This paper discusses South Africa’s recent growth performance and its new growth targets. It analyzes the history of exchange rate volatility, compares it with other countries, and examines the relationship between volatility and trade flows in South Africa. It highlights facts on reserves holding and presents an empirical analysis of a model in assessing the adequacy of South Africa’s reserves. It also analyzes the cyclical balances to determine revenue performance and discusses the penetration of the South African financial conglomerates into State Security Agency and assesses potential vulnerabilities.

Abstract

This paper discusses South Africa’s recent growth performance and its new growth targets. It analyzes the history of exchange rate volatility, compares it with other countries, and examines the relationship between volatility and trade flows in South Africa. It highlights facts on reserves holding and presents an empirical analysis of a model in assessing the adequacy of South Africa’s reserves. It also analyzes the cyclical balances to determine revenue performance and discusses the penetration of the South African financial conglomerates into State Security Agency and assesses potential vulnerabilities.

IV. Cyclically Adjusted Budget Balances: An Application to South Africa1

A. Introduction

1. A proper assessment of economic policies should be based on indicators of discretionary actions that reflect policymakers’ objectives and constraints. In the fiscal realm, it is common to rely on definitions of budgetary balances that capture specific dimensions of fiscal policy. For example, the primary fiscal balance is used in combination with the debt to GDP ratio to assess fiscal sustainability. The current balance captures the contribution of the government sector to national savings; the public sector borrowing requirement makes it possible to assess the impact of government activities on financial markets. And changes in the conventional budget balance are interpreted as an indicator of how the fiscal policy stance affects aggregate demand.2

2. A well-known shortcoming of nominal budget balances is that they automatically respond to developments unrelated to policy actions. The budget is a commitment to allocate public resources to certain programs subject to a well-defined budget constraint. Hence fiscal policy actions are primarily reflected in expenditure variations and in the eventual revenue-raising measures required to satisfy the budget constraint. However, actual revenues, and to a lesser extent, expenditures, reflect events that are beyond the control of government in general and fiscal policy actions in particular. As a result, assessing fiscal policy on the basis of ex post budget balances can be misleading.

3. The main source of endogenous variations in nominal budget balances is the business cycle. Various methods can be used to separate the automatic budgetary impact of the cycle from the effect of discretionary policy changes. Budgetary surveillance conducted by the European Commission, the OECD, and the IMF now routinely relies on cyclically adjusted budget balances (CABs). A growing number of individual countries also publish CABs in official documents to complement the information conveyed by nominal balances. In Chile, the United Kingdom, Sweden, or the Euro area, CABs even play an explicit role in rules-based fiscal frameworks, although, with the notable exception of Chile, they are not primary budgetary targets.

4. In South Africa, budget documents discuss fiscal policy in terms of nominal balances only. A discussion of business cycle and exogenous factors that impact the budget nevertheless helps understand the broad variations in fiscal indicators. The conventional balance, the primary balance, and the current balance of the national government, as well as the borrowing requirement of the nonfinancial public sector receive particular attention.3

5. This chapter applies commonly used methods of cyclical adjustment to the primary and the conventional balance of the national government in South Africa. Because the results broadly support Horton’s analysis (2005) of fiscal policy during the last decade, the discussion here focuses more on the methodological challenges to estimating CABs for South Africa and on the budgetary impact of cyclical developments in recent years.

6. The following main conclusions emerge:

  • Cyclically adjusted balances are generally close to their nominal counterparts, reflecting the relatively small size of the government sector in South Africa, the negligible effect of the business cycle on expenditure, and the moderate deviations of actual GDP from its trend. Still, changes in CABs give a more accurate picture of the underlying policy impulses, and can sometimes differ significantly from changes in nominal balances.

  • CABs obtained with three alternative methods give broadly similar results. The first is a rule of thumb that assumes that tax revenues have a unit elasticity to the output gap. The second allows for different elasticities in broad tax categories, using econometric estimates for South Africa. The third is a disaggregated method that accounts for possible composition effects in GDP cycles. Despite sizeable dissimilarities in the cyclical behavior of GDP components in South Africa, the impact on CAB estimates generally remains limited.

  • While it is assumed that expenditure does not systematically respond to the cycle, the cyclical sensitivity of tax revenues in South Africa is generally comparable to that of other economies. Corporate income tax (CIT) revenues nevertheless exhibit a relatively high elasticity to economic activity, perhaps because econometric estimates may also capture the buoyancy associated with improved compliance in recent years.

  • Projections of CABs for 2006/07 to 2008/09 confirm an expansionary medium-term path of fiscal policy. These estimates may, however, suffer from a downward bias in the estimation of the business cycle, owing to the use of Hodrick-Prescott filtering. The projections are nevertheless robust to the different output gaps obtained under alternative growth scenarios underlying GDP projections.

  • Analysis of cyclical balances provides useful information on what determines revenue performance in South Africa. The recent increase in the tax revenue-to-GDP ratio is only partly explained by cyclical factors, stressing the need for a better grasp of the impact on tax revenues of improved compliance. It would also be useful to better understand the contribution of the informal sector to revenue performance, notably through indirect taxes paid on inputs.

B. Concepts and Methods

7. A cyclically adjusted balance is an estimate of the budget balance that would prevail if GDP was on trend, or at potential, in a given year. Hence, assuming no change in revenue and expenditure policies, any deviation of actual GDP from trend (i.e., any output gap) results in a difference between the CAB and the nominal balance. Because the CAB is expected to be insensitive to the automatic impact of transitory fluctuations in economic activity, it is generally thought to better reflect the discretionary policy stance than the nominal balance.

8. To calculate CABs, the nominal budget balance is decomposed into two parts: the cyclical balance (which captures the automatic effect of the cycle on the budget, and denoted by Bc below), and the CAB (denoted by B*).

B=B*+Bc(1)

If output is on trend, then Bc = 0 by definition, because there is no difference between the nominal balance and the CAB (B = B*). When output is below trend (such as at the end of a recession), revenue is lower than if output were on trend, and Bc < 0. In that case, the CAB shows a lower deficit, or higher surplus, than the nominal balance, that is, discretionary fiscal policy is less expansionary than it appears from the nominal balance (B* > B). When output is above trend, then Bc > 0, and the CAB shows a larger deficit, or smaller surplus, than the nominal balance (B* < B), and discretionary fiscal policy is more expansionary than it appears. Table IV.1 summarizes the relationship between the output gap, the cyclical balance, and the CAB.

Table IV.1.

Budget Balances, the Output Gap, and the Interpretation of Cyclically Adjusted Balances

article image
Source: IMF staff.

9. B* does not exclude all cyclical influence. In particular, since the aim of cyclical adjustment is to better capture policy actions,4 the discretioary response of fiscal policy to the cycle needs to be reflected in B*. Moreover, conventinonal adjustment methods generally ignore variables sensitive to the business cycle that may directly influence the budget independently of the output gap, including commodity prices—particularly relevant where a significant part of the extraction industry is state-owned—interest rates, exchange rates, and asset prices.5 One reason for omitting them is that the cyclical component is difficult to estimate. Also, applying multiple corrections to nominal balances inevitably complicates interpretation of CABs.

10. Cyclical adjustment of nominal budget balances involves mechanical correction of budget items deemed sensitive to the output gap. In general, expenditures are considered not to automatically react to GDP fluctuations; the only exception is unemployment benefits, which depend on the difference between actual and “structural” unemployment rates. However, since the calculations in this paper concern the national government and exclude expenditures of the Unemployment Insurance Fund,6 no correction will be applied to expenditures. Likewise, nontax revenues—in South Africa mainly interest receipts, dividends (potentially cyclical but quite volatile), administrative fees, and the proceeds from financial transactions—will not be adjusted. A specific correction for commodity price cycles also does not seem necessary in view of issues mentioned earlier, and of the modest contribution of the mining sector to corporate tax revenues (see Table IV.2).

Table IV.2.

Structure of Tax Revenues

(1998-2006)

article image
Sources: National Treasury, South African Revenue Service, and IMF staff calculations.

11. Formally, B* is defined as follows:

B*=Σi=1τTi(Y*Y)ηi+NTG(2)

where Ti designates a tax revenue item i (a common decomposition is between corporate income tax, personal income tax, and trade and indirect taxes; see Table IV.2), G, total expenditure, and NT, non tax revenues. Trend GDP is denoted by Y* and actual GDP by Y, and are both expressed in real terms. The parameters i η represent the elasticity of the tax category i to the output gap. To economize on the notation, equation (2) does not account for possible lags in tax collection—an issue that may be particularly relevant for corporate income tax (see below).7

12. Equation (2) can be rewritten in terms of ratios to potential GDP. In the notation that follows, a star designates a cyclically adjusted variable, and lower-case letters denote ratios to GDP for unadjusted variables and ratios to potential GDP for adjusted variables.

b*=Σi=1τti(Y*Y)ηi(YY*)+nt(YY*)g(YY*)b*=Σi=1τti(Y*Y)ηi1+nt*g*(3)

13. From equation (3), two common adjustment methods can be described. The first uses a simple rule of thumb according to which tax revenues automatically respond to GDP with an elasticity of 1, whereas expenditures elasticity is zero. In that case, only an estimate of potential GDP is needed to calculate the CAB. It is clear that assuming ηi = 1 for all i ε [1,τ] in equation (3) implies:

bbRT*=(ntg)(1YY*)(4)bRT*=b(gnt)(YY*Y*)

where a subscript RT refers to the rule-of-thumb method.

14. Incidentally, equation (4) shows that the magnitude of the correction for the cycle is directly proportional to the size of government (as measured by the expenditure ratio net of nontax revenues). Figure IV.1 below gives a flavor of the cyclical balance (i.e., the difference between the nominal balance and the CAB) associated with plausible ranges of government size (15 to 55 percent of GDP) and output gaps (-3 to 3 percent of potential GDP).

Figure IV.1.
Figure IV.1.

Theoretical Range of the Cyclical Balance Under the Rule-of-Thumb Method

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Source: IMF staff calculations.

15. A second correction method inspired by equation (3) is to allow for tax elasticities to differ according to the type of tax. Hence, besides estimate of potential GDP, elasticities of various tax items to the output gap are also needed. These elasticities depend on the sensitivity of revenues to the tax base (denoted below by ηiTBi), and on the sensitivity of the tax base to the output gap (denoted by ηTBiY). Specifically,

ηi=ηiTBiηTBiY(5)

where TBi designates the tax base on which tax i is levied.

16. The elasticity ηiTBi reflects the progressivity of the tax system. For example, because the personal income tax (PIT) is usually progressive, meaning that the average tax rate increases with taxable income, ηiTBi should be greater than 1. Normally, there is a single rate for corporate income tax (CIT), implying neither progressivity nor regressivity. The VAT may be mildly progressive because basic goods may be exempt or zero-rated and luxury items subject to higher rates. In contrast, specific excise duties are regressive because they depend on real consumption. Overall, indirect tax revenues are thus likely to exhibit an elasticity to the tax base close to 1.

17. Elasticities ηiTBi are often calculated using statutory tax schedules and data on income distribution (see Giorno and others, 1995, and Bouthevillain and others, 2001); the elasticity of the tax base to the output gap (ηTBiY) is determined econometrically. Some authors also estimate ηiTBi with econometric techniques (e.g., Momigliano and Staderini, 1999) due to the demanding data requirements of the statutory approach.

18. One assumption underlying the cyclical adjustment methods described in equations (3) and (5) is that GDP growth is balanced in the sense that all its components follow the same cyclical pattern. In practice, however, a given cycle may not resemble previous ones because the primary sources of growth (consumption, net exports, investment) differ. Also, there may be systematic differences in the cyclical behavior of GDP components, suggesting that tax items may be subject to different cyclical influences. For instance, the labor market tends to respond with a lag to developments in the real economy. As a result, movements in taxable labor income are unlikely to be perfectly synchronized with GDP.

19. To account for these composition effects, it has been proposed to adjust each tax category in relation to the cyclical pattern of GDP components that proxy the relevant tax base (see Bouthevillain and others, 2001, Langenus, 1999, and Momigliano and Staderini, 1999).8 Equation (6) describes the corresponding adjustment formula:

bD*=1Y*[Σi=1τTi(Yi*Yi)ηiYi+NTG](6)

where Yi designates a GDP component that proxies the tax base on which tax i is levied, Yi* is the trend value of the proxy base, ηiYi expresses the elasticity of tax i revenues to its proxy base, and the subscript D stands for “disaggregated method.” Some official institutions, such as the European Central Bank, prefer that method to output-gap-based approaches (see Bouthevillain and others, 2001).

20. The discussion of cyclical adjustment methods points to limitations that call for caution when interpreting CABs:

  • Reliable cyclical adjustment depends on properly identifying potential GDP. Yet there is no generally accepted way to do so, and existing statistical tools (see below) yield uncertain and potentially biased results, especially for recent years and forecasts. That problem is particularly relevant in countries like South Africa where structural changes affect potential growth. Hence, CAB forecasts and recent outturns are likely to be significantly revised as new information on the cyclical position of the economy becomes available.

  • Departing from the output gap as the sole basis for adjustment may cause problems. First, the risk of error is increased. For instance, a specific correction for commodity price cycles or for exchange rate effects may duplicate the output-gap-related adjustment to a significant extent. Also, data on GDP components or proxy tax bases (required to implement a disaggregated approach) may be less reliable than aggregated GDP series, compounding the statistical bias associated with estimating cycles. Second, the CABs obtained with such methods are harder to interpret. For instance, assessing the pro- or countercyclical nature of discretionary policy (the relationship between policy and the business cycle) is potentially problematic if the CAB is not based on a consistent definition of the cycle.

  • Different methods can yield different outcomes, and the choice of options is essentially a matter of judgment. As the purpose of CABs is to provide improved information on underlying fiscal policies, a simple and transparent adjustment technique should be preferred.

C. Cyclical Developments and Tax Elasticities

21. Two preliminary steps are needed for calculating CABs. The first is to estimate the trend and cyclical components of GDP and of the proxy tax bases. The second is to determine tax elasticities.

Trends Versus Cycles

22. Various techniques can be used to estimate potential output. They include production functions, statistical filtering, and vector autoregression models (see Arora, 2005, for a comparison of these techniques with South African data). For the sake of simplicity, transparency, and comparability with the literature, this chapter relies on the Hodrick-Prescott (HP) filter, a common univariate filtering procedure to decompose a time series into a trend and a cyclical part.

23. The simplicity and transparency of the HP filter come at a cost. First, as discussed in the Appendix, implementing the HP filter implicitly rests on the analyst’s view of the length of a “typical” cycle. Hence, the extent to which an acceleration, or deceleration, in actual GDP growth will be attributed to structural or cyclical factors is somewhat arbitrary. Second, the HP filter works in such a way that the calculated trend series is too close to the actual series at both ends of the sample. The cyclical component of GDP is therefore systematically underestimated as one gets closer to the end points. To partly alleviate the end-point bias, it is common to expand the sample period with GDP projections. However, estimates of the cyclical position of the economy depend on the scenario underlying those projections (see below).

24. To avoid mechanical adjustments of calendar-year data, fiscal year (FY) potential GDP has been calculated on the basis of quarterly seasonally adjusted data for real GDP published by the South African Reserve Bank (SARB) between 1985 and 2005. IMF staff forecasts through 2009 have also been introduced in the sample. The resulting output gap (defined as the deviation of actual GDP from the HP trend in percent of the latter) is displayed in Figure IV.2. It is broadly in line with previous IMF staff estimates, including those based on alternative methodologies (see Arora, 2005, and Horton, 2005).

Figure IV.2.
Figure IV.2.

Fiscal-Year Output Gaps, 1993/94-2008/09

(Percent of trend GDP)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Source: SARB, and IMF staff calculations.

25. The questionable reliability of output gap measures at the end of the sample is reflected in low numbers and in sensitivity to the underlying economic scenario. The end-point bias is likely to be significant in the last two fiscal years because the sample extends only three quarters beyond the end of FY 2008/09. Moreover, medium-term forecasts assume that GDP growth gradually reverts to potential, which tends to mechanically close the gap. The medium term scenario underlying projections also has a significant impact over FY 2004/05 to 2006/07. Under the growth acceleration scenario, that envisages a gradual acceleration to 6 percent by 2009, the output gap essentially remains closed over that period. In contrast, the somewhat more conservative IMF staff forecasts give a slightly positive output gap (0.4 percent). Finally, a slow-growth scenario, defined as 3 percent annual growth in real terms over the forecasting horizon, yields a positive gap of almost 1 percent of trend GDP in FY 2005/06.

26. Nontrivial revisions of current and past gaps are likely as actual GDP series and the corresponding projections are revised. Figure IV.3 compares output gaps calculated with four different spring vintages of (calendar-year) GDP data for 1970-2009. Between the 2004 vintage (before the two most recent revisions of GDP statistics) and the 2006 vintage, the gap for 2002 is revised downward by 0.75 percentage point; and upward revisions for 2005 and 2006 amount to 0.4 percentage point. Significant corrections go as far back as 2000 (0.25 percentage point).

Figure IV.3.
Figure IV.3.

Output Gap Estimates Under Different Data Vintages, 2000-06

(Percent of trend GDP)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Source: IMF staff calculations.

27. Still, output gaps in South Africa have on average been quite moderate over the last decade, close to or below 1 percent of trend GDP (except for FY 1993/94 and 1996/97). This is in line with the relatively low volatility of South African real GDP growth for the last 15 years (Figure IV.4).

Figure IV.4.
Figure IV.4.

Real GDP Growth Volatility in Selected Emerging and Industrial Countries, 1990-2005

(Standard deviations)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Source: World Economic Outlook database and IMF staff calculations.

28. To implement the disaggregated method, a similar HP filter is applied to GDP components that proxy tax bases, using annual data since 1970. The sample includes IMF staff projections until 2009 on the assumption of balanced growth. This chapter relates three main tax categories to specific GDP components: the proxy base for indirect taxes (VAT and excise duties) is private consumption, the personal income tax is linked to the compensation of employees, and the corporate income tax (including the dividend tax or Secondary Tax on Companies) is linked to the gross operating surplus of firms.9

29. There seem to be important differences between the GDP cycle and that of its components (Figure IV.5).10 In particular, private consumption seems well above trend in 2005/06 and 2006/07, underscoring the strong contribution of consumer spending to growth in the present cycle. As expected, the cycle for the compensation of employees (CE) lags considerably with respect to the output gap, and also looks more persistent, which is consistent with sluggish labor market adjustment. A comparison with earlier cycles suggests that the positive gaps observed in 2006/07 and 2007/08 are likely to be underestimated, because of the end-of-period bias of the HP filter and the assumption of balanced growth in the projections appended to the sample. Until 2000, gross operating surplus (GOS) gaps are well synchronized with output gaps and, as is typical, often fluctuate more. However, the relationship seems to break down in 2000, suggesting that the large exchange rate gyrations experienced after that may have significantly affected export-oriented sectors. Specifically, the protracted weakness of the rand between 2000 and 2002 may have contributed to income windfalls accounting for the large positive GOS gaps in 2001/02 and 2002/03—despite near zero output gap—whereas the strengthening of the rand in 2003 through 2005 could explain continued negative gaps during a strong recovery.

Figure IV.5.
Figure IV.5.

Cyclical Components of Proxy Tax Bases (1993/94-2008/09)

(Percent of trend values)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Source: SARB, and IMF staff calculations.

30. For completeness and consistency, the bottom-right panel of Figure IV.5 compares the output gaps using annual data consistent with the GDP components and the quarterly data used in Figure IV.2. The chart confirms a great similarity between the two estimates.

Tax Elasticities

31. Assumptions for tax elasticities vary depending on the cyclical adjustment method (see Table IV.5). By definition, the rule-of-thumb approach assumes unit elasticity across all tax categories. The tax-specific elasticities needed in the other adjustment methods are based on econometric estimates.

32. Econometric estimation of tax elasticities has limitations. In particular, the deep structural changes affecting the South African economy—such as changes in the sources of growth, shifts in income distribution, and continued efforts to expand the tax base while reducing tax rates—may lead to significant breaks in the statistical relationships. To minimize the problem, evidence of structural breaks (found by applying recursive Chow tests) motivated adjustments in the sample period.11 Estimates are also subject to a potential simultaneity bias (for instance, because discretionary tax policy measures simultaneously affect economic activity and tax revenues); to a measurement error bias as some explanatory variables, such as the output gap are not observed; and to a specification bias because elasticities may themselves be cyclical (Bouthevillain and Quinet, 1999). Finally, the econometric models may capture a systematic discretionary response of tax policy to cyclical developments, with elasticities being overestimated if tax policy tends to be countercyclical, and underestimated in the opposite case.

33. In line with Bouthevillain and others (2001), short-run tax elasticities relative to their respective proxy base (ηiYi) are obtained by regressing nominal tax revenues on real proxy tax bases, using a log-difference specification. 12

Δlog(Ti,t)=β0+β1time+β2Δlog(Ti,t1)+β3Δlog(Yi,t)+εt(7)

where t is an error term. The constant term captures possible trends in the ratio of tax revenues to the proxy base; the time variable accounts for possible changes in that trend; and the lagged dependent variable allows for a delayed response of tax revenues to the cycle.

34. Standard specification tests rejected the presence of a time trend in the CIT equation. The same tests also rejected the introduction of a lagged dependent variable, except for CIT, which is consistent with typical collection lags. The results are generally in line with expectations and with comparable estimates for other countries (Table IV.3).13 The PIT elasticity to the compensation of employees is greater than 1, as expected under progressive taxation. Indirect tax revenues have the expected elasticity to private consumption close to 1, although the estimate is imprecise. The elasticity of CIT (including the Secondary Tax on Companies) to the gross operating surplus is about 2.7, which, although not implausible, is substantially higher than recent estimates obtained with a similar approach for EU countries.

Table IV.3.

South Africa: Tax Elasticities to Proxy Tax Bases(ηiYi)

article image
Sources: National Treasury, SARB, and IMF staff calculations.Note: Standard errors in brackets. Stars indicate the level of significance (*, less than 10 percent; **, less than 5 percent; and *** less than 1 percent).

Samples reflect data availability.

The lagged dependent variable (nonsignificant at the 10 percent confidence level) is not displayed.

35. To recover South-Africa-specific tax elasticities with respect to the output gap, the sensitivity of the proxy tax bases to the output gap (ηYiY) needs to be estimated. Girouard and André (2005) propose to regress the cyclical component of the base on the output gap as follows:

Δlog(Yi,tYi,t*)=γ0+γ1Δlog(Yi,t1Yi,t1*)+γ2Δlog(YtYt*)+ut(8)

where ut is an error term. As already indicated, sample periods were adjusted to avoid structural breaks identified in the early 1980s (Table IV.4). To account for the possible effect of exchange rate fluctuations on exporters’ income, the GOS equation also includes the log-difference of the nominal exchange rate of the rand to the US dollar. The results indicate a positive but nonsignificant exchange rate effect for 2000-05.14

Table IV.4.

South Africa: Elasticities of Proxy Tax Bases to the Output Gap (ηYiY)

article image
Sources: National Treasury, SARB, and IMF staff calculations.Note: Standard errors in brackets. Stars indicate the level of significance (*, less than 10 percent; **, less than 5 percent; and *** less than 1 percent).

Full samples cover the period 1972-2005. Ajustments ensure the absence of structural breaks over the period.

The lagged dependent variable (nonsignificant at the 10 percent confidence level) is not displayed.

The exchange rate variables (nonsignificant at the 10 percent confidence level) are not displayed.

The lagged dependent variable is not displayed.

36. The elasticities of tax revenues to the output gap, which can be calculated as ηi=ηiYiηYiY,, seem plausible in terms of the literature (Table IV.5). CIT elasticity is above the range of OECD countries15 (between 1.08 for Greece and 2.08 for Iceland), whereas the PIT sensitivity to the output gap is slightly higher than the OECD average (1.26) and comparable to estimated values for Ireland and Korea. The elasticity of indirect taxes is very close to the expected value of 1.

Table IV.5.

Summary of Assumptions for Tax Elasticities

article image
Source: IMF staff estimates.

D. CABs and Cyclical Balances

37. The three alternative cyclical adjustment methods yield comparable levels of CABs (Figure IV.6, left-hand panel). Given the relatively small size of the government sector in South Africa, these estimates are generally close to the nominal balances expressed in terms of actual GDP.16 They confirm that the trend improvement in public finances over the last decade can be mainly attributed to discretionary adjustments rather than buoyant economic conditions. The only exception is FY 1996/97, when the mild fiscal tightening (0.1 percent of GDP) identified by the nominal primary balance actually masks a discretionary expansion of between 0.3 and 0.4 percent of GDP (Figure IV.6, right-hand panel). Changes in the cyclically adjusted primary balance also capture quite well the particularly strong consolidation effort undertaken between 1997/98 and 1999/2000, when spending cuts accompanied revenue gains. The shift towards strong primary spending growth initiated in 2000/01 also emerges clearly from the chart, including the expansion currently envisaged in the medium-term expenditure framework. Interestingly, the reductions in the deficit observed in 2004/05 and 2005/06 seem to be only partly related to the cycle in spite of the intent in the initial budget to implement expansionary policies.

Figure IV.6.
Figure IV.6.

Cyclically-Adjusted Budget Balances 1993/94-2008/09

(Percent of potential GDP)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Sources: National Treasury and IMF staff estimates.

38. Nominal and cyclically adjusted balances tend to converge near the end of the sample period, in line with the limited deviations of GDP and of its components from their respective HP trend. The results are robust to the different output gaps corresponding to alternative medium-term growth scenarios (Table IV.6). Discretionary policy seems to be only slightly less contractionary in 2004/05 and 2005/06 under the slow growth scenario. Again, this reflects the fact that the magnitude of automatic revenue stabilizers is proportional to the size of the government, so that only significant variations in the output gap cause meaningful variations in CABs.

Table IV.6.

Rule-of-Thumb CABs Under Alternative Growth Scenarios (2004/05-2008/09)

(Percent of potential GDP)

article image
Sources: National Treasury, South African Reserve Bank, and IMF staff calculations.

39. One important dimension of the present analysis is the attempt to better capture the cyclicality of tax revenues in South Africa by applying different elasticities for broad revenue categories. In comparison to the rule-of-thumb approach, using specific tax elasticities to the output gap slightly increases the absolute value of the cyclical balance because estimated CIT and PIT elasticities are greater than 1 (Figure IV.7). The disaggregated approach sometimes gives fairly different results, reflecting sizeable composition effects, and in particular, the impact of GOS gaps on cyclically adjusted CIT revenues (see below).

Figure IV.7.
Figure IV.7.

Cyclical Balances Under Alternative Cyclical-Adjustment Methods (1993/94-2008/09)

(Percent of potential GDP)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Sources: IMF staff estimates.

40. Decomposing the cyclical balance into the three main tax categories shows that they contribute fairly evenly to the cyclical sensitivity of the budget, although the contribution of CIT has increased (Figure IV.6). This indicates that the higher elasticity of CIT makes up for its smaller, but growing, share of total revenues (Table IV.2). Applying the disaggregated method illustrates the impact of composition effects on the determinants of the cyclical balance. Specifically, the recent, atypical pattern of GOS gaps—partly driven by exchange rate fluctuations—tended to offset other cyclical influences on revenues. The impact of strong private consumption growth on indirect taxes also emerges clearly in 2004/05.

Figure IV.8.
Figure IV.8.

Decomposition of the Cyclical Balance by Tax Categories (1993/94-2008/09)

(Percent of potential GDP)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Sources: IMF staff estimates.

41. To assess the contribution of the business cycle to revenue performance, it is also useful to look at changes in the contributions to the cyclical balance (Figure IV.9 and Table IV.7). The following conclusions emerge from the analysis:

  • The total contribution of cyclical developments to revenue performance rarely exceeded 0.5 percent of GDP.

  • The remarkable tax revenue outturns in FY 2004/05 and 2005/06 (about 1½ percent of GDP in additional revenue each year) seem to be largely independent of cyclical developments, especially so in 2005/06. PIT performance is most closely related to the cycle, with between a third and a half of the cumulative revenue increase attributed to above-potential GDP growth (Table IV.7). In contrast, because only about a quarter of the cumulative revenue gains on indirect taxes and CIT can be attributed to the cycle, these gains seem to be mostly structural or due to one-offs. These conclusions are generally robust to the higher output gaps estimated under the slow-growth scenario, although the size of the cyclical component increases mechanically. Notice in particular that more than half of CIT performance and the entirety of PIT gains in 2004/05 are then be attributable to the effects of the cycle.

  • As the current cycle matures, and barring significant one-offs, any reversal in revenue performance should remain limited, and in any case should not exceed a cumulative 0.2 percent of GDP over the period covered by the medium term expenditure framework. However, that figure doubles if the slow-growth scenario (3 percent a year over the medium term) materializes. This shows that forward-looking assessments of the downside revenue risk should be interpreted as a lower bound because output gaps are likely to be underestimated.

Figure IV.9.
Figure IV.9.

Changes in the Contributions to the Cyclical Balance(1993/94-2008/09)

(Percent of potential GDP)

Citation: IMF Staff Country Reports 2006, 328; 10.5089/9781451841046.002.A004

Sources: IMF staff estimates.

42. Unless the extent of the current cyclical upswing is very seriously underestimated, these results suggest that the exceptional tax buoyancy observed over the last two fiscal years can be better understood by assessing, using the Revenue Service’s microeconomic data, the contribution of compliance efforts, especially in the corporate sector. Indeed, conventional cyclical adjustment methods ignore changes in the actual tax bases. Also, a better understanding of the informal sector’s contribution to tax revenues (mainly through indirect taxes) would be useful. Even though the disaggregated approach captures fairly well the real consumption boom accompanying the current cycle, more than three-quarters of the increase in indirect tax revenues cannot be explained by cyclical developments (Table IV.7). That said, cyclical revenues may well have to be significantly revised in the future as the picture of the current cyclical position of the economy becomes clearer. In the future development of cyclical adjustment tools for South Africa, it would be critical to work with more accurate proxies for the tax bases and to refine estimates of tax elasticities, notably by applying methods based on actual features of the tax code and detailed data on revenue distribution.

Table IV.7.

Cyclical Determinants of Revenue Performance (1994/95-2008/09)

(Percent of GDP, except for ratios)

article image
Sources: National Treasury, and IMF staff estimates.

E. Implications

43. Because cyclically adjusted balances allow for a more accurate assessment of policymakers’ discretionary actions, they can usefully contribute to public debate on the appropriateness of fiscal policy. In particular, they can help policymakers to explicitly incorporate demand-management considerations into formulation of fiscal policy, thereby reducing the risk of procyclical policies.

44. Estimating CABs is challenging, particularly in a rapidly changing economy like South Africa’s where assessing the business cycle is, more than elsewhere, an art rather than a science, and where it is difficult to establish stable relationships between economic and fiscal developments. Besides, the quality of available data affects the quest for reliable estimates, and much remains to be done to properly capture the sometimes subtle dynamics implied by tax collection procedures.

45. Given those challenges and the corresponding uncertainty about estimated CABs, it is not advisable to use them as an explicit fiscal target. However, the results in this chapter suggest that a consistent set of CAB estimates can be obtained for South Africa, using different methods and measures of the output gap. Because they provide additional information on the policy stance, it would therefore be useful to publish CABs on a regular basis in budget documents, along with a discussion of the cyclical influence affecting fiscal performance, and of the contribution of fiscal policy to macroeconomic stability. A simple and transparent cyclical adjustment method seem preferable to more sophisticated refinements. Using specific elasticities of broad tax categories to the output gap seems to achieve a reasonable balance between transparency and the information contents of CABs. That approach also has the advantage to be directly comparable to the approach of many countries and international organizations.

Appendix

46. The HP-filtered series {Xt*} is obtained by minimizing a quadratic objective function that balances the need for the trend series to remain reasonably close to the original one (first argument in equation (9)) while exhibiting only limited variations in its growth rate (the smoothing objective captured by the second argument in (9)). Formally, {Xt*} minimizes:

Σt=1T(XtXt*)2+λΣt=2T((Xt*Xt1*)(Xt1*Xt2*))2(9)

where Xt and Xt* are expressed in natural logarithms.17

47. Implementing the HP filter requires an appropriate choice of the smoothing parameter, λ. If that is close to zero, the smoothed series converges to the actual one: changes in Xt are mostly attributed to the trend and the filter does not allow for much of a cycle. In contrast, λ → ∞ yields an almost linear trend because the filter heavily penalizes variations in the growth rate of Xt* and thereby allows for possibly very long cycles to be deemed “cyclical.” Hence, the choice of λ very much reflects the analyst’s view of the “typical” length of an economic cycle in a given economy (Morh, 2006).

48. There is a vast literature on the appropriate value for λ (see Röger and Ongena, 1999, for a survey). While there is broad agreement that λ = 1600 is appropriate for quarterly data, the recommended values for annual data vary between the 3-5 range (Pedersen, 2001) and 100 (Hodrick and Prescott, 1997). In line with the suggestion of King and Baxter (1999), this paper retains the conventional λ = 1600 for quarterly data18 and uses l = 10 for annual data. This implies that cycles lasting up to 8 years will be fully attributed to the cyclical component of the filtered series and longer cycles will be deemed structural (i.e., capturing acceleration and deceleration in potential growth) and incorporated into the trend {Xt*}.19

49. Another important practical issue in implementing the HP filter is the end-point bias. HP filtering amounts to deriving the trend series {Xt*} as a moving weighted average of actual observations with symmetrically distributed and decreasing weights (King and Rebelo, 1993). In finite samples, the distribution of weights becomes highly asymmetric at the end points; excessively large weights are attributed to extreme observations. The calculated trend values at the extremes of the sample are therefore artificially close to actual observations, and the cyclical component of the series is correspondingly underestimated. To alleviate endpoint bias, it is common to complement the sample with forecasts, even though a substantial bias seems to remain (see Mohr, 2006).

References

  • Arora, Vivek, 2005, “Economic Growth in Post-Apartheid South Africa: A Growth Accounting Analysis,” Post-Apartheid South Africa—The First Ten Years, ed. by Michael Nowak and Luca Antonio Ricci (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Bouthevillain, Carine, and others, 2001, “Cyclically Adjusted Budget Balances: An Alternative Approach,” ECB Working Paper No 77 (Frankfurt: European Central Bank).

    • Search Google Scholar
    • Export Citation
  • Bouthevillain, Carine, and Alain Quinet, 1999, “The Relevance of Cyclically-Adjusted Public Balance Indicators—The French Case,” in Indicators of Structural Budget Balances (Rome: Banca d’Italia).

    • Search Google Scholar
    • Export Citation
  • Giorno, Claude, Pete Richardson, Deborah Rosevaere, and Paul van den Noord, 1995, “Potential Output, Output Gaps, and Structural Budget Balances,” OECD Economic Studies, No 24 (Paris: Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation
  • Girouard, Nathalie, and Christophe André, 2005, “Measuring Cyclically-Adjusted Budget Balances for OECD Countries,” Economics Department Working Papers, No 434 (Paris: Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation
  • Hagemann, Robert, 1999, “The Structural Budget Balance: The IMF’s Methodology,” in Indicators of Structural Budget Balances (Rome: Banca d’Italia).

    • Search Google Scholar
    • Export Citation
  • Hodrick, Robert, and Edward Prescott, 1997, “Postwar U.S. Business Cycle: an Empirical Investigation,” Journal of Money, Credit and Banking, Vol. 29, pp. 1-16.

    • Search Google Scholar
    • Export Citation
  • Horton, Mark, 2005, “Role of Fiscal Policy in Stabilization and Poverty Alleviation,” in Post-Apartheid South Africa—The First Ten Years, ed. by Michael Nowak and Luca Antonio Ricci (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Jacobs, Davina Frederika, 2002, “Suggestions for Alternative Measures of Budget Balance for South Africa,” IMF Working Paper No 02/110 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • King, Robert, and Marianne Baxter,, 1999, “Measuring Business Cycles: Approximate Band Pass Filters for Economic Time Series,” Review of Economics and Statistics, Vol. 81, pp. 575-93.

    • Search Google Scholar
    • Export Citation
  • King, Robert, and Sergio Rebelo, 1993, “Low Frequency Filtering and Real Business Cycles,” Journal of Economic Dynamics and Control, Vol. 17, pp. 217-31.

    • Search Google Scholar
    • Export Citation
  • Langenus, Geert, 1999, “The NBB’s Work on Structural or Cyclically-Adjusted Fiscal Indicators,” in Indicators of Structural Budget Balances (Rome: Banca d’Italia).

    • Search Google Scholar
    • Export Citation
  • Momigliano, Sandro, and Alessandra Staderini, 1999, “A New Method of Assessing the Structural Budget Balance: Results for the Year 1995-2000,” in Indicators of Structural Budget Balances (Rome: Banca d’Italia).

    • Search Google Scholar
    • Export Citation
  • Morh, Matthias, 2006, “The Missing Cycle in the HP Filter: Implications for the Measurement of Cyclically Adjusted Budget Balances,” manuscript (Frankfurt: European Central Bank).

    • Search Google Scholar
    • Export Citation
  • Pedersen, Torben Mark, 2001, “The Hodrick-Prescott Filter, the Slutsky Effect, and the Distortionary Effect of Filters,” Journal of Economic Dynamics and Control, Vol. 8, pp. 1081-1101.

    • Search Google Scholar
    • Export Citation
  • Ravn, Morten, and Harald Uhlig, 2002, “On Adjusting the HP Filter for the Frequency of Observations,” Review of Economics and Statistics, Vol. 84, pp. 371-80.

    • Search Google Scholar
    • Export Citation
  • Röger, Werner, and Hedwig Ongena, 1999, “The Commission Services’ Cyclical Adjustment Method,” in Indicators of Structural Budget Balances (Rome: Banca d’Italia).

    • Search Google Scholar
    • Export Citation
1

Prepared by Xavier Debrun (FAD). This chapter benefited from insightful comments by IMF colleagues and staff of the South African National Treasury, and in particular Fabrizio Balassone, Mark Horton, Matthew Simmonds, Joan Stott, and Noekie Steyn.

2

Jacobs (2002) provide an extensive survey of useful fiscal balances in the South African context.

3

Outturns for the consolidated general government are also presented.

4

Of course, policy actions as captured by the CAB may differ from policy intent, often for reasons that escape policymakers, such as natural disasters that require emergency relief, the cost of social unrest, etc. Other transitory components also affect CABs. Removing these items from CABs gives “structural” budget balances.

5

In Chile, however, CAB estimates account for the direct budgetary impact of copper price cycles.

6

Annual expenditures of the Unemployment Insurance Fund amount to about 0.2 percent of GDP (Table IV.2), while other social transfers to households (about 4.7 percent of GDP in 2005/06) do not primarily serve to insure against cyclical fluctuations (Horton, 2005). Hence, a correction for cyclical unemployment would not significantly affect CAB estimates for the general government in South Africa. This chapter’s focus on national government and the corresponding assumption of zero elasticity of public expenditure to the output gap thus seem reasonable. As a comparison, and despite broader social safety nets in the OECD, Girouard and André (2005) find expenditure elasticities of the general government between −0.02 (Luxembourg) and −0.23 (the Netherlands), with an average −0.1. Bouthevillain and others (2001) obtain similar numbers.

7

See for instance Hagemann (1999), or Girouard and André (2005).

8

The method still leaves out the effect of changes in relative price deflators accompanying GDP cycle (see Langenus, 1999, for a more detailed discussion).

9

Gaps are calculated with series in real terms. For the last two series, nominal figures have been deflated by GDP prices.

10

The HP filter is applied to calendar-year data. FY gaps result from a mechanical correction of the original series and its HP trend.

11

While no sample adjustment seemed necessary for the equations linking revenue elasticities to the tax base (Table IV.3), statistically significant breaks were identified in the early 1980s for the relationships between the output gap and two proxy tax bases (i.e. the compensation of employees and the gross operating surplus).

12

An alternative to (7) is to perform a cointegration analysis to explicitly disentangle short-run and long-run elasticities (see Bouthevillain and others, 2001). Given the small size of the sample, a parsimonious specification seemed more appropriate here.

13

Looking at European Union member states, Bouthevillain and others (2001) find CIT elasticities to the gross operating surplus of between 0.7 (Belgium) and 1.5 (France); PIT elasticities to worker’s compensation between 1.2 (France) and 2.6 (the Netherlands); and elasticities of indirect taxes to private consumption between 0.7 (Luxembourg) and 1.2 (Sweden).

14

The model uses two exchange rate variables, one interacted with a dummy for years 2000-05, and the other interacted with a dummy for previous years.

15

See Girouard and André (2005).

16

Total expenditures of the national government represent close to 28 percent of GDP. The fact that tax policy has aimed at enlarging the tax base while lowering tax rates may also have dampened somewhat cyclical influences on the budget.

17

See Hodrick and Prescott (1997).

18

Ravn and Uhlig (2002) argue that this would imply a value as low as 6.5 for annual data.

19

The more common assumption of 100 implies that the filter would fully accommodate cycles of up to 16 years, which seems a rather liberal definition of a cycle (Röger and Ongena, 1999).

South Africa: Selected Issues
Author: International Monetary Fund
  • View in gallery

    Theoretical Range of the Cyclical Balance Under the Rule-of-Thumb Method

  • View in gallery

    Fiscal-Year Output Gaps, 1993/94-2008/09

    (Percent of trend GDP)

  • View in gallery

    Output Gap Estimates Under Different Data Vintages, 2000-06

    (Percent of trend GDP)

  • View in gallery

    Real GDP Growth Volatility in Selected Emerging and Industrial Countries, 1990-2005

    (Standard deviations)

  • View in gallery

    Cyclical Components of Proxy Tax Bases (1993/94-2008/09)

    (Percent of trend values)

  • View in gallery

    Cyclically-Adjusted Budget Balances 1993/94-2008/09

    (Percent of potential GDP)

  • View in gallery

    Cyclical Balances Under Alternative Cyclical-Adjustment Methods (1993/94-2008/09)

    (Percent of potential GDP)

  • View in gallery

    Decomposition of the Cyclical Balance by Tax Categories (1993/94-2008/09)

    (Percent of potential GDP)

  • View in gallery

    Changes in the Contributions to the Cyclical Balance(1993/94-2008/09)

    (Percent of potential GDP)