The data are from secondary sources, including Statistics South Africa, the Organization for Economic Cooperation and Development (OECD) and the South African Reserve Bank (SARB). Data extraction is quarterly, from 1998 to 2020. The selected period and sources are based on data availability.

The study employs the fiscal policy theoretical framework of Musgrave (1959). The framework advances how policymakers are conscious that macroeconomic indicators (y_i) like employment, economic growth, inflation and poverty can be impacted by variations in policy instruments (x_j) (Gachari & Korir 2020; Tendengu, Tsegaye & Kapingura 2022). The general functional form is (Equation 1):

Additionally, Gachari and Korir (2020:24) stipulate that should a minimal or realistic change in a policy instrument result in significant alterations to a macroeconomic indicator that instrumental variable is efficient concerning that indicator. Fiscal variables in the form of deficits are indirect, while taxes and spending signify direct forms of control. Analysis can also expand, to include nonfiscal instruments, such as interest rates, exchange rates and capital formation (Tendengu et al. 2022:4). This allows for the Musgrave (1959) theoretical foundation to incorporate nonfiscal policy impacts, such as monetary policy, on macroeconomic indicators (Gachari & Korir 2020).

The executive and legislature are responsible for conveying fiscal policy, which is affected by the political scene, while central banking authorities conduct monetary policy independently. Both policies are essential and strategic elements that can influence production levels. This mandates proper coordination between fiscal and monetary policy authorities to achieve desired economic outcomes. Given theoretical and empirical considerations, the study modifies the functional models of Idris (2019). Research by Idris examined the relative effectiveness of stabilisation policy impacts on Nigerian production efforts. The study by Idris underpinned the modelling of policy impacts in other developing nations, such as Ethiopia (Woldsemayat 2020) and the ASEAN-5 (Indonesia, Malaysia, Philippines, Thailand and Vietnam) (Samsuddin & Selvia 2021), thus highlighting its relevance to the present research.

The linear model depicting the combined, collaborative impact of fiscal and monetary policy on manufacturing industry performance in South Africa, as premised in the work of Idris (2019) is specified as (Equation 2):

where aggregate manufacturing production, or production within industry i of the manufacturing sector, (MANO_i), is dependent on total tax revenue (TTR), total government expenditure (TGE), deficit financing (DF), lending rates (LR), the nominal effective exchange rate (NEER) and money supply (M3). β₀, β₁ … β_n indicate the parameters to be estimated and v_it the error term at time t.

Research relating to the effect of oil prices on manufacturing activities is also well-documented (Gummi, Hassan & Mu’azu 2018; Karadag 2021). Oil serves as a primary input in industrial production. The inclusion of Brent crude oil prices (OIL) and log terms for significantly sized variables modifies Equation 2 (see Equation 3):

No universal criterion is sufficient to define manufacturing (MANO_i). Broadly, it is the process through which products are altered, renovated or reconstructed. Manufacturing indices (2015 = 100) of physical production volumes are used. Total manufacturing production is used as a baseline against which selected industries within the sector can be compared. Selected industries within the sector are those considered to hold a significant share of aggregate manufacturing production in South Africa. These industries, and their contributions to total manufacturing (IDC 2020b), are: (1) metals, metal products, machinery and equipment (22.5%); (2) food, beverage and tobacco (21.4%); (3) petroleum products, chemicals, rubber and plastic (14%) and (4) wood and paper, publishing and printing (10.4%).

Fiscal policy instruments incorporate tax revenue, government expenditure and deficit financing. The TTR dataset (in Rmilllion’s) includes revenue from all three South African sources, personal income, company and value-added tax (VAT). Consumers are becoming increasingly price sensitive; the Nielsen State of the Retail Nation Report for the first quarter of 2024 revealed South Africa to be one of the most price-sensitive countries in the world (NielsenIQ 2024). Literature dictates that increased TTR reduces output (Ahumada & Villarreal 2023). Total government expenditure (measured as a percentage of GDP) is synonymous with public consumption and investment spending (Fourie & Burger 2015:397), with a positive relation to production expected. Production should be increased via raised government spending by boosting private investment (Osinowo 2015). Deficit financing (measured as a percentage of GDP) relates to generating funds and financing deficits resulting from excessive expenditures over given revenues. Surveys in the sector noted that more than 50% of respondents have increased debt levels, in both short- and long-term finance. Given domestic economic conditions, this is likely the result of strained cash flows (Manufacturing Circle 2016). According to Tkalec and Vizek (2010), DF may crowd out production.

Monetary variables in this context include interest rates, exchange rates and M3. Nominal interest (in percentage terms) signifies the rand value borrowers are obligated to pay, or lenders are entitled to receive for monies lent (Parkin 2020). Nominal interest in the form of the repurchase rate is relevant for the present study, given that all other variables in the model are expressed in nominal form. Literature dictates that the impact of LR on output is negative (Agbonrofo & Olusegun 2023; Kutu & Ngalawa 2016a). Interest rate tightening discourages production. Modifications in rates cause variations in investment and consumption, influencing output and prices (Kutu & Ngalawa 2016a). The NEER is an unadjusted trade-weighted index calculated using nominal rates; it is the rate at which one country’s currency exchanges for a basket of currencies (Appleyard, Field & Cobb 2017).

Research by Loweald (2020) identified a strong link between exchange rate depreciations and manufactured imports in South Africa, solidifying the need for incorporation of the exchange rate into the present analysis. The impact of NEER on manufacturing may vary (Sahoo et al. 2024). The sign of the relation depends on industrial technological intensity; depreciations in low-tech industries invigorate production efforts (Tkalec & Vizek 2010). The NEER dataset used is in indexed form (2015=100). For M3 (measured as a percentage change), the preferred monetary aggregate of SARB is M3. This is the single most comprehensive measure of money in South Africa (Mahadea et al. 2017). The impact of M3 on output is generally positive in the literature (Omolade & Ngalawa 2016; Matola 2023). There has also been extensive research on the impact Brent crude oil prices (OIL) have on manufacturing (Gummi et al. 2018; Karadag 2021). South Africa is an oil-importing nation; hence, growth tends to be vulnerable to oil prices. Brent crude oil is measured in Rands per barrel.

Should there be any doubt concerning the presence of a trend, formal stationarity techniques by the likes of ADF and PP can be employed (Stock & Watson 2015:603). Null hypotheses for ADF testing relate to a time series containing unit roots (δ = 0), which implies nonstationarity and leads to spurious regressions. Alternatively, (δ < 0) implies that a series is in actuality stationary (Stock & Watson 2015:603). While ADF enables the correction of higher-order serial correlation via the addition of lagged differenced terms on the right-hand side of an equation, PP tests correct the t-statistic of the coefficient (γ) in accounting for serial correlation present in the error term (e_t) (Asteriou & Hall 2016:357). To confirm the order of integration, this study uses both ADF and PP.

After stationarity is confirmed, various techniques can be used in cointegration testing. The ARDL procedure, however, has several advantages over Engle-Granger and Johansen (Kripfgranz & Schneider 2023). ARDL does not require all variables included in a model to be of the same integration order; it is more efficient in handling small and finite sample data sizes, and unbiased estimates of long-run models are obtainable (Kripfganz & Schneider 2023:983–994). After determining the appropriate lags, long- and short-run estimates of the ARDL model are estimated. Based on Equation (3), the long-run ARDL model for examining the combined impact of fiscal and monetary policy on manufacturing sector production at the aggregate and industry levels in South Africa is given by (Equation 4):

The ARDL bounds test is premised on the joint F-statistic. The bounds test has an asymptotic nonstandard distribution under the null hypothesis of no cointegration. Should there be evidence of a cointegrating vector, the ARDL model may be reparametrised into an ECM to acquire the short-run dynamics connected with long-run estimates (Kripfganz & Schneider 2023:984). The derivation of short-run elasticities, as they pertain to the combined model for fiscal and monetary policy impacts, results in the formulation of the ECM (see Equation 5):

Diagnostic testing based on residuals allows for checking the adequacy of functional form and signals the validity of model assumptions. These assumptions include no serial correlation, normality and homoscedasticity. Following the estimation of ECM, stability testing in the form of CUSUM and CUSUM of squares is conducted for parameter stability (Asteriou & Hall 2016).

This article does not contain any studies involving human participants performed by any of the authors. Ethical clearance was obtained from the University of Fort Hare’s Inter-Faculty Research Ethics Committee (IFREC) on 26 March 2018.

Formal stationarity testing is completed at levels, using ADF and PP with intercept, trend and intercept and no intercept or trend (Table 1). LTOTAL is stationary at the 10% significance level for intercept only under the ADF testing procedure. Regarding the PP results, LTOTAL illustrates stationarity at the 1% level of significance for intercept and trend and intercept and nonstationarity for none. Noticeably, TGE is stationary at the 1% level under intercept for ADF and for varying levels for intercept, trend and intercept and none for PP. Deficit financing is nonstationary according to ADF but stationary under PP at the 1% level. Alternatively, LR is stationary under ADF at the 1% significance level but is nonstationary under PP, except for none, where it is stationary at 10%.

Following differencing, formal testing shows that all variables are stationary for intercept, trend and intercept, and none for the ADF and PP techniques (Table 2). Estimation using ARDL is appropriate for cointegration, given that variables are integrated of mixed orders I(0) and I(1). No variables included are integrated with orders greater than I(1). The ARDL estimation produces consistent estimates of the long-run coefficients that are asymptotically normal irrespective of whether the underlying regressors are integrated of order (I[1]) only, integrated of order zero (I[0]) only or a mixture of I(1) and I(0) (Kripfganz & Schneider 2023:985).

The study uses the Akaike information criterion (AIC) for determining the lag structures, for the LTOTAL, LMETAL, LFOOD, LCHEM and LWOOD policy models (Table 3). According to Asteriou and Hall (2016:73), models that adopt the lowest AIC values are the best fit.

Computed F-statistic values for joint fiscal and monetary models for LTOTAL and all selected industries concerned exceed the upper critical values at the 5% level of Pesaran for bounds testing (Table 4). This implies a rejection of the null hypothesis of no cointegration for LTOTAL and all manufacturing industry subcategories, providing evidence of long-term relationships among variables.

Long-run ARDL regressions are then estimated (Table 5). Long-run empirical results indicate positive, significant impacts of LTTR on LTOTAL, LFOOD and LWOOD. The positive coefficient associated with LTTR does not conform with a priori expectations but is aligned with the findings of Stoilov (2017) and Islam (2019). Higher taxes may suppose greater levels of public expenditure, some of which could foster increased growth. This is prevalent with personal income taxes and social contributions, which in South Africa constitute the largest and most significant source of tax revenue (Mohr 2020) and could explain this positive link. In contrast, LTTR is negative and significant for LMETAL, conforming to a priori expectations. Higher national tax revenue (TTR) is expected to lead to decreased output growth within manufacturing (Bakare-Aremu & Osobase 2015). For LCHEM, the fiscal LTTR variable is insignificant.

The linkage of TGE with LTOTAL, LMETAL and LWOOD is significant and of a negative nature. The relationship between TGE and manufacturing output is expected to be positive (Aghioin, Hemous & Kharroubi 2014; Osinowo 2015). The negative signage could be because of crowding out (Fourie & Burger 2015; Tkalec & Vizek 2010). Insignificance in the LFOOD model concurs with prior research; there were findings of insignificance of TGE for the food and beverage industry in Croatia (Tkalec & Vizek 2010), countries that were part of the European Union before the 2004 expansion (EU15) and New Member States that joined the EU at a later date (NMS10) (Zislin & Barrett 2009). For LCHEM, a significant positive link is observed with TGE conforming to expectations (Aghioin et al. 2014; Osinowo 2015). Positive relations were identified in the chemicals industry in Croatia (Tkalec & Vizek 2010) and Japan (Zislin & Barrett 2009).

Concerning DF, there are findings of a significant, positive link for LCHEM. The positive signage does not conform to a priori expectations or industry research on NMS10 countries (Zislin & Barrett 2009). DF is expected to crowd out manufacturing production (Tkalec & Vizek 2010). DF is insignificant for LTOTAL, LMETAL, LFOOD and LWOOD in the long run. Insignificance in LMETALS concurs with the findings for the basic metals sector in Croatia, Japan and the United States (Tkalec & Vizek 2010; Zislin & Barrett 2009).

In terms of LR, the relation with manufacturing output is expected to be negative (Agbonrofo & Olusegun 2023; Kutu & Ngalawa 2016a). Empirical results for LTOTAL, LMETAL and LWOOD conform to significant negative relations. This aligns with findings on the US metals industry. The metals industry has strong linkages with construction activities, which are also strongly associated with residential investment. Residential investments are perceived to be highly sensitive to interest rate changes (Ahmad & Rangaraju 2020). Lending rates is insignificant for LFOOD and LCHEM. However, the insignificance of LR for LFOOD industries concur with the findings of prior research on the US food and beverages industry. Nondurable industries often display delayed or very mild responses towards monetary policy variations (Ahmad & Rangaraju 2020).

With LNEER, results for LMETAL find the variable to be insignificant. The responses of LTOTAL, LFOOD and LWOOD are significant and positive. Contrastingly, negative signage was observed in the context of the food and beverage industry for NMS10 nations within the EU (Zislin & Barrett 2009). Results for LWOOD conform to the significant, positive relation from prior research on the wood production industry in Croatia (Tkalec & Vizek 2010) but not with research in the United States (Thorbecke 2018). Concerning LCHEM, LNEER is significant at the 5% level but negative. The impact of exchange rates on manufacturing performance could vary (Sahoo et al. 2024). A significant positive linkage indicates that an appreciating exchange rate positively affects domestic manufacturing production levels.

In the long-run relationship between LTOTAL, LMETAL and M3, the positive significant coefficients concur with a priori expectations (Matola 2023; Omolade & Ngalawa 2016). However, LFOOD and LCHEM have a negative link with M3. This could suggest M3 is not being efficiently utilised in managing monetary variables in the long run and that the form of credit extended towards these industries is not conducive towards productivity (Aiyedogbon, Ohwofasa & Anyanwu 2015). Money supply is insignificant in the LWOOD industry. This is indicative that M3 occupies no significant role in LWOOD production.

The coefficient for LOIL is negative and significant at the 1% and 10% levels for the long-run LFOOD and LWOOD models, respectively. In the South African LCHEM industry, there is a significant positive link. The LOIL variable is relevant in explaining LCHEM at the 5% level. A negative relationship is expected between production and LOIL (Sarmah & Bal 2021). Findings thus contrast a priori expectations. Concerning LTOTAL, LOIL is insignificant. Following long-run estimations, the ECM models are obtained to confirm speed of convergence (Table 6).

Findings reveal that the coefficient of the ECM is negative and statistically significant at the 1% level for LTOTAL and all four industries within the sector, LMETAL, LFOOD, LCHEM and LWOOD, confirming the adjustment procedure over the short term. It indicates a correction of 92.01% (LTOTAL), 95.8% (LMETAL), 90.4% (LFOOD), 80% (LCHEM) and 88.9% (LWOOD) of uncertainty from the preceding period within the next quarter.

Concerning LMETAL, all variables except LOIL are significant. The LTTR coefficient is positive in the short run; this does not conform to a priori expectations or the long run but does align with Stoilov (2017) and Islam (2019). The positive coefficient of TGE also does not conform to long run findings but is per a priori expectations (Aghioin et al. 2014; Osinowo 2015). Unlike long run estimations, DF is significant and aligns with a priori expectations. A negative link is observed (Tkalec & Vizek 2010). The positive significant relation between LR and LMETAL in the short run conforms to neither long run findings nor a priori expectations.

The positive relations observed for LNEER with LMETAL concurs with long run estimations and indicate that an appreciating exchange rate positively affects local manufacturing production in the short term. Similarly to the short run findings of the model for LTOTAL, the negative link observed between LMETAL and M3 could suggest that M3 is not being efficiently utilised in managing monetary variables in the short run, and that the form of credit extended towards LMETAL is not conducive towards productivity (Aiyedogbon et al. 2015 & Anyanwu 2015). LOIL is both positive and insignificant in the short term, thus mirroring long run findings.

differentiated term the case of the LFOOD, LTTR, DF and LNEER are significant in the short run. The LTTR coefficient is negative in the short run; this does not conform to the long run but is in alignment with a priori expectations (Bakare-Aremu & Osobase 2015). Unlike long run estimations, DF is significant in the short run but does not conform to a priori expectations with a positive link being observed. The negative linkage observed for LNEER does not concur with long run estimations but indicates that an appreciating exchange rate, in the short run, negatively affects manufacturing.

Concerning LCHEM, LTTR, TGE and DF are significant in the short term. The coefficient of LTTR is negative; this does not conform to long run estimations but is in alignment with a priori expectations, as well as short run findings observed for LTOTAL, LMETAL and LFOOD. Unlike long run estimations, TGE is negative; this does not conform to a priori expectations but could be explained as the crowding-out effect, with attempts to stimulate economic activity via increased public spending harming private spending (Fourie & Burger 2015). The sign of the relationship for DF does conform to a priori expectations but contrasts long run estimations; DF is expected to crowd out production (Tkalec & Vizek 2010).

For LWOOD, TGE and DF are significant, positive fiscal variables in the short run. The positive links do not conform to long run estimations. In terms of a priori expectations, however, the relationship between TGE and manufacturing output is expected to be of a positive nature (Osinowo 2015). A negative relation is predicted in that DF will crowd out manufacturing production. However, in the domestic context, positive relations have been empirically observed (Ocran 2011). In addition, the signs for LR and M3 do not conform to either a priori expectations or long run estimations but interestingly concur with short run findings for LMETAL.

Diagnostic tests include those examining normality (Jarque-Bera), heteroscedasticity (Breusch-Pagan-Godfrey) and serial correlation (Lagrange multiplier). The p-values for all tests exceed the 5% level for LTOTAL and the selected industry policy models (Table 7). The null hypotheses of normally distributed residuals, homoscedasticity and no serial correlation are accepted.

Visual representations of CUSUM and CUSUMQ for LTOTAL and all industry subcategories are provided (Figure 1 to Figure 5). The null cannot be rejected where statistics remain within the critical bounds of 5%. All plots for CUSUM and CUSUMQ stay within the critical bounds for LTOTAL, LMETAL, LFOOD, LCHEM and LWOOD, confirming the stability of long-run regressor coefficients.