Inequality and Fiscal Policy
Chapter

Chapter 22. Fiscal Adjustment and Income Inequality in Brazilian States

Author(s):
Benedict Clements, Ruud Mooij, Sanjeev Gupta, and Michael Keen
Published Date:
September 2015
Share
  • ShareShare
Show Summary Details
Author(s)
JOão Pedro Azevedo, Antonio C. David and Fabiano Rodrigues Bastos 

Introduction

This chapter assesses the links between fiscal policy and income inequality in Brazil for the period 1995–2011 by combining fiscal data with household survey information at the state level.1 The period under analysis is marked by important changes in subnational fiscal policy and institutions because states had to increase their primary balances to comply with debt renegotiation programs that they agreed to with the federal government in the late 1990s. Even those states that did not have significantly large debt levels were bound by new fiscal rules designed to mitigate long-standing fiscal risks stemming from subnational entities.

This fiscally constrained environment has profoundly reshaped revenue and expenditure policies at the subnational level since 2000. Against this backdrop, income inequality decreased significantly. The Gini coefficient declined from 57.7 in 1995 to 52.2 percent in 2011, although it remains among the highest in the world. Microeconomic studies have linked inequality reduction in Brazil to changes in labor income, including changes in both the supply of and demand for skilled workers, and to the emergence of effective social transfer programs at the federal level (Azevedo and others 2013; Pecora and Menezes 2013; Lopez-Calva and Rocha 2012). Azevedo and others (2013) show that approximately 40 percent of the reduction of inequality in Brazil between 2001 and 2011 can be attributed to changes in the labor markets, in particular to higher hourly earnings of low-skilled workers. Transfers (public and private) and noncontribu-tory pensions contributed 20 percent and 18 percent, respectively, to the reduction of inequality. Demographic factors are one last important component in the reduction of inequality in this period.

This chapter examines to what extent subnational fiscal policy is associated with inequality dynamics, after controlling for a number of determinants of inequality already established in the literature. The results indicate that a tighter fiscal stance in Brazilian states, measured by changes in their cyclically adjusted primary balances, is not linked to a deterioration in inequality. This conclusion differs from the results of several papers that analyze the impact of fiscal consolidation on inequality at the national level for Organisation for Economic Co-operation and Development (OECD) countries, which typically conclude that fiscal consolidation is associated with increases in inequality. The findings in this chapter caution against extrapolating policy implications of the literature focusing on advanced economies to other settings.

A Brief Survey of the Literature

The recent empirical literature on the effects of the fiscal policy stance on inequality has focused mostly on OECD or advanced economies and uses data at the national level. Agnello and Sousa (2012) look at the impact of fiscal consolidation on inequality in a panel of 18 industrial countries and find that inequality increases during periods of fiscal consolidation. In addition, consolidation is particularly detrimental to inequality if led by expenditure cuts. In contrast, fiscal consolidation driven by revenue increases is associated with reductions in inequality.

Ball and others (2013) find that both expenditure- and taxed-based fiscal consolidations at the national level have typically raised inequality for a panel of OECD countries, even if the distributional effects of spending-based adjustments tend to be larger relative to tax-based adjustments. These conclusions are largely confirmed for a broader panel of countries that also includes emerging markets in a study by Woo and others (2013). These authors find positive and statistically significant elasticities of spending-based consolidations on inequality (of about 1.5 to 2), but the coefficients for tax-based consolidations are not statistically significant.

These findings are likely to differ for developing economies, where the impact of fiscal policy on inequality is shaped by lower levels of taxes and transfers relative to advanced economies. This differential impact is compounded by greater reliance on regressive taxes (such as consumption taxes) and low coverage and benefit levels of transfer programs (Bastagli, Coady, and Gupta 2012). Furthermore, overall in-kind public expenditures (on health and education, for example) have been found to be regressive in several developing countries, reflecting lower-income households’ lack of access to public services (Bastagli, Coady, and Gupta 2012).

The literature on fiscal policy and inequality in Brazil has mainly concentrated on static incidence analysis. Lustig, Pessino, and Scott (2012), using data from the 2009 household budget survey (Pesquisa de Orçamento Familiar, POF) conclude that direct taxes (such as personal income taxes) in Brazil appear to be progressive, but their impact on inequality is relatively small because of their small size relative to GDP, whereas indirect taxes are regressive. Lustig, Pessino, and Scott (2012) also examine the incidence of direct cash transfers and conclude that it varies significantly depending on the program. Bolsa Familia is well targeted to the poor, but other programs such as the Special Circumstances Pensions (SCP) benefit the top quintile relatively more.

Ferreira, Leite, and Ravallion (2010) examine the importance of macroeconomic and redistributive policies on poverty dynamics in Brazil for the period 1985–2004, focusing on state-level data, as does this chapter. They find that state and municipal “social” public expenditures have had an adverse effect on poverty (regressive incidence), whereas state-level investment spending had no significant effect.

Fiscal Institutions and Fiscal Structure in Brazil

Brazil’s current fiscal federalism arrangements were shaped by the 1988 Constitution, which created an environment of fiscal decentralization with revenue-sharing transfer mechanisms inspired by equity concerns among states. Fiscal policy among different levels of government (federal government, states, and municipalities) is not formally coordinated. A sequence of subnational fiscal crises occurred during the first 10 years after the 1988 Constitution was adopted. These crises originated in great part from a lack of fiscal discipline and the existence of moral hazard associated with federal bail-out packages. High subnational indebtedness turned into a macroeconomic risk factor at the country level.

In 1997, the federal government and the states engaged in a debt-restructuring agreement, whereby the center would take on most of the states’ debt stock and the states would be given 30 years to repay, under special conditions. The debt-restructuring agreement introduced binding constraints on the fiscal behavior of the states and required hard commitments to fiscal goals. These changes were reinforced by the 2000 Fiscal Responsibility Law, which is considered to be a landmark of fiscal reform in Brazil and an essential part of broader macroeconomic reforms (the country’s move toward inflation targeting and floating exchange rates). Among other features, the 2000 Fiscal Responsibility Law imposed quantitative restrictions such as caps on payroll expenditure, as well as limits on the debt level as a share of tax revenue (Sturzenegger and Werneck 2006).

Hence, since 2000 fiscal decentralization has been accompanied by constraints on subnational debt growth. States’ and municipalities’ fiscal performance improved quickly in this new environment. However, the subnational fiscal adjustment was not without cost. States and municipalities faced sizable current expenditures with considerable downward rigidity, so the limitation on new borrowing meant that public investment became the main adjustment variable in many instances. This situation led to the accumulation of infrastructure weaknesses as well as other repressed investment needs. From the revenue side, states began to compete with each other and offer tax exemptions to attract firms, which further compressed fiscal space.

The regressive nature of the overall tax system in Brazil is emphasized by Soares and others (2009). As of 2007, based on national accounts data, these authors estimate that indirect taxes accounted for more than 40 percent of the total gross tax burden (excluding government transfers), whereas income and property taxes accounted for less than 30 percent.2 Most of the taxation of income and property can be decomposed into corporate income taxes (close to 39 percent of the total taxation of income and property); property taxes, including the financial transactions tax (close to 27 percent); and personal income taxes (21 percent). In addition, these authors estimate that income and property taxes accounted for close to 44 percent of the increase in tax buoyancy (tax-to-GDP ratio) between 1997 and 2007, whereas indirect taxes accounted for about 33 percent of the increase.

The main state-level tax is a value-added tax on the consumption of goods and services (Imposto sobre a Circulação de Mercadorias e Prestação de Serviços, ICMS). Moreover, as discussed in Sturzenegger and Werneck 2006, most federal revenue transfers to state governments stem from the sharing of the revenue from a tax on manufactured products (Imposto sobre Produtos Industrializados, IPI). Thus, Brazilian states rely mostly on tax revenue from so-called indirect taxes rather than taxes on income or property. However, the relative importance of the ICMS and federal revenue transfers varies widely across states. The median ICMS-to-state-GDP ratio in 2012 was about 8 percent. For richer states, the ICMS corresponds to more than 50 percent of total revenue, while federal revenue transfers amount to very little. The opposite is true for states with less developed economic bases. On the expenditure side, the median public-investment-to-state-GDP ratio is about 1.5 percent with strong variation as well; for instance, in 2012, the lowest investment-to-GDP ratio was 0.3 percent while the highest was 8.2 percent. Another important expenditure category is compensation of employees, which reached about 9 percent of GDP in 2012 for the median state.

A First Look at Inequality and Fiscal Policy at the Subnational Level in Brazil

The Data

Annex 22.1 presents data definitions and sources and Annex 22.2 provides descriptive statistics for selected variables.3 The income per capita inequality measure and the employment rate variable were constructed using data from an annual household survey (Pesquisa Nacional por Amostra de Domicilios, PNAD) undertaken by Brazil’s Institute of Geography and Statistics (IBGE) and compiled by the Socio-Economic Database for Latin America and the Caribbean. PNAD is a representative survey at both the national and state levels.

Because of data availability, the measures of inequality in this analysis are based on income per capita after transfers, but before taxes. The PNAD contains information on household market income and transfers (both public and private). In this context, one could argue that this chapter is focusing on the “macroeconomic” effects of fiscal adjustments. It might have been of interest to use measures of inequality on a net-of-taxes basis to also pick up the effect of fiscal adjustment on direct taxes, but this information is not available from the PNAD.

Lustig, Pessino, and Scott (2012) attempt to infer the disposable income of Brazilian households based on current tax legislation. These authors also try to impute other in-kind transfers, such as public expenditure on health and education, using a different data set, the household budget survey (Pesquisa de Orçamento Familiar, POF). However, the POF is published every five years, not annually as is the PNAD.

Lustig, Pessino, and Scott’s (2012) approach could not be implemented to obtain disposable incomes in this analysis for several reasons. In addition to the limitations in the time dimension of the POF data, the imputations performed by these authors are only available for one year. More generally, the use of disposable income as a dependent variable would hinge on the assumption that the tax legislation is actually binding. Furthermore, any attempt to apply Lustig and colleagues’ methodology to other POFs would still encounter the problem of geographical representativeness, since the 1994/95 and 2002/03 POFs are not representative at the state level (they are only representative at a broader geographical level for the five macro regions).

State-level fiscal indicators are constructed based on a data set compiled by the National Treasury Department at the Ministry of Finance. The data set provides comprehensive information on revenue, expenditures, assets, and liabilities for the Brazilian states. The exercise also uses additional Treasury data for information on “social” public expenditures at both the state and municipal levels. These expenditures are defined as spending on education and culture, health and sanitation, and social security and social assistance (similar to the categories used in Ferreira, Leite, and Ravallion 2010).

In addition, this analysis also uses Regional National Accounts Statistics from IBGE to obtain a series for state-level GDP and the respective deflators. Finally, it uses information on federal social transfers at the state level obtained from the Institute for Applied Economic Research (Instituto de Pesquisa Economica Aplicada, IPEA). This comprises information on three main federal social programs—Bolsa Família, Benefício de Prestação Continuada, and Renda Mensal Vitalícia—all of which are direct cash transfers to households.

Recent Trends and Stylized Facts

Figure 22.1 depicts trends in a number of inequality measures during the period of interest. As panel 1 illustrates, overall per capita income inequality at the national level has declined significantly irrespective of the measure considered (the Gini coefficient, as well as the GE(0) and GE(1) measures4 are shown). Median inequality across states has also fallen during the period. Moreover, Azevedo and others (2014) show that for the GE(0), GE(1), and GE(2) indicators, the “between states dimension” accounts for only 3–8 percent of total inequality in Brazil and remains relatively constant over time. Therefore, the analysis presented in subsequent sections of this chapter focuses on within-state inequality.

Figure 22.1Income Inequality in Brazil, 1995–2011

Source: Authors’ calculations.

Note: GE refers to the generalized entropy index of inequality. GE(0) and GE(1) are measures of the index taking the parameter alpha to be 0 and 1, respectively.

As mentioned, the first decade of the 2000s was a time of fiscal adjustment for Brazilian states. The median primary balance moved to surplus in the late 1990s, and hovered around 1 percent of state GDP for most of the subsequent decade, before falling to near zero after the global financial crisis (Figure 22.2, panel 2). In contrast, the federal government primary balance has been in surplus since 1995 (Figure 22.2, panel 1).

Figure 22.2Evolution of Fiscal Balances and Fiscal Stance, 1995–2011

Source: Authors’ calculations.

An emerging literature links the cyclicality of fiscal policy to social outcomes. Vegh and Vuletin (2014) use evidence from a number of Latin American and European countries to show a causal link from countercyclical (procyclical) fiscal policy at the national level and improvement (deterioration) in social indicators, including inequality indicators; that is, countercyclical policies are expected to reduce inequality.

This chapter follows their approach and finds that the overall correlation between the cyclical component of real government expenditure at the state level and the output gap is very weak, pointing to an essentially acyclical fiscal policy at the state level during the period (Figure 22.2, panel 3). However, considerable variation in the cyclicality of policies can be discovered by looking at the state-by-state correlation between the cyclical component of real government expenditures and the output gap (Figure 22.2, panel 4). In fact, 14 of the 27 states demonstrate a positive correlation between expenditures and the output gap, suggesting procyclical policies.

Figure 22.3 presents simple scatter plots including the log of the Gini coefficient, the level of the primary fiscal balance at the state level, and changes in the cyclically adjusted primary balance, which is the preferred measure of the fiscal stance in this analysis (see discussion in the next section). Note the negative correlation between the level of the primary balance and the Gini for the entire sample. This negative link seems to be stronger before the enactment of the Fiscal Responsibility Law (in a comparison of panels 2 and 3 of Figure 22.3). In fact, the association between the level of the primary balance and the Gini practically disappears if only the period after 2000 is considered. Finally, the simple correlation between the Gini and changes in the cyclically adjusted primary balance (the measure of fiscal adjustment in this analysis) is also weak; there is no indication that fiscal adjustment is associated with increases in inequality at the subnational level.

Figure 22.3Scatter Plots of Inequality and Primary Balances

Source: Authors’ calculations.

Modeling Approach and Results

This investigation follows an empirical specification summarized in equation (22.1) for i = 1, …, N states, t = 1, …, T time periods, and m = 1, …, M control variables. The variable yi, t represents the log of the Gini index at the state level; Δpb are changes in the cyclically adjusted primary balance as a share of state GDP (a measure of the fiscal stance); X is a set of controls; αi are state-specific fixed effects; and εit is the error term, assumed to be white noise. The estimation of equation (22.1) also includes time dummies and a time trend in the specification.

The following macroeconomic and fiscal variables are considered as controls in the baseline regressions: the employment rate, the state-level GDP per capita growth rate, inflation (measured by changes in the state GDP deflator), state and municipal social expenditure as a share of state GDP, and federal social transfers as a share of state GDP. To mitigate possible endogeneity problems, lagged values of the control variables are used and these control variables are therefore assumed to be weakly exogenous. Cross-sectional dependence problems are addressed by including time effects in the regressions and by using Driscoll and Kraay (1998) corrected standard errors.

When constructing the cyclically adjusted primary balance we focus on adjustments of revenue and expenditure to the output gap and do not consider the impact of asset or commodity price fluctuations. This is justified given the results discussed in IMF 2011 that suggest that, at the national level, revenue elasticities relative to cycles in equity and commodity prices appear to be very small. The analysis follows the “aggregated method” described in Bornhorst and others 2011 and focuses on the sensitivity of aggregate revenues and expenditures to the output gap at the state level. The revenue and expenditure elasticities estimated by Arena and Revilla (2009) for Brazilian states for the period 1991–2006 are used. These authors estimate a total revenue elasticity of about 1.7 and total expenditure elasticity of 1.3.

This chapter’s measure of the fiscal stance differs from the action-based fiscal consolidation measures constructed by Devries and others (2011) and used in several papers in the literature on fiscal consolidation and inequality for OECD economies (Ball and others 2013). The construction of a similar action-based measure for Brazilian states is more difficult given that historical policy documents providing information on discretionary changes in taxes and expenditures are not readily available at the subnational level. Furthermore, this examination also considers the effects of gradual continuous adjustment rather than focusing on large consolidation episodes, as is frequently done in the literature (Alesina and Ardagna 2012).

A structurally adjusted primary balance (that is, the primary balance adjusted not only for cyclical fluctuations but also for one-off fiscal operations) would provide a better measure of the fiscal stance. However, lack of information and accounting challenges make it impossible to remove one-off fiscal operations at the state level in a consistent and systematic way.

Results for the Baseline Specification

The results presented in Table 22.1 indicate that the employment rate is an important factor in explaining inequality for the period. Furthermore, both real GDP per capita growth and the inflation rate are linked to increases in inequality, but only the coefficients obtained for the inflation rate are statistically significant, mirroring some of the results of the literature. These findings highlight the importance of macroeconomic stabilization in inequality reduction, a well-established result in the Brazilian context.

Table 22.1Fixed Effects Regressions, 1995–2011
12345678
GiniGiniGiniGiniGiniGiniGiniGini
Lagged Dependent Variable0.455***0.455***0.430***0.433***0.425***0.422***0.420***0.415***
[0.100][0.100][0.096][0.096][0.098][0.094][0.098][0.099]
Δ(Cyclically Adjusted Primary Surplus/GDP)t-1−0.181**−0.182**−0.179**−0.176**−0.162**−0.153**−0.153**−0.157**
[0.067][0.066][0.071][0.068][0.064][0.060][0.061][0.062]
(GDP Growth per Capita)t-10.0250.0210.0370.0350.0490.0500.044
[0.071][0.074][0.078][0.076][0.074][0.074][0.077]
(Employment Rate)t-1−0.284**−0.294**−0.251*−0.307**−0.301**−0.272**
[0.102][0.105][0.118][0.119][0.117][0.118]
(Inflation)t-10.073**0.076**0.076**0.076**0.078***
[0.028][0.027][0.028][0.028][0.026]
(Subnational Social Expenditure)t-10.217**0.348***0.345***0.348***
[0.077][0.073][0.074][0.074]
(Federal Social Transfers/GDP)t-1−1.073***−1.036**−0.915**
[0.344][0.369][0.388]
(Years of Education)t-1−0.0070.027
[0.018][0.020]
(Returns to Education)t-1−0.094
[0.070]
Constant0.0000.0000.0000.0000.0000.0000.0000.000
[0.000][0.000][0.000][0.000][0.000][0.000][0.000][0.000]
Time DummiesYesYesYesYesYesYesYesYes
Time TrendYesYesYesYesYesYesYesYes
Pesaran (2004) Cross-Sectional Dependence Test−1.87*−1.89*−2.05**−2.04**−2.19**−2.28**−2.28**−2.30**
Number of Observations405405405405405405405405
Number of Groups2727272727272727
R20.5890.5890.6060.6090.6140.6210.6210.622
Source: Azevedo and others (2014).Note: Driscoll-Kraay standard errors in brackets. Null hypothesis of Pesaran (2004) test for regression residuals is cross-sectional independence. Time effects coefficients not reported to save space.*** p < 0.01, ** p < 0.05, * p < 0.1.

More important for the purposes of this volume, fiscal variables seem to matter. Somewhat intriguingly, a tighter fiscal stance is associated with less inequality. In addition, as expected, the highly progressive social transfers at the federal level are strongly linked to reductions in inequality with economically large coefficients. Social expenditures at the state and municipal levels appear to be associated with higher inequality, with positive and statistically significant coefficients for all specifications, in line with the results obtained by Lim and McNelis (2014)5 for Latin American countries and with results by Ferreira, Leite, and Ravallion (2010) for Brazil.

The link between the cyclically adjusted primary balance and inequality remains statistically significant when average years of education and average returns to education at the state level are controlled for (specifications 7 and 8 in Table 22.1). The short-term coefficients for the effects of changes in the primary surplus as a share of GDP on inequality range from -0.18 to -0.15 (the long-term effects are about -0.4). These coefficients can be interpreted as semi-elasticities.

It is possible that the results obtained with fixed effects specifications are due to shortcomings in statistical techniques. Hence, the models are reestimated using generalized method of moments (GMM) techniques, namely the system (Blundell-Bond) GMM estimator,6 which allow the potential endogeneity of some regressors to be handled by using lagged values of these variables as instruments. We transform instruments using forward orthogonal deviations and present robust standard errors, which are consistent in the presence of heteroscedasticity and autocorrelation. The bias introduced by high instrument count is mitigated by replacing instruments with their principal components.

The results obtained (Table 22.2) confirm the association between inequality measures, the employment rate, and federal social transfers, as well as the link between the cyclically adjusted primary balance and reductions in inequality. However, the coefficient for the inflation rate is no longer statistically significant. The estimated coefficient (semi-elasticity) for changes in the primary balance continues to be statistically significant and is larger than what was obtained in the fixed effects regressions, ranging from about -0.26 to -0.51.

Table 22.2System (Blundell-Bond) Generalized Method of Moments Regressions, 1995–2011
GiniGiniGiniGiniGiniGiniGiniGini
Lagged Dependent Variable0.595***0.670***0.648***0.654***0.543***0.624***0.565***0.593***
[0.096][0.093][0.081][0.077][0.107][0.094][0.122][0.120]
Δ(Cyclically Adjusted Primary Surplus/GDP)t-1−0.512***−0.390***−0.442***−0.392***−0.256**−0.309**−0.257**−0.292**
[0.098][0.117][0.100][0.110][0.103][0.111][0.119][0.116]
(GDP Growth per Capita)t-10.0150.033−0.086−0.119−0.099−0.072−0.065
[0.107][0.118][0.096][0.081][0.090][0.095][0.090]
(Employment Rate)t-1−0.329**−0.283*−0.206**−0.316***−0.294***−0.288***
[0.155][0.149][0.091][0.085][0.071][0.078]
(Inflation)t-10.0210.0180.0200.0120.027
[0.058][0.061][0.064][0.057][0.058]
(Subnational Social Expenditure)t-10.286**0.337**0.368**0.377***
[0.125][0.136][0.140][0.135]
(Federal SocialTransfers/GDP)t-1−0.761**−1.108**−1.109***
[0.358][0.403][0.381]
(Years of Education)t-1−0.001−0.008
[0.029][0.029]
(Returns to Education)t-10.016
[0.057]
Constant−5.302−5.172−5.92412.53825.931**19.51918.85913.879
[8.187][15.363][16.386][14.812][11.323][12.777][14.220][17.910]
Time DummiesYesYesYesYesYesYesYesYes
Time TrendYesYesYesYesYesYesYesYes
Pesaran (2004) Cross-Sectional Dependence Test−1.99**−1.90*−2.00**−1.91*−2.08**−2.08**−2.16**−2.16**
Sargan Test98.24***153.3***159.9***171.8***165.1***181.7***173.1***177.0***
Hansen Test10.7012.3316.4114.738.5192.8732.1421.627
Arellano-Bond AR(2) Test0.3540.5630.5470.8370.6570.6030.4990.515
Number of Observations405405405405405405405405
Number of Groups2727272727272727
Source: Azevedo and others (2014).Note: Heteroscedasticity and autocorrelation consistent (HAC) robust standard errors in brackets. Null hypothesis of Pesaran (2004) test for regression residuals is cross-sectional independence. Null of Arellano-Bond test is that first-differenced errors exhibit no second-order serial correlation. Sargan and Hansen tests of the validity of overidentifying restrictions. GMM = generalized method of moments.*** p < 0.01, **p < 0.05, *p < 0.1.

Diagnostic tests reject the null of cross-sectional independence in the residuals for most specifications, despite the inclusion of time dummies; therefore, the results should be interpreted with caution. In addition, autoregression tests do not indicate the presence of serial correlation of the residuals. As far as the validity of instruments is concerned, the Hansen test suggests that overidentifying restrictions are valid, but the Sargan test rejects the validity of these restrictions. However, one should bear in mind that the Sargan test statistic is not robust to heteroscedasticity or serial correlation.

Azevedo and others (2014) also present a number of robustness checks on the baseline specification. The authors start by considering additional control variables, including the share of prime-age workers in the informal sector, the share of employment in agriculture, the share of employment in manufacturing, as well as some demographic variables (the dependency ratio and the average household size), and the share of employment in all sectors (excluding manufacturing). None of the additional control variables present statistically significant coefficients, but otherwise the results are broadly similar to the ones obtained previously.

Moreover, Azevedo and others (2014) also consider specifications that include the log of the average labor income of low-skilled workers (defined as workers with eight years of education or less), the log of average labor income of high-skilled workers, and the total labor income of low-and high-skilled workers as control variables. As expected, the coefficient for average earnings of high-skilled workers is positive and significant, and the earnings for low-skilled workers present a negative coefficient. More important for the purposes of this chapter, the coefficient for the primary balance continues to be negative and significant in all specifications, although its statistical significance is reduced to the 10 percent level in fixed effects regressions (but not in GMM ones).

Finally, alternative estimation methods and different modeling of deterministic components were also tried. Random effects models do not perform well (in the sense that most variables are not statistically significant), but the link between changes in the cyclically adjusted primary balance and inequality remains negative and statistically significant. Models that allow both the intercept and slope coefficients to vary across panel members were also considered, following the random coefficient model proposed by Swamy (1970). For these models, the coefficient for the primary balance continues to be negative, but is no longer statistically significant.

Specifications that model deterministic components (that is, time trends and time effects) in a different way were also tried. Regressions were estimated including national, regional, and statelevel polynomial time trends, and results are qualitatively similar to those already presented. Overall, the robustness checks confirm that there is no evidence that fiscal adjustment at the subnational level is positively linked to inequality in Brazil.

Disentangling the Effects of Changes in Expenditures and Changes in Revenue

A number of papers in the literature on fiscal consolidation at the national level tend to find different effects on inequality depending on whether the consolidation is expenditure based or revenue based (Agnello and Sousa 2012). This section explores these differential channels for the Brazilian case by separately including changes in revenues and changes in primary expenditures in the regressions.

The evidence at the national level for Brazil and other Latin American countries points to a greater reliance on indirect (regressive) taxes relative to income taxes, which would suggest that higher primary surpluses driven by higher tax revenues would tend to be associated with increases in inequality (Bastagli, Coady, and Gupta 2012; Goñi, López, and Servén 2011). However, the evidence at the national level also indicates that a large share of social spending is captured by the better off, and thus reductions of these expenditures would not necessarily lead to increases in inequality.

Before the regression results are presented, it is important to note that revenues played an important relative role in those years in which the state-level primary surpluses experienced a positive change. Expenditures, however, have been relatively more relevant in those years in which the state-level primary surpluses experienced negative changes. Hence, fluctuations of the overall fiscal adjustment process have embedded in them changing roles for revenue and expenditure. Also, from a descriptive perspective, the fiscal adjustment process at the state level appears to have had a relatively stronger revenue-side component in aggregate, though this pattern can vary significantly by state.

The regression results are presented in Table 22.3. As before, aggregate revenues and expenditures are adjusted to the cycle using the elasticities estimated by Arena and Revilla (2009), but because of the considerable uncertainty regarding these disaggregated elasticity estimates specific revenue and expenditure items are not adjusted. In fixed effects regressions (specification 1 in Table 22.3), changes in revenues are negatively associated with inequality, whereas changes in primary expenditures show a positive association. Further disaggregation of revenues and expenditures suggests that these results are driven by changes in revenues linked to revenue transfers to states7 and by changes in investment expenditure (specification 2), with coefficients for these variables being significant at the 1 percent level. These relationships are robust in GMM regressions as well; however, only changes in primary expenditures are statistically significant.

Table 22.3Effects of Changes in Expenditures and Changes in Revenue, 1995–2011
1234
Fixed EffectsFixed EffectsSystem GMMSystem GMM
GiniGiniGiniGini
Lagged Dependent Variable0.421***0.423***0.642***0.774***
[0.091][0.093][0.126][0.097]
Δ(Primary Revenues)t-1−0.184***−0.260
[0.058][0.162]
Δ(Primary Expenditure)t-10.149**0.350***
[0.068][0.124]
Δ(Tax Revenue)t-1−0.037−0.262
[0.242][0.268]
Δ(Revenue Transfers)t-1−0.293**−0.314
[0.104][0.223]
Δ(Other Revenue)t-10.0490.101
[0.114][0.258]
Δ(Current Expenditure)t-10.0780.407*
[0.115][0.232]
Δ(Investment Expenditure)t-10.215***0.263**
[0.057][0.117]
Δ(Other Expenditure)t-10.1970.194
[0.158][0.365]
(GDP Growth per Capita)t-10.0420.049−0.027−0.039
[0.068][0.075][0.082][0.095]
(Employment Rate)t-1−0.309**−0.311**−0.182**−0.103
[0.120][0.112][0.077][0.086]
(Inflation)t-10.074**0.064*0.0510.065
[0.025][0.031][0.058][0.066]
(Subnational Social Expenditure)t-10.356***0.346***0.2510.133
[0.076][0.082][0.161][0.136]
(Federal Social Transfers/GDP)t-1−1.083***−1.091***−0.386−0.104
[0.349][0.354][0.350][0.273]
Constant0.0000.0009.5607.317
[0.000][0.000][12.738][16.464]
Time TrendYesYesYesYes
Time DummiesYesYesYesYes
Pesaran (2004) Cross-Sectional Dependence Test−2.29**−2.26**−2.07**−1.84*
Sargan Test210.3***293.0***
Hansen Test1.6428.002
Arellano-Bond AR(2) Test0.7280.792
Number of Observations405405405405
Number of Groups27272727
R20.6210.626
Source: Authors’ calculations.Note: Heteroscedasticity and autocorrelation consistent (HAC) robust (GMM regressions) or Driscoll-Kraay (fixed effects regressions) standard errors in brackets. Null hypothesis of Pesaran (2004) test for regression residuals is cross-sectional independence. Null of Arellano-Bond test is that first-differenced errors exhibit no second-order serial correlation. Sargan and Hansen tests of the validity of overidentifying restrictions. GMM = generalized method of moments.*** p < 0.01, ** p < 0.05, * p < 0.1.

In this context, it can be concluded that revenue increases in Brazilian states were not typically linked to increases in inequality during the period of analysis. Furthermore, reductions in primary expenditures also do not seem to have had deleterious impacts on inequality measures. The behavior of revenue transfers to states and investment expenditure seems to be particularly important in explaining these results. Changes in revenues linked to revenue transfers are associated with reductions in within-state income inequality, though between-state inequality has remained broadly stable (as discussed previously). The rules that govern federal revenue transfers to states in Brazil favor poorer states (as measured by GDP per capita), which tend to present higher inequality indicators as well.

Changes in investment expenditure also seem to have played a role in inequality dynamics, but in the direction of increasing inequality. As argued by Lim and McNelis (2014), capital spending (especially infrastructure spending) enhances the returns to capital and might contribute to an increase in inequality. In Brazil, public investment does appear to have an infrastructure bias.

Finally, it is important to note that the scope for efficiency gains in revenue mobilization at the state level a decade ago was substantial, so it would have been plausible for states to raise revenues without necessarily exacerbating existing inefficiencies. In addition, current public spending could also have been captured by the better off, and thus controlling the growth of these expenditures would not necessarily lead to increases in inequality. Overall, the results of this analysis are consistent with the view that the observed fiscal adjustment process has contributed to a more judicious and efficient use of public resources.

One important caveat regarding the disaggregated results is that the quality of the fiscal information deteriorates as one moves into a higher degree of desegregation of revenues and expenditures. This arises from the potential misclassification of data and lack of harmonized classification practices across states, an issue that is difficult for the federal government to resolve alone.

Conclusion

This chapter finds that a tighter fiscal stance in Brazilian states, measured by changes in the cyclically adjusted primary balance, is not associated with an increase in inequality during the period 1995–2011. This conclusion is in contrast to the results of several papers that analyze the impact of fiscal consolidations at the national level for OECD countries (Ball and others 2013). The results also suggest that revenue increases in Brazilian states were not associated with increases in inequality. Similarly, reductions in primary expenditures do not seem to have had deleterious impacts on inequality measures. Further disaggregation indicates that revenue increases due to revenue transfers to states are linked to decreases in inequality, whereas changes in investment expenditure are positively linked to inequality measures.

The different conclusions obtained in this chapter with respect to the rest of the literature could be explained by the fact that several of these studies employ measures of fiscal adjustment that are different from the ones used here. The differences could also be linked to the definition of income used to measure inequality. Some studies use measures based on disposable income, while income after transfers and before taxes is used for Brazil.

Nevertheless, the bulk of the difference in the results is likely to be explained by differences in structural characteristics (fiscal, social, and economic) of Brazilian states compared with OECD economies. The chapter does not attempt to establish the precise mechanism linking fiscal policy and inequality, but possible differences driving the result include higher initial levels of inequality, lesser reliance on progressive taxation, the absence of extensive social safety nets and other automatic stabilizers, scope to significantly improve the efficiency of public spending and the quality of public services, and the regressive nature of some forms of public expenditure at the state level. Furthermore, fiscal adjustment at the state level might also have been achieved through efficiency gains in revenue collection with no discernible impact on inequality.

Finally, it is worth noting that all measures of inequality used in this chapter came from the same survey, which is conducted using the exact same questionnaire, with the same field work protocols, and using the same period of reference. As Beegle and others (2012) rigorously demonstrate, survey design and implementation effects can have significant impacts on the final indicators, which can confound any cross-country analysis in this field.

Future research could focus on drilling down on the mechanisms linking the fiscal stance and inequality dynamics. This would be important to ascertaining how the relationship is expected to evolve as the country’s macroeconomic and social conditions change, thus more precisely informing policymaking. A central message, however, is that the results linking fiscal adjustment to an increase in inequality in advanced economies cannot be easily generalized to developing countries, given the Brazilian experience.

Annex 22.1 Variable Definitions and Sources
VariableDescription and NotesSource
Income InequalityComprises different measures of inequality in household income per capita (after transfers, but before taxes) including the log of the Gini coefficient and the generalized entropy indices GE(0) and GE(1)Authors’ calculations based on PNAD data
Employment RateShare of employed population at prime working age (25–65 years) by stateAuthors’ calculations based on PNAD data
State GDP Growth per CapitaLog of change in real GDP per capita at the state levelAuthors’ calculations based on IBGE data
InflationChange in GDP deflator at the state level; log of (1 + (state inflation)/100).Authors’ calculations based on IBGE data
Cyclically Adjusted Primary BalanceSee main text for details of variable construction. State-level primary balance (revenues minus expenditures net of interest payments) as a share of state GDPAuthors’ calculations based on Treasury Department’s database
Subnational Social ExpendituresSum of state and municipal expenditures on education and culture; health and sanitation; and social security and social assistance as a share of state GDPAuthors’ calculations based on Treasury Department’s database
Federal Social TransfersComprises information at the state level on three main federal social programs: Bolsa Familia, Beneficio de Prestacao Continuada, and Renda Mensal Vitalicia; values in the data set are for December of each year and have been multiplied by 12 to obtain annual figuresIPEADATA (www.IPEADATA.org)
Years of EducationLog of average years of education of prime working-age individuals (25–65 years) by stateAuthors’ calculations based on PNAD data
Returns to EducationReturns to education at prime working ageAuthors’ calculations based on PNAD data
Note: IBGE = Institute of Geography and Statistics; PNAD = Pesquisa Nacional por Amostra de Domicilios.
Annex 22.2 Descriptive Statistics for Selected Variables
VariableMeanStandard DeviationMinimumMaximumObservations
Log of Gini Coefficientoverall4.0060.0723.7624.182N = 459
between0.0553.8674.105n = 27
within0.0483.8354.109T = 17
GE(0)overall0.5550.0880.3260.807N = 459
between0.0700.4030.712n = 27
within0.0540.3970.710T = 17
Shared Prosperityoverall0.0970.101−0.4330.526N = 432
between0.0170.0470.123n = 27
within0.099−0.3840.554T = 16
GDP per Capita Growthoverall0.0130.040−0.1440.144N = 405
between0.0070.0000.033n = 27
within0.039−0.1490.140T = 15
Employment Rateoverall0.6570.0420.5190.765N = 459
between0.0350.5690.710n = 27
within0.0240.5220.731T = 17
Inflationoverall0.0830.046−0.0780.269N = 405
between0.0090.0630.099n = 27
within0.045−0.0610.254T = 15
Subnational Social Expenditure/GDPoverall0.1230.0470.0350.258N = 459
between0.0400.0520.204n = 27
within0.0260.0330.188T = 17
Cyclically Adjusted Primary Balanceoverall0.0020.020−0.1250.079N = 432
between0.006−0.0080.017n = 27
within0.019−0.1180.066T = 16
Federal Social Transfers/GDPoverall0.0110.0110.0000.054N = 459
between0.0080.0020.026n = 27
within0.008−0.0070.039T = 17
Years of Educationoverall1.8400.2131.2802.911N = 459
between0.1631.5532.193n = 27
within0.1401.4352.941T = 17
Returns to Educationoverall0.7330.1330.4341.143N = 459
between0.1230.5411.052n = 27
within0.0570.5270.969T = 17
Source: Authors’ calculations.
References

    Agnello, L., and R. M.Sousa.2012. “How Does Fiscal Consolidation Impact on Income Inequality.Banque de France Working Paper No. 382, Paris.

    Alesina, A., and S.Ardagna.2012. “The Design of Fiscal Adjustments.NBER Working Paper No. 18423, National Bureau of Economic Research, Cambridge, Massachusetts.

    Arena, M., and J. E.Revilla.2009. “Pro-Cyclical Fiscal Policy in Brazil: Evidence from the States.World Bank Policy Research Working Paper No. 5144, World Bank, Washington.

    Azevedo, J. P., M. E.Davalos, C.Diaz-Bonilla, B.Atuesta, and R.Castaneda.2013. “Fifteen Years of Inequality in Latin America: How Have Labor Markets Helped?World Bank Policy Research Working Paper No. 6384, World Bank, Washington.

    Azevedo, J. P., A. C.David, F.Rodrigues Bastos, and E.Pineda.2014. “Fiscal Adjustment and Income Inequality: Sub-National Evidence from Brazil.IMF Working Paper No. 14/85, International Monetary Fund, Washington.

    Ball, L., D.Furceri, D.Leigh, and P.Loungani.2013. “The Distributional Effects of Fiscal Consolidation.IMF Working Paper No. 13/151, International Monetary Fund, Washington.

    Bastagli, F., D.Coady, and S.Gupta.2012. “Income Inequality and Fiscal Policy.IMF Staff Discussion Note No. 12/08, International Monetary Fund, Washington.

    Beegle, K., J.De Weerdt, J.Friedman, and J.Gibson.2012. “Methods of Household Consumption Measurement through Surveys: Experimental Results from Tanzania.Journal of Development Economics98 (1): 318.

    Bornhorst, F., G.Dobrescu, A.Fedelino, J.Gottschalk, and T.Nakata.2011. “When and How to Adjust beyond the Business Cycle? A Guide to Structural Fiscal Balances.IMF Technical Note and Manuals No. 11/02, International Monetary Fund, Washington.

    Devries, P., J.Guajardo, D.Leigh, and A.Pescatori.2011. “A New Action-Based Dataset of Fiscal Consolidation.IMF Working Paper No. 11/128, International Monetary Fund, Washington.

    Driscoll, J. C., and A. C.Kraay.1998. “Consistent Covariance Matrix Estimation with Spatially Dependent Panel Data.Review of Economics and Statistics80 (4): 54960.

    Ferreira, F., P. G.Leite, and M.Ravallion.2010. “Poverty Reduction without Economic Growth? Explaining Brazil’s Poverty Dynamics, 1985–1994.Journal of Development Economics93 (1): 2036.

    Goñi, E., H.López, and L.Servén.2011. “Fiscal Redistribution and Income Inequality in Latin America.World Development39 (9): 155869.

    International Monetary Fund (IMF). 2011. “Brazil: Selected Issues Paper.International Monetary Fund, Washington.

    Lim, G. C., and P. D.McNelis.2014. “Income Inequality, Trade and Financial Openness.Paper presented at the conference “Macroeconomic Challenges Facing Low-Income Countries,”Washington, January30–31.

    Lopez-Calva, L. F., and S.Rocha.2012. “Exiting Belindia? Lessons from the Recent Decline in Income Inequality in Brazil.Poverty, Equity and Gender Unit, Latin America and the Caribbean, World Bank, Washington.

    Lustig, N., CarolaPessino, and JohnScott.2012. “The Impact of Taxes and Social Spending on Inequality and Poverty in Argentina, Bolivia, Brazil, Mexico, and Peru: A Synthesis of Results.Working Paper No. 1313, Tulane University, New Orleans, Louisiana.

    Pecora, A. R., and N.Menezes.2013. “O Papel Da Oferta E Da Demanda Por Qualificação Na Evolução Do Diferencial De Salários Por Nível Educacional No Brasil.Texto para Discussão USP. São Paulo.

    Pesaran, H.2004. “General Diagnostic Tests for Cross Section Dependence in Panels.Cambridge Working Papers in Economics No. 0435, University of Cambridge.

    Roodman, D.2009. “How to Do xtabond2: An Introduction to ‘Difference’ and ‘System’ GMM in Stata.Stata Journal9 (1): 86136.

    Soares, S., F. G.Silveira, C. H.dos Santos, F. M.Vaz, and A. L.Souza.2009. “O Potencial Distributivo do Imposto de Renda-Pessoa Física (IRPF).IPEA Texto para Discussão (Working Paper) No. 1433, Rio de Janeiro.

    Sturzenegger, F., and R. L. F.Werneck.2006. “Fiscal Federalism and Procyclical Spending: The Cases of Argentina and Brazil.Económica52 (1–2): 15194.

    Swamy, P.1970. “Efficient Inference in a Random Coefficient Regression Model.Econometrica38 (2): 31123.

    Vegh, C. A., and G.Vuletin.2014. “Social Implications of Fiscal Policy Responses during Crises.NBER Working Paper No. 19828, National Bureau of Economic Research, Cambridge, Massachusetts.

    Woo, J., E.Bova, T.Kinda, and Y. S.Zhang.2013. “Distributional Consequences of Fiscal Consolidation and the Role of Fiscal Policy: What Do the Data Say?IMF Working Paper No. 13/195, International Monetary Fund, Washington.

This chapter is largely based on the analysis presented in Azevedo and others 2014. The views expressed in this chapter are those of the authors and do not necessarily represent those of the IMF or IMF policy or those of the World Bank and World Bank policy. The authors thank Benedict Clements and Maura Francese for excellent comments and suggestions.

Brazil is organized politically and administratively as a federal system consisting of 26 states and 1 federal district. The states are characterized by heterogeneous levels of inequality and fiscal outcomes, but share common institutions and federal regulations.

Indirect taxes are estimated by Soares and others (2009) to amount to about 10 percent of GDP.

See Azevedo and others 2014 for a richer and more detailed description of the data and of the construction of the variables.

GE refers to the generalized entropy index of inequality. We present two measures of the index taking the parameter alpha to be 0 and 1, respectively. GE(1) is the so-called Theil index.

Note that Lim and McNelis (2014) focus on overall spending rather than social spending, and they use country-level data.

See Roodman 2009 for a discussion.

Note that the federal revenue transfers go to state governments, whereas federal social transfers also considered in the regressions are direct cash transfers to households; thus, these components have very different implications for inequality. Revenue transfers from the federal government to states are made according to revenue-sharing mechanisms. The bulk of revenue transfers originate from the “States’ Participation Fund,” which comprises revenues from income taxes and the IPI tax on manufactured products (21.5 percent of the revenues linked to these taxes is allocated to the Fund). States with lower GDP per capita receive a relatively larger share of the transfers.

    Other Resources Citing This Publication