February 2018 · Number 1 Identifying the Drivers of Demand for Government in Southern African Customs Union Countries Sébastien C. Dessus, Johannes Herderschee, average for Sub-Saharan Africa. In 2015, SACU’s unweighted average government size Marko Kwaramba, Nyasha Munditi1 exceeded 39 percent of GDP, against less than 25 percent for the 32 other Sub-Saharan Introduction countries considered in this analysis. Lesotho The size of government – defined as the share and Namibia rank respectively first and second of the public expenditure in GDP – measures among the 37 countries considered, while the the magnitude of resources that are under the other three SACU countries rank among the control of the public sector and available for seven largest countries in Sub-Saharan Africa reallocation across sectors, income groups and (SSA).2 geographical locations. The size of the government is determined by both demand Brief Literature Review and supply factors. Supply factors include the Empirical Literature on the determinants of ability to mobilize resources (tax and non-tax government sizes comprises two main revenue, borrowing) and capacity to spend branches. 3 Both use cross-country analysis to them (public financial management, test their theoretical assumptions. None of procurement). Demand for public services vary them, however, focus specifically on sub- by country, depending on – among others - its Saharan African countries. level of development, exposure to shocks, On the one hand, government size is demography, income distribution, political explained by the capacity to mobilize systems and social fractionalization. revenues. Fenochietto and Pessino (2013) 4 This note focuses on the determinants of develop a stochastic tax frontier analysis on governments’ size in Southern African panel data from 1991-2012 to estimate the tax Customs Union (SACU) countries: Botswana, potential of 113 countries (first for 96 non- Lesotho, Namibia, South Africa, and natural resource dependent countries and then Swaziland. The size of SACU governments (as an addition of 17 resource-dependent measured by central government expenditures economies). The estimated tax potential is then over GDP) tends to be significantly above the 1 This practice note was cleared by Mathew growth. This branch does not however attempt at Verghis, Practice Manager, MTI. explaining the actual size of government and why it 2 For the sake of comparability across the note, a few could differ from estimated optimal levels. Barro, SSA were excluded from the sample as missing the Robert (1990), “Government spending in a simple necessary data for the econometric analysis: Cabo model of endogenous growth”, Journal of Political Verde, Democratic Republic of Congo, Equatorial Economy, 98(5), part 2, S103-125. 4 Guinea, Eritrea, Republic of Congo, Rwanda, Sao Fenochietto, Ricardo and Pessino, Carola (2013), Tome, South Sudan. Among them, Equatorial “Understanding Countries’ Tax Effort”, Working Guinea and Republic of Congo exhibit government Paper #13/244, International Monetary Fund, sizes comparable with SACU countries. Washington D.C. 3 A third branch, initially promoted by Barro (1990), concentrates on the optimal size of governments for compared to actual tax collection to measure over 100 countries from 1970-2000, he the ‘tax effort’). Two econometric methods are conducted cross sectional analysis on total and used, the Truncated Normal Heterogeneous in individual categories of expenditure (such as Mean and Decay Inefficiency (TNH) and the defense, education, health care) for different Mundlak version of the Random Effects Model levels of government (central and local) to test (REM) measure, yielding similar results: tax the following theories: potentials significantly depend positively on • Rodrik’s theory of trade openness, GDP per capita, education levels (proxied with whereby greater openness would be public education expenditure in GDP), trade associated with greater exposure to openness, and negatively on inequalities and external shocks and a related demand the size of agriculture in GDP. Thus, most from citizens for greater social European countries were found to be near their protection.7 tax potential while high levels of exemptions • The “Wagner’s Law” relating higher and tax evasion and low tax rates largely account for low levels of tax effort in some per capita GDP to higher demand for developing countries. In another related study, complex and luxury public goods such Langford and Ohlenburg (2016)5 employed the as regulatory services (needed in same econometric method to estimate the tax complex economies) or cultural capacity for 85 non-resource-rich economies, in enhancement services.8 a 27-year panel from 1984-2010. They also • Alesina-Wacziarg theory of country found the tax capacity of low income countries size, whereby sharing non-rivalrous to be on average smaller than that of high public goods across large populations income countries. Only two SACU countries generate economies of scale (thus were covered in these reports, South Africa and lower demand); and whereby large Namibia, for the year 2011. Namibia was found populations exhibit more to be close to its potential, with a tax effort heterogeneous preferences for public around 90% (25.3% of GDP collected in tax revenue, against about 28% of potential). South goods, hence agree on lower demand Africa was found to be farther away from its for public goods.9 potential: its tax potential was estimated to • Alesina et al. theory on the role of range between 36 and 38 percent of GDP, but ethnic and other forms of social tax collection only amounted to 28% (hence a fractionalization, reflecting different tax effort ranging between 73 and 77%). preferences for public goods (and thus On the other hand, government size is lower public spending).10 explained by a combination of structural • Meltzer and Richard’s theory of the factors driving demand for public role of inequality and Benabou’s expenditures. Shelton (2007) 6 used a extension of this theory to include combination of demand variables that have political rights, where higher repeatedly been shown to be correlated with patterns in government expenditure to test inequality combined with higher their econometric robustness. Using data for political rights lead to higher demand for fiscal redistribution.11 5 Langford, Ben and Ohlenburg, Tim (2016), “Tax Scottish Journal of Political Economy 41 (3), 286– Revenue Potential and Effort: An Empirical 298. Investigation”, January, Working Paper, 9 Alesina, Alberto, and Wacziarg, Romain (1998), International Growth Center, London and Oxford. “Openness, country size and government”, Journal 6 Shelton, C. (2007), “The size and composition of of Public Economics 69 (3), 305–321. government expenditure”, Journal of Public 10 Alesina, Alberto, Baqir, Reza, and Easterly, Economics, 91, 2230–60. William (1999), “Public goods and ethnic 7 Rodrik, Dani (1998) “Why do more open divisions”, Quarterly Journal of Economics 114 (4), economies have bigger governments?” The Journal 1214–1284. of Political Economy 106 (5), 997–1032. 11 Meltzer, Allan, and Richard, Scott (1981) “A 8 Oxley, Les (1994), “Cointegration, causality, and rational theory of the size of government”, The Wagner's Law: a test for Britain 1870–1913”, Journal of Political Economy 89 (5), 914–927; Lijphart, Arend (1997), “Unequal participation: February 2018 · Number 1 · 2 • Persson and Tabellini, and Milesi- South Africa through the SACU revenue- Feretti, Peroti and Rostagno theories sharing formula. This formula implicitly on the role of electoral rules and compensates Botswana, Lesotho, Namibia and government types, whereby Swaziland (BLNS) for relinquishing their majoritarian and presidential regimes individual rights for trade policy design to South Africa. Thus, actual tax revenue in require less expenditure to acquire Botswana, Lesotho, Namibia and Swaziland political power than proportional and (BLNS) may not be solely determine by their parliamentarian regimes.12 tax potentials and efforts. Second, SACU Shelton (2007) finds evidence that total countries are among the most unequal expenditure increases strongly with openness countries in the world, and testing the impact in both industrialized and less-developed of inequalities and political systems on the countries. However, in less developed demand for public expenditure may be countries the increases are not in categories that particularly enlightening. And third, because insure for social risk but in such sectors as analyses combining high and low income transportation and education. He also finds economies makes difficult the interpretation of supportive evidence for the hypotheses that results, given the heterogeneity of institutions greater inequality and better political rights are between these groups that simple indicators associated with more redistribution, may not fully capture. Hence the choice to particularly in industrialized democracies. As restrict the analysis to Sub-Saharan Africa, seen for the Wagner’s Law, Shelton’s results suggest as more homogeneous institutionally. that it is in reality driven by the demographic Using a cross-country sample of 37 Sub- structure of countries, where richer countries Saharan Africa countries, we regress the share tend to have more elderly and thus spend more of central government expenditure over GDP on social security and other forms of social (in 2015) on a set of structural factors. Given our focus on the impact of inequalities, we protection which boosts their total expenditure. However, the law fails to hold retain the Gini coefficient as an explanatory variable in all our regressions and test with it when social security is excluded from total other potential structural factors. Simple expenditure as spending declines with per correlation analysis suggests that inequalities capita income. He also concludes that in more are not significantly correlated with any other populous countries and in countries with greater ethnic fractionalization, spending on explanatory variables. many categories of public goods (education, Most tested variables have the expected sign healthcare, public order and safety) is and are statistically significant. The GINI associated with higher spending by local coefficient (Source: WDI) is of expected sign governments (not necessarily fully reflected in and statistically different from zero in all central government expenditures in the form of specifications. Results suggest that a one transfers to local governments). Finally, the percentage point increase in the GINI paper finds majoritarian governments are coefficient (i.e. worsened inequalities) increase statistically associated with reduced public expenditure by 0.26-0.37 percentage expenditures and without any bias towards or points of GDP. The trade openness variable against any type of spending. (TROP, measured as the sum of exports and Methodology and Results imports of goods and services over GDP, In the rest of this note, we aim at replicating averaged over the period 2010-15; Source: Shelton’s econometric approach on a selected WDI) is expected sign and statistically different set of Sub-Saharan countries, including all from zero. Results suggest that a one SACU countries, for a number of reasons: percentage point increase in the TROP ratio (i.e. First, because SACU countries derive a large increased exposure) increase public part of their tax revenue from transfers from expenditure by 0.15 percentage points of GDP. democracy's unresolved dilemma”, American Ferretti, Gian-Maria, Perotti, Roberto, and Political Science Review 91 (1), 1–14. Rostagno, Massimo, (2002), “Electoral systems and 12 Persson, Torsten, and Tabellini, Guido (2004), public spending”, Quarterly Journal of Economics “Constitutional rules and fiscal policy out-comes. 117 (2), 609–657. The American Economic Review 94, 25–46; Milesi- February 2018 · Number 1 · 3 The population variable (POP, expressed in Lesotho records much higher public millions; Source: WDI) is of expected sign and expenditure than all other countries in the statistically different from zero, suggesting that sample. However, except for Nigeria lying at more populated countries have lower demand the other extreme of the spectrum, one cannot for public expenditure. In the absence of other infer from the model statistically significant control variables than GINI, the GDP per capita differences in predicted government sizes variable (GDPC, expressed in constant PPP$ across countries. and averaged over the period 2010-15; Source: The model can finally be used to identify the WDI) is of the expected sign but not statistically respective contributions of the different significant. The political rights variable (POLR, variables to differences observed across ranging from 0 to 1, worst to best; Source: countries in government sizes. Variables are Freedom House) is also of the expected sign, regrouped in 3 different groups: but is not either statistically significant. In • Inequalities reflects the contribution of contrast, the majoritarian political regime the GINI variable to the observed variable (MAJS, taking the value of 1 for difference between countries’ majoritarian systems and 0 otherwise; Source: predicted government sizes and Persson and Tabellini, 1999) and the sample average. presidential political regime variable (PRSS, • Political systems and rights reflects the taking the value of 1 for presidential systems combined contributions of the POLR, and 0 otherwise; Source: Persson and Tabellini, MAJS, PRSS and ETHFR variables to 1999) are both of the expected sign and are statistically significant. 13 Likewise, the ethnic the observed difference between fractionalization variable (ETHFR; Source: countries’ predicted government sizes Alesina et al., 2003) is of the expected sign and and sample average. is statistically different from zero: less • Economy and population reflects the homogeneous societies tend to demand less combined contributions of the TROP, goods and services from central POP, and GPDC variables to the governments. 14 Taken all together (Table 1, observed difference between Column 9), these variables explain two-third of countries’ predicted government sizes the observed differences in government sizes and sample average. between the 37 sub-Saharan countries considered in the sample. In this specification, Results suggest that SACU countries’ the POP variable loses its statistical comparatively high government sizes are significance, as strongly (inversely) correlated driven by the combination of high with the TROP variable. In contrast, the GDPC inequalities and conducive political systems variable becomes significant and of the to press governments for more fiscal expected sign. redistribution. SACUs’ government sizes are on average about 10 percentage points higher Using this model (Column 9, Table 1) allows than the Sub-Saharan average (as measured by to predict quite accurately the ranking of this sample); and out of this difference, more SACU countries. Lesotho continues to rank than 8 percentage points of GDP can be first, while other SACU countries rank among attributed to the combination of inequalities the first 9 nations in terms of predicted (3.5 percentage points of GDP) and political government sizes. Given standard errors of systems (4.2 percentage points of GDP), predictions, the model also allows to conclude confirming the strong role played by the fiscus that the considered structural variables can to redistribute resources in Namibia and South explain with statistical significance why Africa. 15 In contrast, economic and 13 15 Persson, Torsten, and Tabellini, Guido (1999), Using the Commitment to Equity framework, “The size and scope of government: comparative World Bank found that fiscal systems contributed to government with rational politicians”, European reduce the Gini coefficient (measuring income Economic Review 43, 699–735. inequalities) by respectively 20.6 and 17.5 14 Alesina, Alberto, Devleeschauwer, Arnaud, percentage points in Namibia and South Africa. Easterly,William, Kurlat, Sergio, and Wacziarg, World Bank (2017), “Does Fiscal Policy Benefit the Romain (2003), “Fractionalization”, Journal of Poor and Reduce Inequality in Namibia”, June, Economic Growth 8, 155–194. Washington, DC.; World Bank (2014), “Fiscal February 2018 · Number 1 · 4 demographic variables only contribute for 1.9 factors driving the demand for public percentage point of GDP to the observed expenditures also play a role. In particular, the difference with the sample’s average. Within response of political systems to inequalities SACU, the large population in South Africa, appears to be a main driver of public and the relatively high per capita GDP (except expenditures, reflecting the ongoing social for Lesotho), contribute to lowering demand contract in these countries built since for public expenditure. independence. Thus, fiscal consolidation Conclusion programs designed to restore fiscal As all SACU countries currently face difficult sustainability without affecting the role played macroeconomic situations, 16 a better by governments to reduce inequalities have understanding of the structural drivers of probably a greater chance of achieving public expenditure is important to possibly sustainable results. identify realistic fiscal consolidation pathways. While high levels of government expenditures About the author(s): in BLNS are often associated with the Sébastien C. Dessus, Program Leader, World opportunity provided by SACU transfers from Bank’s Africa Region South Africa, our results suggest that structural sdessus@worldbank.org Johannes Herderschee, Senior Economist, World Bank’s Macroeconomics, Trade & Investment Global Practice jherderschee@worldbank.org Marko Kwaramba, Economist, World Bank’s Macroeconomics, Trade & Investment Global Practice mkwaramba@worldbank.org Nyasha Munditi, Consultant, World Bank’s Macroeconomics, Trade & Investment Global Practice nmunditi@worldbank.org Policy and Redistribution in an Unequal Society”, BLNS Countries”, Southern Africa Programmatic South Africa Economic Update 6, November, Fiscal Work (P148373), June 2017. Washington, DC. 16 World Bank (2017), “South Africa’s Recent Economic Developments and SACU Revenue for February 2018 · Number 1 · 5 Appendix 1 – Tables and Figures Source: Government Financial Statistics, International Monetary Fund. Table 1: Determinants of Demand for Central Government Expenditures (% of GDP, 2015) 1 2 3 4 5 6 7 8 9 C 10.009 2.089 14.853 10.560 8.035 15.621 16.921 22.743 13.409 1.39 0.38 2.17 1.44 1.10 2.32 1.94 2.68 1.84 GINI 0.379 0.266 0.320 0.352 0.368 0.355 0.317 0.289 0.257 2.39 2.25 2.17 2.12 2.33 2.48 1.95 1.89 2.44 TROP 0.16 0.15 5.51 4.73 POP -0.106 -0.029 -2.71 -0.96 GDPC 0.000 -0.001 0.59 -2.71 POLR 8.030 2.532 1.21 0.57 MAJS -7.342 -5.010 -3.03 -2.92 PRSS -4.980 -4.691 -1.37 -1.76 ETHFR -13.174 -1.900 -2.46 -0.45 Adj R2 0.116 0.519 0.293 0.099 0.127 0.283 0.137 0.227 0.664 Source: World Bank staff calculations. Note: in italic are T-Student statistics. February 2018 · Number 1 · 6 Source: World Bank staff calculations. Source: World Bank staff calculations. February 2018 · Number 1 · 7