The paper used net savings of corporations and households rather than their total deposits (Table 2). Dadam and Viegi (2018) used total deposits of the corporate sector as a measure of precautionary savings, but the paper argues that total deposits overestimate the level of savings as they do not take into account the liabilities of the sector. Thus, using net savings moves us closer to the underlying relationship that has to be examined. Further, existing literature that looks at the fiscal deficit-private savings nexus tends to use corporate savings data while excluding household savings data (Aggarwal et al. 2022; Dadam & Viegi 2018). Surely, household savings are a significant component in the savings story that has to be captured in the modelling. The paper accounts for these matters in attempting to model the relationship between fiscal deficits and private savings.

Table 2 describes some of the control variables used in the study. Fiscal deficit is the nominal overall fiscal deficit defined as total revenue minus total government spending including debt servicing costs expressed as a percentage of GDP. The paper also uses the nominal interest rate defined as the real interest rate plus the inflation rate. Thus, the nominal interest rate captures the effect of both the real interest rate and inflation. As South Africa is organised around a minerals-energy industrial complex, the exchange rate plays a crucial role in corporate incomes and thus, savings. Finally, the paper controls for household disposable income per capita that captures both the ability of households to save but also to express effective demand, which influences corporate incomes and savings.

After unit root tests (Table 5) confirmed that the variables were integrated of different orders, a Toda-Yamamoto Vector Autoregressive (TY-VAR) model was developed so as to retain test statistics, ‘t’ and ‘F’, that follow standard distributions (Toda & Yamamoto 1995). Toda and Yamamoto (1995) have shown that when series in a system of equations are integrated of different orders, the ‘t’ and ‘F’ statistics follow non-standard distributions. To correct that problem, they devised an estimation procedure that essentially overfits the VAR. They proposed using the maximum order of integration in the system to augment the lag length in such a manner that the lag order of the VAR becomes q = p + d_max, p being the optimal lag order and d_max being the maximum order of integration in the system. The non-causality test is still based on the optimal lag structure, but the estimated model has a lag order q. The choice of the VAR as a modelling framework arises from the observation that all the variables controlled for in the article potentially have reverse causal influences on each other theoretically.

The TY-VAR models that were estimated Equation 1 for net household savings and Equation 2 for the net corporate savings were: Ht=[hst,dt,it,et,yt][Eqn 1] Ct=[cst,dt,it,et,yt][Eqn 2]

In Equation 1, hs_t,d_t,i_t,e_t,y_t represent net household savings, deficit, nominal interest rate, the log of the nominal exchange rate and the log of household disposable income per capita, respectively. In Equation 2, cs_t,d_t,i_t,e_t,y_t represent net corporate savings, deficit, nominal interest, the log of nominal exchange rate and the log of household disposable income per capita.

The TY-VAR system for net household savings was represented as: Ht=α+∑i=1n−1∏iHt−i+∑j=ndmax∏jHt−i  +∑k=1mβkDumkt+εt[Eqn 3]

In Equation 3, α is a constant, Π_i is an n × n matrix of coefficients, ɛ_t ≈ iid(0, Ω), dmax is the maximum order of integration and Dum_kt are dummy variables.

The TY-VAR system for net corporate savings was represented as: Ct=δ+∑i=1n−1∏iCt−i+∑j=ndmax∏jCt−i  +∑k=1mβkDumkt+ωt[Eqn 4]

In Equation 3, δ is a constant, Π_i is an n × n matrix of coefficients, ω_t ≈ iid(0, Ω), dmax is the maximum order of integration and Dum_kt are dummy variables.

The article controlled for a number of dummy variables that potentially had effects on both corporate and household savings through economic growth and altering expectations and thus, saving behaviour. The base period was 1960–1966. The article controlled for the divestment dummy, which captures the protracted campaigns against investment in apartheid South Africa, which ran from 1967 to about 1987. By creating uncertainty, most likely this campaign caused corporate savings to increase as firms reduced their investment rate but reduced household savings through poor economic growth. Similarly, the sanctions of the period 1985–1993 would have resulted in rising corporate savings but perhaps undermined household savings as disposable incomes fell.

The article also controlled for periods of significant commodity price movements such as the falling commodity prices of the 1980s, rising commodity prices of the 2000s and the oil crisis of the 1973–1974 period. As the South African economy is built around a minerals-energy complex, one would expect a commodity price bust to result in job losses, decreasing household incomes and so, falling household savings, and the story would be more or less the same for corporations as their earnings shrink significantly. The converse is true for the times of commodity price booms.

The controlled for the democratic transition that is expected to have induced greater household participation in the economy. Increased access to credit because financial liberalisation that followed this transformation would have reduced household savings and increased reliance on debt. Firms would have saved less as they invested more in pursuit of new business opportunities that the demise of the apartheid signalled. A number of financial crisis dummies were controlled, which include the rand crises of 1996–1998 and 2001, the Asian financial crisis and the global financial (economic) recession. The rand crisis created uncertainty and probably reduced household savings, but increased net corporate savings as firms would have adopted a wait-and-see approach to investment. However, the effect of the exchange rate on corporate savings behaviour is often ambiguous because it tends to benefit exporters while penalising importers; the net effect depends on which of the two responds more to the depreciating rand. The Dotcom bubble was a period of good economic performance and is expected to have increased household savings because of rising incomes but reduced corporate savings as firms undertook more investment activities. The COVID-19 pandemic is expected to have increased corporate savings because of the postponement of investment decisions, while household savings would have declined because of household loss of incomes. Further, the article controlled for the inflation targeting framework dummy and the adoption of the MTEF to capture changes in both the monetary and fiscal regimes.

The Schwarz Information Criterion (SIC) and the Hannan-Quinn information criterion (HQC) determined an optimal lag order of one. The Akaike Information Criterion (AIC) determined three lags but given the size of the sample, the article chose to rely on the SIC and the HQC. The sample period was 1960–2021, and the data had annual frequency. This period was chosen simply for data availability reasons.

The article presents the analysis in terms of impulse response analysis and as well as Granger non-causality test. The Cholesky ordering for the impulse response analysis was informed by the heterodox postulation that government spends first before taxing (Cesaratto 2016). This implies that the deficit has an underlying causal effect on the other variables. The neoclassical view is that interest rates increase because of deficits (Bernheim 1989), but the heterodox view is that interest rates decrease in response (Kelton 2020). As South Africa is an open economy, the exchange rate responds to the movement in interest rates. The combination of these impulses influences household incomes via the performance of the economy through investment and economic growth: dt⇒it⇒et⇒yt⇒hst[Eqn 5] dt⇒it⇒et⇒yt⇒cst[Eqn 6]

This article followed all ethical standards for research without direct contact with human or animal subjects.

In general, Table 3 shows that corporations saved more than households, but the variability of the savings was more or less the same for both firms and households. The statistical ranges were very large for the private sector and more or less matched the range for the fiscal deficit. Table 3 shows that net household savings were lower in the post-democracy period than the apartheid period – averaging 0% and 3% of GDP, respectively. The maximum levels of net household savings for the two sub-periods were different, averaging 5% (1960–1993) and 2% (1994–2021), suggesting the perennial problem of household indebtedness in the post-democracy period.

Net corporate savings behaved the same way across the two sub-periods, averaging 4%. The maximum net corporate savings were nearly the same – averaging 11% (1960–1993) and 10% (1994–2021). Even the median of the net corporate savings distributions was nearly the same – estimated at 3% (1960–1993) and 4% (1994–2021).

Table 3 also shows that the fiscal deficit was more or less equal – averaging −3% (1960–1993) and −4% (1994–2021). The behaviour of nominal interest rates was essentially the same for both sub-periods regardless of whether the mean, median or maximum is considered. However, the exchange rate became much more volatile in the 1994–2021 period than it was in the apartheid period – averaging R/USD1.00 (1960–1993) and R/USD9.00 (1994–2021) and the maximum for the post-democracy period was as high as R/USD16 relative to R/USD3.00 for the apartheid period. The rand has experienced several crises in the 1990s and 2000s. Average household disposable incomes were much higher under the democratic dispensation than the apartheid era.

Table 4 reveals important correlation patterns that deserve a comment. Whether the sample is broken into the pre- and post-democracy period, as well as considered in full, fiscal deficits have a negative correlation with net corporate savings – pre-democracy (insignificant), after democracy (−0.8) and full sample (−0.4). The strong negative post-democracy correlation seems to agree with Heterodox Economics literature that states that fiscal deficits are negatively correlated with net corporate savings. Net household savings are negatively correlated (−0.5) to fiscal deficits in the pre-democracy era but insignificant for the post-democracy era and the full sample. Thus, fiscal deficits have an inverse relationship to both household and corporate savings just as Figure 1 demonstrates. There is also a positive correlation (0.3) between household and corporate savings.

In a somewhat surprising outcome, the nominal interest rate had a statistically insignificant relationship with net corporate savings for both sub-samples and the full sample. The nominal interest rate had a weakly significant negative correlation (−0.4) with household savings (1960–1993). The nominal interest rate had a positive correlation with household savings, estimated at 0.7 (1994–2021) and 0.6 (1960–2021). The nominal exchange rate was negatively correlated with corporate savings in the pre-democracy sub-sample (−0.5) and in the full sample (−0.3).

The correlation between household disposable income per capita and net household savings is statistically significant and negative in the post-democracy sub-sample (−0.7) and in the full sample (−0.8). Theoretically, one would expect a growing household disposable income per capita to result in higher household savings but rising incomes could be an entry visa into credit markets so that household debt rises at the expense of savings. Literature suggests that South Africa is affected by a Veblenian conspicuous consumption problem, which should result in rising incomes being associated with falling household savings (Chipp, Kleyn & Manzi 2011; Van Wyk & Posel 2019).

Table 4 also shows that the nominal exchange rate and household income per capita were positively correlated to nominal interest rates, estimated at 0.9 (1960–1993) for both. However, in the post-democracy the correlation became negative but with the same absolute sizes, estimated at −0.9 (exchange rate) and −1.0 (disposable income per capita). The full sample shows that the both correlations remain negative and fall to −0.5. As income and the exchange rate have almost perfect linear correlations, the empirical model dropped them to avoid the problem of severe multicollinearity (Gujarati & Porter 2009).

Table 5 shows that the variables are integrated of mixed orders. The combination of integrated processes of order zero and one paved way for a Toda-Yamamoto VAR modelling framework as presented already. However, for the household savings regression, a standard VAR was estimated as all variables were integrated of order zero.

Table 6 reports Granger non-causality test results for the corporate and household savings VAR systems. In the household savings VAR system, fiscal deficits granger cause household savings, which is expected from both the heterodox and neoclassical systems of thought, although they differ in the nature of the effect. Fiscal deficits also unidirectionally Granger-cause nominal interests.

With respect to the corporate savings VAR system, as is expected in both heterodox and neoclassical thought, fiscal deficits Granger-cause the nominal interest rate. The difference between these schools of thought is that the heterodox view says deficits reduce interest rates, while the neoclassical thought says deficits increase interest rates.

Figure 2 reports the impulse response functions for the household savings VAR system based on Kilian’s unbiased bootstrapping procedure. In conformance to heterodox economic theory, a one standard deviation decrease in fiscal deficits (i.e. positive shock, moving towards surplus) is associated with decreasing household savings (Figure 2b). The size of the short-run multiplier is about −0.15, and the long run multiplier is about −0.51. To put it differently, a 1% point decrease in fiscal deficits reduces household savings by 0.15% in the short run and 0.51% points in the long run. Conversely, a 1% point increase in household savings increases fiscal deficits by between 0.1% points in the short run and 0.3% points in the long run (Figure 2d). This conforms to the prediction of the two-gap model: private savings build-up are associated with increasing public sector deficits.

A one standard deviation positive shock to nominal interest rates induces a decrease in household savings (Figure 2c). This result seems atheoretical, but the household savings measure is adjusted for both debt and interest payments. As such, net household savings in a society with highly indebted households would experience a decline in savings as interest rates increase. In the short run, a 1% point increase in the nominal interest rate reduces household savings by 0.19 percentage points, but the effect declines to 0.07 percentage points in the long run. On the other hand, a 1% point increase in household savings reduces the nominal interest rate by −0.9 percentage points in the short run and 1.2 percentage points in the long run (Figure 2g). According to the loanable funds theory, as the supply of loanable funds increases, interests decrease.

Figure 2h also reveals that a one standard deviation positive shock in fiscal deficits (moving towards surplus) increases interest rates. A 1% point reduction in fiscal deficits increases nominal interest rates by 0.9 percentage points in the short run and 2.1 percentage points in the long run. While this finding seems theoretically problematic from a neoclassical perspective, it conforms to the prediction of heterodox thought, such as Modern Monetary Theory, which shows that deficits create financial assets that increase savings (and so liquidity) and the effect is to reduce interest rates. These views are advanced by thinkers such as Kelton (2015, 2020) and Sawyer (2020). However, an increase in the nominal interest rate increases fiscal deficits by between 0.26% in the short run points and 0.42% points in the long run, and most likely the effect works via the income effect and investment effect such that the tax base shrinks (Figure 2f).

Figure 3b shows that a one standard deviation positive shock in the fiscal deficit (a move towards surplus) reduces corporate savings. A 1% point decrease in fiscal deficits induces a 0.3% point decrease in corporate savings in the short run, but the effect becomes positive in the long run (0.5% points). This prediction entirely agrees with the heterodox view that fiscal surpluses reduce the stock of financial assets in the economy and so, savings. This is evident also in that a positive shock to the fiscal deficit (moving towards surplus) increases interest rates. Conversely, a 1% point increase in corporate savings has no effect on fiscal deficits in the short run but increases fiscal deficits in the long run by 0.22% points (Figure 3d). The corporate savings model shows, as did the household savings model, that as fiscal deficits move towards surplus, the nominal interest rate increase – the short run multiplier estimated at 0.8 and the long run multiplier reaching 3.4 (Figure 3h). This confirms the heterodox claim.

The article posed three questions. Firstly, do South African private savings-fiscal deficits nexus conform to the neoclassical hypothesis? Secondly, do fiscal deficits have the same effect on both household and corporate savings? Thirdly, has the savings-fiscal deficit relationship undergone structural change since the 1990s. With respect to the first question, evidence suggest that the net household savings-fiscal deficits nexus conforms to the heterodox view that increasing fiscal deficits increase household savings through the income channel. The neoclassical view finds no support. In response to the second question, the Granger causality analysis showed that fiscal deficits do not have a causal effect on corporate savings, but they have a causal effect on household savings. As to the third question, various dummies relating to the period of the 1990s were insignificant except the democratic transition dummy and the Dotcom bubble dummy. The effects of both dummies suggest that corporate savings and household savings declined in the 1990s.