increases despite negligible increases or decreases in overall labour productivity (Habanabakize et al. 2019).

Wages, defined in the state labour market and private labour market models, are given by Equation (31)

$$W_A=\left(W_{E,S} + W_{E,P}\right)/N_E$$

(31)

We can substitute Equation (31) into Equation (12) for $W_A$ to yield Equation (32)

$$Z_1=Z_2+Z_3$$

(32)

where

$$Z_1=\left({\displaystyle \frac{B_{-S}}{N_L}}\right)+\left({\displaystyle \frac{N_S}{N_L}}\ast W_S\right)$$

(33)

$$Z_2=\left({\displaystyle \frac{N_E}{N_L}}\ast \left[\left(W_{E,S} + W_{E,P}\right)/N_E\right]\ast T_{R,P}\right)+\left({\displaystyle \frac{G_C}{N_L}}\ast T_{R,V}\right)$$

(34)

$$Z_3={\displaystyle \frac{\left(C_P\ast T_{R,C}\right)+OTH}{N_L}}$$

(35)

When $Z_1=Z_2+Z_3$, the budget deficit or surplus is zero, and government income and expenditure is at parity. When $Z_1>Z_2+Z_3$ there is a budget deficit. When $Z_1<Z_2+Z_3$ there is a budget surplus. Figure 1 provides a schematic of the relationships between variables in the overall theoretical model framework, showing the dependence between labour, population, and tax variables on fiscal output variables.

The proposed model assumes the form of a multiple linear regression. Ordinary least squares estimation is used to determine the relevant model coefficients. Model goodness-of-fit and validity are determined through testing for normality, constant error variance, and independence (Altman and Krzywinski 2016). These include inspecting visual plots, such as the normal quintile–quintile (Q-Q) plot, actual versus predicted values plot, the distribution of actual values versus residuals, and the frequency of observed budget deficits or surpluses versus predicted budget deficits or surpluses. The Kolmogorov–Smirnov (KS) test is used to determine whether the actual versus predicted distributions are statistically different (Berger and Zhou 2005). The Granger causality test is used to test for predictive causality between input and output variables, which indicates non-stationarity (Shojaie and Fox 2022).

The data used in this analysis was sourced for the period 1994 to 2021. All variables, except for population count and the tax rate variables, used in this study are observed in South African Rand at nominal value across the sample period. Population counts are observed as numerical counts, and tax rate variables are observed as percentiles. Annual observations were used, given the constraints of data availability on an intra-year basis. SWE excludes pensions and non-labour-force grants, focusing on social wage expenditure linked to unemployment benefits and social support.

4. Results

Results for the study are divided into two sections. Model results are presented first, followed by tests for model goodness-of-fit and validity.

4.1. State Labour Market Model

4.1.1. Model Results

The results for Equation (15) of the state labour market model are shown in

Table 1. Results for Equation (15) ($N_{E,S}=m_1 lo{g}_{\alpha }\left(N_P^{*}+k_1\right)+k_2$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$m_1$	4.344 × 105	1.234 × 105	3.522	0.00154	0.2894
$k_1$	0
$k_2$	−9.943 × 106	3.158 × 106	−3.149	0.00398
$\alpha$	2

Source: Authors’ calculations.

.

The $\alpha$ parameter base for $\log \left(\alpha \right)$ is arbitrarily defined as 2. Table 1 indicates that both model coefficients are significant within the linear model at the 5% significance level. The adjusted R² is relatively good, since both coefficients are statistically significant (Ozili 2023). The results for Equation (17) of the state labour market model are shown in

Table 2. Results for Equation (17) (${\displaystyle \frac{\Delta N_{E,S}}{m_1}}= m_2\left[{\displaystyle \raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$[0.1+ N_P \ln \left(\alpha \right)+k_1 \ln \left(\alpha \right)]$}\right.}\right]+k_3$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$m_2$	−5.232 × 107	1.447 × 107	−3.617	0.00121	0.3015
$k_3$	2.098	4.151 × 10−1	5.056	2.620 × 10−5

Source: Authors’ calculations.

.

Table 2 indicates that both model coefficients are significant within the linear model at the 5% significance level. The adjusted R² is relatively good, since both coefficients are statistically significant (Ozili 2023). The results for Equation (20) of the state labour market model are shown in

Table 3. Results for Equation (20) (${\displaystyle \frac{R}{GDP}}= m_3\left[{\displaystyle \frac{1}{GD{P}_{N_P}}}\left({\displaystyle \frac{m_2}{\left[\frac{\Delta N_{E,S}}{m_1}-k_3\right]}}-\left(k_1+0.1\right)\ln \left(\alpha \right)\right)\right]+k_4$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$m_3$	0.934	0.005	174.47	<2 × 10−16	0.9991
$k_4$	−78.575	6.062	−12.96	4.17 × 10−13

Source: Authors’ calculations.

.

Table 3 indicates that both model coefficients are significant within the linear model at the 5% significance level. The adjusted R² indicates a very strong fit. The results for Equation (21) of the state labour market model are shown in

Table 4. Results for Equation (21) ($B_{E, N_P}= {\displaystyle \raisebox{1ex}{$B_E$}\!\left/ \!\raisebox{-1ex}{$ N_P$}\right.} = m_4 GD{P}_{N_P}+k_5$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$m_4$	2.966 × 10−1	5.424 × 10−3	54.694	<2 × 10−16	0.9907
$k_5$	−6.723 × 102	3.055 × 102	−2.201	0.0365

Source: Authors’ calculations.

.

Table 4 indicates that both model coefficients are significant within the linear model at the 5% significance level. The adjusted R² indicates a very strong fit. The results for Equation (23) of the state labour market model are shown in

Table 5. Results for Equation (23) ($W_{E,S}=m_5 N_P\left[{\displaystyle \frac{m_3 m_4 N_{E,S}^{*}}{\left(GD{P}^{*}-k_4\right)}}+k_5\right]+k_6$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$m_5$	2.497 × 10−1	5.077 × 10−3	49.185	<2 × 10−16	0.9886
$k_6$	2.084 × 1010	7.073 × 109	2.946	0.00655

Source: Authors’ calculations.

.

Table 5 indicates that both model coefficients are significant within the linear model at the 5% significance level. The adjusted R² indicates a very strong fit.

4.1.2. Model Goodness-of-Fit and Validity

Model goodness-of-fit results for Equation (23) of the state labour market model are given in Figure 2.

Figure 2 indicates that there is a degree of non-linearity present in the model (Kozak and Piepho 2018). Circles refer to each specific year in the time period under analysis, with the line indicating the trend. This effect, however, may also be a factor of the specific period over which the analysis is performed, since the normal Q-Q and scale-location plots indicate an increased propensity for a marginal degree of non-linearity. In comparing the theoretical distribution proposed by the model with the empirical distribution, the results for the Kolmogorov–Smirnov (KS) test are given in

Table 6. Results for the KS test for Equation (23) ($W_{E,S}=m_5 N_P\left[{\displaystyle \frac{m_3 m_4 N_{E,S}^{*}}{\left(GD{P}^{*}-k_4\right)}}+k_5\right]+k_6$).

D	P
0.10345	0.9985

Source: Authors’ calculations.

.

In Table 6, both D and P statistics at the 5% significance level indicate that the theoretical and empirical distributions are drawn from the same distribution (Teegavarapu 2019). This is based on the relatively small magnitude of the D statistic, which is indicative of the largest vertical distance on the y-axis between the cumulative distribution functions of the theoretical and empirical distributions. The large P statistic indicates that the observed difference between the distributions could reasonably be attributed to random variation within the same distribution, and we fail to reject the null hypothesis. The results for the Granger causality test are given in

Table 7. Results for the Granger causality test for Equation (23) ($W_{E,S}=m_5 N_P\left[{\displaystyle \frac{m_3 m_4 N_{E,S}^{*}}{\left(GD{P}^{*}-k_4\right)}}+k_5\right]+k_6$).

F Statistic	Degrees of Freedom	P (>|F|)
10.1230	−1	0.003886

Source: Authors’ calculations.

.

In Table 7, the low p-value of 0.003886 at the 5% significance level leads to a strong rejection of the null hypothesis. Therefore, past changes in population size, state labour employment levels, and GDP are reliable predictors of future changes in state labour market wages.

4.2. Private Labour Market Model

4.2.1. Model Results

The results for Equation (25) of the private labour market model are shown in

Table 8. Results for Equation (25) ($N_{E,P}=M_1{{\displaystyle \sum }}_{t=1}^u{\left[{{\displaystyle \sum }}_{i=1}^j(C_{A,i}-C_{L,i})\right]}_t+z_1$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$M_1$	2.307 × 10−5	4.042 × 10−6	5.706	1.16 × 10−5	0.5892
$z_1$	5.891 × 106	2.972 × 105	19.822	4.49 × 10−15

Source: Authors’ calculations.

.

Table 8 indicates that both model coefficients are significant within the linear model at the 5% significance level. The adjusted R² indicates a strong fit.

The results for Equation (26) of the private labour market model are shown in

Table 9. Results for Equation (26) (${\overline{W}}_{E,P,t}=M_2\left[{\overline{W}}_{t=0} \left(1+{{\displaystyle \sum }}_{i=1}^j{I}_t\right)\left(\sqrt{(1+\Delta GD{P}_{t-1}}\right)\right]+z_2$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$M_2$	2.289 × 10−1	1.426 × 10−2	7.589	2.89 × 10−13	0.9211
$z_2$	4.515 × 104	5.950 × 103	16.053	1.90 × 10−7

Source: Authors’ calculations.

.

Table 9 indicates that only the variable model coefficient is significant within the linear model at the 5% significance level. The intercept statistic is significant at the 5% significance level. The adjusted R² indicates a strong fit. The results for Equation (27) of the private labour market model are shown in

Table 10. Results for Equation (27) ($W_{E,P}=M_3\left[M_1\left[\left(C_t^{\gamma }\right)\left(M_2+z_2\right)\right]+z_1\left[\left(M_2{I}_t^{*}\right)+z_2\right]\right]+z_3$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$M_3$	1.989	8.508 × 10−2	23.383	<2 × 10−16	0.9612
$z_2$	−4.824 × 1011	6.604 × 1010	−7.305	3.43 × 10−7

Source: Authors’ calculations.

.

Table 10 indicates that only the variable model coefficient is significant within the linear model at the 5% significance level. The intercept statistic is significant at the 5% significance level. The adjusted R² indicates a strong fit.

4.2.2. Model Goodness-of-Fit and Validity

Model goodness-of-fit results for Equation (27) of the private labour market model are given in Figure 3.

Figure 3 indicates that there is a degree of non-linearity present in the model (Kozak and Piepho 2018). Circles refer to each specific year in the time period under analysis, with the line indicating the trend. This effect, however, may also be a factor of the specific period over which the analysis is performed, since the normal Q-Q and scale-location plots indicate an increased propensity for a marginal degree of non-linearity. In comparing the theoretical distribution proposed by the model with the empirical distribution, the results for the Kolmogorov–Smirnov (KS) test are given in

Table 11. Results for the KS test for Equation (27) ($W_{E,P}=M_3\left[M_1\left[\left(C_t^{\gamma }\right)\left(M_2+z_2\right)\right]+z_1\left[\left(M_2{I}_t^{*}\right)+z_2\right]\right]+z_3$).

D	P
0.34783	0.1242

Source: Authors’ calculations.

.

In Table 11, both D and P statistics at the 5% significance level indicate that the theoretical and empirical distributions are drawn from the same distribution (Teegavarapu 2019). This is based on the relatively small magnitude of the D statistic, which is indicative of the largest vertical distance on the y-axis between the cumulative distribution functions of the theoretical and empirical distributions. The large P statistic indicates that the observed difference between the distributions could reasonably be attributed to random variation within the same distribution, and we fail to reject the null hypothesis. The results for the Granger causality test are given in

Table 12. Results for the Granger causality test for Equation (27) ($W_{E,P}=M_3\left[M_1\left[\left(C_t^{\gamma }\right)\left(M_2+z_2\right)\right]+z_1\left[\left(M_2{I}_t^{*}\right)+z_2\right]\right]+z_3$).

F Statistic	Degrees of Freedom	P (>|F|)
9.4002	−1	0.006359

Source: Authors’ calculations.

.

In Table 12, the low p-value of 0.006359 at the 5% significance level leads to a strong rejection of the null hypothesis. Therefore, past changes in working capital balances, the initial wage level in the private labour market at time t = 0, and inflation, are reliable predictors of future changes in private labour market wages.

4.3. Budget Deficit Model

4.3.1. Model Results

The results for Equation (32) of the budget deficit model are shown in

Table 13. Results for Equation (32) ($Z_1=Z_2+Z_3$).

Year	Z1	Z2	Z3	Z1 − (Z2 + Z3)	BDF Predicted (Millions)	BDF Empirical (Millions)	Variance (P – E) (Millions)
2000	10,572	8301	1967	303	6663	13,364	-6700
2001	11,232	8845	2425	−38	−856	13,600	−14,456
2002	13,420	10,282	3493	−355	−7625	10,606	−18,231
2003	13,791	10,071	3865	−145	−3385	9314	−12,699
2004	15,715	11,356	4133	226	5120	26,159	−21,038
2005	17,485	12,339	4971	176	3964	13,479	−9515
2006	20,122	13,830	6416	−124	−2722	−650	−2072
2007	22,385	15,186	8188	−989	−21,776	−25,323	3546
2008	25,319	16,477	11,381	−2538	−56,473	−30,596	−25,877
2009	29,032	16,977	12,071	−16	−355	15,137	−15,492
2010	32,518	17,639	10,307	4571	108,870	149,163	−40,293
2011	34,478	18,811	10,978	4689	113,179	131,796	−18,617
2012	37,456	20,613	12,290	4553	111,438	147,262	−35,824
2013	39,901	22,051	12,945	4904	122,095	151,670	−29,575
2014	41,200	22,549	13,418	5233	137,255	147,744	−10,489
2015	43,650	23,872	13,909	5869	156,474	145,605	10,869
2016	46,733	24,863	14,565	7306	200,187	174,640	25,547
2017	49,013	26,987	15,648	6378	174,872	161,309	13,563
2018	52,074	30,117	16,370	5587	155,280	188,476	−33,196
2019	55,038	31,181	16,649	7208	203,470	218,915	−15,445
2020	59,654	30,240	16,341	13,072	383,928	445,053	−61,125
2021	62,143	32,460	14,504	15,179	451,208	538,055	−86,847
2022	65,175	35,117	20,897	9161	270,115	325,044	−54,929

Source: Authors’ calculations.

.

The results in Table 13 indicate that the model predicts an increasing budget deficit over the model period. The distribution shape of the results obtained aligns directionally with the empirical budget deficit over this period. The model aligns strongly with empirical data during stable economic periods; however, the model diverges from the empirical distribution during economic shocks, such as the COVID-19 pandemic period of 2020–2022. The model tends to predict a lower budget deficit than empirically observed, indicating the presence of additional variation that the theoretical model does not account for.

The results for the linear regression of the empirical budget deficit versus the predicted budget deficit for Equation (32) of the budget deficit model are shown in

Table 14. Results for the linear regression of the empirical budget deficit versus the predicted budget deficit ($BD{F}_{Empirical}=M\ast BD{F}_{Predicted}+z$).

Coefficients	Estimate	Standard Error	t Value	P (>|t|)	Adjusted R2
$M$	8.865 × 10−1	2.761 × 10−2	32.102	<2 × 10−16	0.9791
$z$	−5.295 × 109	5.352 × 109	−0.989	0.334

Source: Authors’ calculations.

.

Table 14 indicates that only the variable model coefficient is significant within the linear model at the 5% significance level. This indicates that the predicted budget deficit scales almost one-to-one with the empirical budget deficit. The intercept statistic is insignificant at the 5% significance level, which indicates that the intercept term is insignificant in the linear regression. The adjusted R² indicates a strong fit.

4.3.2. Model Goodness-of-Fit and Validity

Model goodness-of-fit results for Equation (32) of the budget deficit model are given in Figure 4.

Figure 4 indicates that there is a degree of non-linearity present in the model due to the presence of right skewness as per the normal Q-Q plot (Boylan and Cho 2012; Kozak and Piepho 2018). Circles refer to each specific year in the time period under analysis, with the line indicating the trend. This effect, however, may also be a factor of the specific period over which the analysis is performed, since the normal Q-Q and scale-location plots indicate an increased propensity for a marginal degree of non-linearity.

5. Discussion

Model results for both the state labour market and the private labour market models possess adjusted R² statistics of >0.95. The variable and intercept coefficients for both models are also significant at the 5% level of significance. The budget deficit model possesses an adjusted R² statistic of >0.95. The variable coefficient is significant at the 5% level of significance. The intercept coefficient for this model is insignificant at this confidence level. These statistics indicate that these models explain more than 95% of the variance in their respective output variables and are empirically acceptable (Ozili 2023).

The results for the state labour market model suggest that the variables population size, number of state employees, and the size of the economy (GDP) are all independently positively associated with the level of state wages (given other variables are kept fixed). The total effect of the variable term is marginally positively associated with the level of state labour market wages (coefficient estimate of 2.497 × 10⁻¹). This result indicates that although there is a degree of association, a decoupling effect may exist in the state labour market, through which the level of state wages is determined in part independently of the model variable set. Several studies support these findings (Oliver and Yu 2018; Garnero 2021; Giupponi and Machin 2024), whereby state labour market wages are decoupled from the general economy due to factors such as increased rates of unionisation and collective bargaining in the public labour market (Anzia and Moe 2015). Additionally, state employees typically experience a higher degree of employment security relative to the private labour market (Kopelman and Rosen 2016).

Model goodness-of-fit tests indicate a degree of non-linearity present in the model. The residuals for the model are largely normally distributed; however, the normal Q-Q plot suggests the presence of heavy tails since the residual distribution deviates in the extremes of the distribution. This suggests the presence of a degree of skewness. The Kolmogorov–Smirnov test indicates that the theoretical and empirical distributions are drawn from the same distribution, indicating that the model prediction is empirically acceptable relative to empirical observations.

Results for the private labour market model suggest that the variables working capital balances, cumulative inflation, and the annual growth rate in GDP are all independently positively associated with the level of private labour market wages (given other variables are kept fixed). The total effect of the variable term is positively associated with the level of private labour market wages (coefficient estimate of 1.989). These results indicate that private labour market wages are generally positively associated with labour productivity, as increasing working capital balances over time requires efficient utilisation to generate sufficient output to support wage growth in the long run (Sawarni et al. 2020).

Model goodness-of-fit tests indicate a degree of non-linearity present in the model. The residuals for the model are largely normally distributed; however, the normal Q-Q plot suggests the presence of heavy tails since the residual distribution deviates in the extremes of the distribution. The residuals versus fitted values plot follows a U-shaped distribution, which suggests non-linearity present that is not captured in the existing model. The Kolmogorov–Smirnov test indicates that the theoretical and empirical distributions are drawn from the same distribution. This indicates that the model prediction is empirically acceptable relative to empirical observations.

The labour productivity and wage growth models suggest the presence of a decoupling effect from labour productivity in the state labour market, which forms a large proportion of the formal labour market. The decoupling of wages and labour productivity observed is inconsistent with the wider literature regarding the relationship between these two variables, which, in general, are found to be positive (Nikulin 2015). However, Gil-Alana and Skare (2018) found evidence of decoupling between labour productivity and wages in several developed economies, with the level of the decoupling being the highest in the United States and Japan and the lowest in Norway and Germany. They also found that the decoupling effect is shared by all countries in the sample and that the intensity of the effect varies by individual country.

However, in the case of South Africa, the literature indicates that the relationship between wages and labour productivity is substantially weaker compared to other economies. Despite Tsoku and Matarise (2014) finding no evidence of causality between labour productivity and wages in South Africa, Mawejje and Okumu (2018) found a positive association between productivity and wages for a panel of 39 African countries. They found that education and sex factors are leading determinants of wage levels and wage growth through the mechanism of employment opportunities and progression. These factors are strong contributors to the weaker relationship between wages and labour productivity in South Africa since the composition of the labour market is skewed towards females, and the labour force at large possesses relatively low skills (Posel and Casale 2019). Where employment opportunities do exist, these are largely concentrated within the informal labour market from a volume perspective, combined with a low wage base (Gradín 2021b).

The structure of the South African labour market is thus conducive to supporting a decoupling effect between labour productivity and wages (Klein 2012). Unemployment and low skills development are concentrated in the Black South African population, which constitutes most of the labour force. These factors, combined with little bargaining power across the informal employment sector, lead to a higher likelihood of depressed wage growth for the total labour market (Festus et al. 2016; Jain 2019). The large degree of income inequality in South Africa supports the results obtained in that there is a bifurcated income growth mechanism present in the labour market, which hampers upward mobility and provides only marginal economic stability for most in the labour force (Adato et al. 2013). The effect observed in the decomposition of the relationship between labour productivity and wage growth can be ascribed to younger workers being systematically compensated below their productivity levels and the labour share declining as a contributor to national income through increasing supplantation of income accrual and concentration to the higher end of the income distribution (Van Biesebroeck 2014). This effect yields an increasing rate of income inequality.

The decomposition between wages and labour productivity is also driven by structural factors specific to the general economy’s composition (Schröder 2020). Labour productivity and wage growth correlations can vary dramatically between different economic sectors, from 0.846 for the construction industry to −0.125 for social services (Klein 2012). Additional factors that support labour productivity and wage decoupling include increased technology-driven labour productivity and recent policy direction related to minimum wages, inflation, interest rates, economic growth, and unemployment (Gil-Alana and Skare 2018; Compagnucci et al. 2021; Nasir et al. 2022). These factors present considerable challenges for businesses to operate with confidence, especially given the track record of poor governance and support for total structural economic reform in South Africa to support the economic inclusion of the largely low-skilled, unemployed Black majority (Mlambo and Masuku 2020; Mantzaris and Pillay 2019). The degree of economic growth experienced over the past two decades also does not necessarily contribute to more employment opportunities as growth increases; however, higher economic growth does contribute to higher overall wages (Bassier and Woolard 2021). To stimulate greater labour absorption and consequently decrease demand for social income, investment-enhancing policies, such as low interest rates and a favourable economic environment, should be put in place to accelerate growth (Meyer and Sanusi 2019).

These results infer that despite an increase in labour productivity, the South African economy will not necessarily experience a rise in overall wage growth, which is affected largely by the sheer scale of income inequality and consequently has little impact on reducing social welfare expenditure. In addition, this result supports the outcome that the South African economy is concentrated largely in economic sectors with low correlations between labour productivity and wage growth. Simply expanding these economic sectors without the necessary structural and policy reforms will yield little change in wage growth at the overall economy level. With respect to fiscal and taxation policies, the decoupling of labour productivity and wages further indicates that simply targeting economic growth and increasing tax revenue will not necessarily offset social welfare dependency growth in South Africa. The decoupling effect ensures that any specific increase in productivity is not necessarily rewarded with an increase in wages, which reinforces inequality and prevents a decline in social welfare requirements. Increased social welfare expenditure has limited capability to improve the quality of life of South Africans, with limited capability to reduce poverty in the long run. Inefficient strategies to transform infrastructure development and quality of education are strong contributors to this inability to reduce poverty and inequality (Sidek 2021).

The net effect of these factors on South Africa’s budget deficit indicates that, unless there are specific policy interventions to change the fiscal and economic structure of South Africa, the developed model predicts that the budget deficit could potentially increase by 66% by 2035. The combined effects of persistently high unemployment, little change in the effect of a decoupled local labour market, and the necessary adjustments for cost-of-living increases on social welfare grants over time ensure a growing social welfare expenditure budget. To meet this growing demand for social welfare income, an increase in such expenditure will rely on either cannibalisation in the national budget from other expenditure items, such as healthcare, cost, or infrastructure, or higher taxes on personal income, corporate, or value-added goods, or a combination of the two. These results are consistent with those obtained in the literature, which further highlight that an increase in tax revenue driven by a larger income-generating labour force, consumption economics afforded through greater income inequality, and improved business and investor sentiment are all factors the South African government should prioritise to reduce the risk of incurring escalating budget deficits (Murwirapachena et al. 2013; Redda 2020; Banday and Aneja 2022).

In countries with similar economic structures, such as Brazil, China, India, and Nigeria, social welfare and redistribution policies have had mixed effects on inequality and poverty reduction, despite considerable budget expenditure allocations (Maiorano and Manor 2017; Ichoku and Anuku 2019). Several studies on the Brazilian economy’s fiscal policy found that despite significant growth in social welfare expenditure, redistributive policies disintegrated as the distributive conflict between labour and capital intensified amidst declining tax revenues associated with economic slowdown (Dweck and Teixeira 2017; Martins and Rugitsky 2021). This directly contributed to further increases in net social expenditure, with a net decline in social transfers at the individual level since austerity measures have been adopted (Marquetti et al. 2020). The impact on the poor has led to indirect taxes often exceeding direct and indirect grants and subsidies received through social welfare programmes (Higgins and Pereira 2014).

Padhan et al. (2022) investigated the impact of taxation and government expenditure on income inequality in India. They suggest that economic growth, globalisation, and urbanisation indirectly promote income inequality in the Indian economy. This effect arises due to the exclusion of low-skilled labourers during economic booms since firms are focused on hiring skilled labour, thus reducing low-skilled labour demand and absorption and increasing the reliance on social redistribution. Like South Africa, India also possesses a large proportion of low-skilled labour. The promotion of skills development in the general population is proposed as a contributor to a more equalised income distribution outcome (Sehrawat and Singh 2019).

There exists a strong consensus in the literature that the redistributive impact of taxes is not wholly effective in reducing income inequality as deeply as theoretically proposed, particularly in the case of developing nations (Martinez-Vazquez et al. 2012; Padhan et al. 2022). Policies targeting income inequality require the government to consider both expenditure and sources of taxation in tandem before framing any policies related to income distribution (Padhan et al. 2022). Fiscal policy must consider both long-run economic growth objectives and short-term output fluctuations (Halkos and Paizanos 2016).

These implications directly support the notion of structural transformation that is necessary in the South African economy along two time horizons. In the short and medium term, the government should prioritise the absorption of labour in industries that require low skills and that possess relatively low barriers to entry with strong market demand (Islam 2017). This will require investment in industrialisation of the country, with a specific focus on aligning the development of such industries to easily accessible points of distribution, whilst considering the geospatial distribution of labour across the country. Such developments, termed special economic zones, have led to significant increases in local economic growth, employment, and income where these have been developed in line with longer-term economic objectives aligned to global labour demand trends (Wang 2013). Since the South African government faces limited capability to ensure such investment, it is necessary to attract foreign investment on a much larger scale. This will require policies that support investor confidence and that can position South Africa as an attractive investment destination. Such policies will require careful consideration to ensure a fair trade-off between attractive taxation rates on income earned from foreign investments, flexibility on certain labour policies, and the net economic effect on the community of such special economic zones over a specified time period.

In the long term, prioritisation should focus on the next wave of technological and industrial advancement, with a particular focus on developing a large, highly skilled labour force that can compete globally to meet the labour demand of growing industries. This does not necessarily imply an additional investment at the national budget level for education since the government already spends a considerable proportion on education (USAID 2024). To achieve this objective, the government must prioritise much stronger alignment between the skills required in the future economy and reallocate funding and investment into such degrees and vocational programmes on a much greater scale (Allais 2022). Youth must be incentivised to enter professions that support the long-term objectives of the country to specialise and provide expertise in future-proofed economic sectors.

Additional areas of future research should consider model sensitivities related to non-linear effects, which are not accounted for in this study. These may induce additional effects that should be considered beyond the inferences from this study. The effects of economic emigration and immigration factors are additional areas of focus that should be considered from both tax and expenditure perspectives, which both affect the budget deficit. The mechanism through which labour productivity and wages decouple is of interest in the wider literature, with additional considerations on which factors affect the relationship (Škare and Škare 2017). Such factors include capital investment, education, labour market formalisation and organisation, the labour force’s age and geographical distributions, types of employment in the labour market, and technology and innovation factors (Islam et al. 2015; Fatuła 2018; Katovich and Maia 2018). Understanding how such additional factors affect and reinforce this negative feedback loop in the South African labour market can provide additional opportunities for potential remediation. Additionally, the period under analysis may be widened for future studies.

6. Conclusions

This study developed and empirically tested a theoretical model linking labour productivity, market wages, and social welfare expenditure (SWE) in a national budget, with a specific focus on how these different factors contribute either to a budget deficit or surplus. The study also aimed to investigate the implications of the primary objective on fiscal and tax policies adopted by South Africa in recent years.

The core empirical finding reveals a dual-sector wage–productivity dynamic. Whilst the private sector adheres to economic fundamentals, linking remuneration positively to productivity, working capital, and inflation, the public sector exhibits a significant decoupling. The results indicate that a 1% increase in labour productivity and GDP output is associated with a 0.4% increase in private labour market wages.

State labour wages are driven predominantly by institutional factors rather than efficiency gains, creating an economically rigid wage bill that acts as a structural drag on the fiscus. This rigidity not only inflates government expenditure but also sets a costly, non-market anchor for wages across the broader economy. Additionally, our findings unequivocally confirm that the fiscal crisis is structural, not cyclical, rooted in a dysfunctional and dualistic labour market that fails to generate the necessary revenue base to fund its social commitments. The results indicate that a 1% increase in labour productivity and GDP output is associated with a 0.8% increase in state labour market wages.

These results indicate that simply boosting aggregate labour productivity is an insufficient solution. Due to deep structural inequality, weak labour absorption capacity, and a pervasive skills mismatch, general labour productivity increases do not necessarily translate into the broad-based, high-wage employment necessary to reduce the population’s reliance on SWE. The current policy trajectory locks South Africa into a vicious cycle, characterised by a rigid public wage bill that consumes revenue, structural unemployment that drives up SWE, and the resulting budget deficit crowding out productive investment needed to break the cycle.

To address the persistent budget deficit and mitigate rising welfare dependency, this study highlights the need for specific and targeted structural policy interventions across three key focus areas.

Firstly, policy must aggressively pivot from general growth targets to a strategy of job-rich industrial expansion. The promotion of labour in economic sectors capable of absorbing low- and mid-skilled labour at scale, such as infrastructure development and manufacturing, can effectively integrate the structurally unemployed into the productive economy.

Secondly, an overhaul of the education and vocational training system is essential to resolve the skills mismatch. Curricula and tertiary offerings must be structurally aligned with the specific, forward-looking demands of high-growth private sector industries to ensure productivity gains are matched with a relevant supply of human capital.

Thirdly, new fiscal rules must be implemented to manage the state labour wage bill, ensuring its growth is constrained to a rate demonstrably below national labour productivity growth. The fiscal space created by this restraint must then be ring-fenced and directed toward productive public capital investment programmes, focused on infrastructure, energy, and logistics, which yield the highest economic multiplier and long-term employment benefits across the economy.

In summary, solving South Africa’s budget deficit requires recognising that it is the product of a systematically broken economic structure. The future stability of the fiscus hinges not on austerity measures alone, but on a decisive commitment to structural reform that re-links productivity, wages, and employment to create a self-sustaining, high labour absorption economy that promotes expenditure from SWE to investment, and a consequent promotion of the quality of life of citizens. Future research should focus on simulating the precise multiplier effects of sector-specific industrial policies on employment absorption rates across different skill quartiles.