- Email: [email protected]
Tax burden, government expenditures and income distribution in Brazil
The Quarterly Review of Economics and Finance 48 (2008) 345–358
Tax burden, government expenditures and income distribution in Brazil Werner Baer ∗ , Antonio Fialho Galv˜ao Jr. University of Illinois at Urbana-Champaign, United States Received 13 December 2006; received in revised form 13 December 2006; accepted 13 December 2006 Available online 25 March 2007
Abstract This article attempts to explain the seeming paradox of a country with a high tax burden and a continually concentrated distribution of income. By means of a structural quantile regression model we analyze the distributional impact of government expenditures on the Gini index, and it is shown that Brazil’s redistribution expenditures has a relatively smaller impact for low quantiles of the conditional distribution of income inequality. It is also noted that both the country’s tax and expenditure structure are, in part, responsible for the country’s continuous concentration of income. © 2007 Board of Trustees of the University of Illinois. All rights reserved. Keywords: Income distribution; Taxes; Government expenditures; Brazil
1. Introduction Toward the end of the 19th century the German political theorist Adolph Wagner devised his law of expanding state activity, also known as Wagner’s Law. This law stated that the size of the public sector in the economy grows as per capita income rises. Although this “law” was somewhat controversial, the data show that there is a tendency for government expenditures as a share of GDP to be larger for rich than for poor countries. For instance the World Bank’s World Development Report of 1994 showed that central government expenditures as a share of GDP was 17% for low-income countries and 32% for high-income countries. Analysts have also noted that wealthy countries “. . . have larger social welfare programs that cause subsidies and other transfers to be a much higher share of expenditures than in most lower-income countries.”1 ∗ 1
Corresponding author. Tel.: +1 217 333 8388; fax: +1 217 333 1398. E-mail address: [email protected] (W. Baer). Perkins et al. (2001), p. 423.
1062-9769/$ – see front matter © 2007 Board of Trustees of the University of Illinois. All rights reserved.
doi:10.1016/j.qref.2006.12.012
346
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Similarly, in his review of world fiscal systems Musgrave noted that “. . . the rise in the overall expenditure to GNP ratio has been due primarily to the growth of social services. This reflects in particular a growing re-distributional concern, rather than a rising demand for social goods as such. It appears that the underlying forces of social and political change greatly outweighed any tendency . . . for the transfer share to fall with rising per capita income.”2 Furthermore, he found that the “. . . rise in the relative importance of social services, which make heavy use of transfer payments, suggests a considerable increase in the importance of transfer payments relative to public purchases.”3 One would thus expect to find the tax burden of most low income countries to be lighter than the tax burden of wealthy countries, and at the same time that the distribution of income to be more equitable in higher than in lower income countries. It is thus striking to find that in Brazil, one of the major emerging countries, the tax burden is similar to that of many advanced industrial countries, its income distribution is among the most concentrated in the world. It is the purpose of this article to throw some light on this seeming paradox. Some previous studies focused on some of the elements which might lead to an explanation of the determinants of inequality linked to the fiscal system, especially its social expenditures. For instance, De Mello and Tiongson (2006) cross-section study found that “. . . re-distributive spending may be inefficient as an instrument to reduce poverty and to improve income distribution because the benefits of public spending may be captured by the non-poor.” (p. 303). Carneiro, De Mello, and Tiongson (2002) made a study of 37 industrial and developing countries in 1972–1987 and found that “. . . the countries where redistributive public spending is more needed were found to be the ones that are less likely to redistribute income through public policies.” (p. 103). Another recent study by Siqueira, Nogueira and Levy (2002) attempted to identify the types of households which benefit from federal government expenditures in Brazil. They came to the conclusion that Brazil’s tax and benefits system is deceptively small given the substantial amount of resources involved. Clements (1997) found that government social expenditures in Brazil have exacerbated income inequalities. Another set of studies (Barros & Foguel, 2000; Hoffmann, 2003; Ramos, 2000; von Amsberg et al., 2000) showed that one of the reasons that the Brazilian government did not succeed in eliminating poverty was due to the inadequate targeting of public expenditures. Within this framework we shall expand the study of the relation of the fiscal structure to the distribution of income. We examine some evidence with regard to the distribution of the tax burden4 of the tax system in Brazil. In addition, we estimate the impact of government expenditures on the country’s income distribution5 using a structural quantile regression model. An examination of descriptive statistics on taxes and government expenditures give strong evidence that both the tax burden and government expenditures favor the higher income classes, which means that the country’s fiscal system has a relatively low redistributional impact. In addition, the results of estimates of the effects of government expenditures on income inequality, using municipality data, present evidence that in order to reduce income inequality, the government would have to emphasize in its expenditure patterns those programs which benefit more municipalities with high-income inequality.
2
Musgrave (1969), pp. 93–96. Musgrave (1969) p. 96. 4 By “tax burden” is meant the ratio of total taxes paid to the GDP. 5 We are aware that the country’s tax structure may also contribute to the concentration of income, which we discuss below. 3
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
347
Fig. 1. Tax burden vs. Gini coefficient.
The article is organized as follows. Section 2 presents an overview of tax burden and income inequality around the world. Section 3 presents some evidence for Brazil’s tax burden and government expenditure. Section 4 proposes an econometric model of Brazil’s government expenditures and income distribution. Finally, Section 5 contains a general conclusion. 2. Tax burden and income distribution: world trends In this section we shall take a brief look at the income distribution and tax burdens prevailing in a large sample of countries around the globe. It will be noted that Brazil does not fall into the generally observed pattern, that is, the higher the tax burden, the more equitable is the distribution of income. We first plot the Gini index and the tax burden for 50 countries.6 Data for Gini coefficient is of a cross-section nature, which are available in the Human Development Report 2000. The International Monetary Fund provides data for the tax burden. We use tax burden data for 1998. The tax burden is simply total tax revenues as a share of GDP. Fig. 1 shows the scatter diagram of Gini coefficients and tax burdens. It will be noted that Brazil’s income distribution was one of the most concentrated in the world with a relatively high tax burden. This becomes even clearer in Table 1, which describes the gap between the rich and poor for selected countries. Thus the evidence presented shows Brazil as being an important case to be studied, given that it appears to have a very concentrated income and relatively high tax burden.
6
A list of the countries examined can be found in Appendix A.
348
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Table 1 Gap between rich and poor for selected countries (% share of income)
Sierra Leone Brazil Mexico United States Germany Sweden
Lowest 20% of income groups
Highest 20% of income groups
1.1 2.2 3.5 4.2 8.2 9.6
63.4 64.1 57.4 46.9 38.5 34.5
Source: http://www.infoplease.com/ipa/A0908770.htm1. Table 2 Brazil’s tax burden as % of GDP 1988
1994
2000
2002
2003
Federal State Municipal Social security
16.1 5.7 0.6
13.4 7.8 1.0 5.63
14.8 8.35 0.84 7.12
17.1 8.84 1.11 7.45
16.4 8.61 1.20 7.41
Total
22.4
27.9
31.6
34.9
34.0
2004
2006
35.9
38.0
Source: IPEA, Secretaria da Receita Federal.
3. Some evidence for Brazil’s government expenditure In this section we present some evidence of how Brazil’s fiscal system contributes to the continued concentration of income. Let us begin by examining the country’s tax burden. As is clear from Table 2, Brazil’s tax burden has grown substantially. It amounted to 22.4% in 1988 and reached 38% in 2006. This burden was disproportionately born by the country’s lower income groups, which becomes clear when looking at the origins of these taxes. It will be noted in Tables 3 and 4 that the tax system was quite regressive. In 2002 direct taxes made up only 15.7% of the federal government’s revenues, whereas indirect taxes accounted for 42.1%. Table 4 (part (a)) shows that whereas 86% of direct taxes were paid by the top three deciles, these only paid 58% of indirect taxes. In part (b) of that table, it will be noted that the indirect Table 3 Federal taxes as share of total government receipts in 2002 Total federal taxes Direct taxes Personal income tax Social security contributions (employees) Indirect taxes Tax on industrial products (IPI) Contributions to the financing of social securitya Social integration program (PIS) and program for public servants (PASEP) Education salary Social security contributions of firms Source: Lei Orc¸amentaria (2002) and Rezende and Cunha (2002). a Confins.
57.8% 15.7% 7.7% 8.0% 42.1% 7.4% 16.7% 4.1% 1.2% 12.7%
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
349
Table 4 1
2
3
4
5
7
8
9
10
Total
3 0 6 7
5 0 10 10
7 0 14 13
13 3 22 16
66 97 38 29
100 100 100 100
5
6
9
11
16
40
100
(b) Taxes as a % of gross income by income groups (deciles) Direct taxes 0 1 2 3 4 Indirect taxes 22 17 16 14 14
4 13
5 13
5 12
6 11
10 8
(a) Distribution of tax receipts by income groups (deciles) (%) Direct taxes 0 1 1 2 2 IRPF 0 0 0 0 0 Soca 0 1 2 3 4 Indirect taxes 3 4 5 6 7 Total
2
3
4
4
6
Source: Rezende and Cunha (2002), p. 104 (for part (a)) and Rezende and Cunha (2002), p. 106 (for part (b)). a Social security contributions of employees.
tax burden falls more heavily on the lower than on the upper income groups. Thus, though the general tax burden of Brazil was high, the proportion of taxes born by the lower income groups was even higher. Taxes are only part of the story, as the beneficiaries of government expenditures also need to be examined. How do these expenditures affect the distribution of income? It will be noted in Table 5 (part (a)) that in 2005 over half of government expenditures were for amortization of the debt and debt servicing, which are not of a redistributive nature. Social security expenditures and “Other Current Expenditures” category, which includes health, education, etc. and which might be thought of as having a redistributive function, constituted about 18–20% of government expenditures in the years 2002–2005. However, even this category, when examined more carefully, will show that its redistributive function is extremely limited in the case of Brazil. For instance, another way of examining government expenditures is presented in Table 5 (part (b)). It shows in a more disaggregate way for the year 2002 the distribution of federal benefits as a percent of total public sector income. It is clear that most of the monetary benefits go towards pensions, while the non-monetary benefits going towards health care are twice as high as those going to education. Let us examine the manner in which the social security and other social expenditures were distributed among different income groups. Part (a) of Table 6 presents the distribution of monetary benefits according to 10 income distribution deciles. It is clear that most of the monetary benefits went to the top two income deciles, and in the case of pensions, which make up most of these benefits, over half went to the top income decile. Finally, 70% of non-monetary health benefits went to the lower 5 deciles (which made up only 7.1% of total public sector income), while 66% of educational non-monetary benefits went to the top four deciles. Table 7 offers some further direct and indirect evidence about the distributional impact of government expenditures. It will be noted that the non-attendance at publicly supported educational institutions is much higher among the lower than the upper quintiles of the population. This means that public expenditures on various levels of education favor the higher income groups. This is even more dramatically the case with respect to the distribution of government scholarships. Finally, the table also shows how 65% of government pensions go to the upper fifth income quintile of the population. This has led Camargo and Ferreira (2002) to conclude that the “. . . highly regressive pattern of incidence of ‘social expenditures’, which on the whole are dispropor-
350
Salaries
Debt servicing
Transfers to states and municipalities
(a) Brazil: federal government expenditures by major categories (%) 2002 11 8 11 2003 9 8 9 2004 10 8 10 2005a 9 8 10
Social security
Other current expenditures
Capital expenditures
Amortiz. and refinancing of debt
Total
13 12 13 12
7 6 7 6
15 13 12 7
35 43 40 48
100 100 100 100
(b) Federal benefits as a % total public sector income Benefits Monetary benefits Pensions Old age benefits Unemployment insurance Salary supplement Family wage supplement Family school scholarship Food scholarship Special children scholarship Non-monetary benefits Health Education Source: Ministerio da Fazenda, Secretaria do Tesouro Nacional (for part (a)) and IPEA and Rezende and Cunha (2002), p. 93 (for part (b)). a January–July 2005.
49.1 38.2 34.7 0.6 1.7 0.3 0.1 0.6 0.1 0.1 10.9 7.1 3.8
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Table 5
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
351
Table 6 1
2
3
4
5
6
7
Brazil: distribution of government monetary benefits by income groups (deciles—%) Total benefits 2 3 3 3 5 7 7 Pensions 0 1 2 3 4 7 7 Unemployment insurance 6 6 8 12 12 9 13 2 8 11 13 13 12 10 Family supporta 7 12 28 14 39 0 0 Old age supportb 35 38 19 7 1 0 0 Children supportc 1
2
3
4
5
6
7
Brazil: distribution of non-monetary benefits by income groups (deciles—%) Health 17 16 14 12 11 9 7 Education 6 6 6 6 5 5 10
8
9
10
Total
9 9 11 10 0 0
15 16 14 9 0 0
46 51 9 12 0 0
100 100 100 100 100 100
8
9
10
Total
6 12
5 17
3 27
100 100
Source: Rezende and Cunha (2002), p. 95. a Abono salarial de salario-familia. b Amparo ao idoso. c Bolsas-escola, alimentac ¸ a˜ o e crianc¸a cidad˜ao.
tionately appropriated by the middle-classes and the rich, is partially responsible for the failure to reduce poverty significantly.” (p. 97) One should also consider the distributional impact of the government debt. Over the last two decades, Brazil’s government debt increased substantially, reaching 53% of the GDP. Although the foreign debt has declined somewhat, a substantial portion of the domestic debt is also foreignowned. Such a high debt represents an expenditure burden on the government. This burden increased substantially since the devaluation of 1999, as the central bank has adopted a policy of extremely high interest rates, not only to control inflation, but also to attract foreign investors. The high interest rate has substantially increased the debt-servicing burden on the government (it will be noted in Table 5 (part (a))) that debt servicing amounted to 8% of total federal government expenditures. If one considers the fact that the owners of the government debt consist in large part of banks, investment funds (where individuals in the upper income brackets have the great representation), then the government’s debt servicing will contribute to increase the redistributing function of the fiscal system in favor of the higher income groups (taxes being collected in a regressive manner, while debt servicing favors the higher income groups).
Table 7
Not attending public primary school Not attending secondary school Not attending higher education Distribution of government scholarships Distribution of government pensions Distribution of government expenditures on scholarships
First quintile
Second quintile
Third quintile
Fourth quintile
Fifth quintile
31.2% 94.7% 100.0% 0.1%
12.9% 90.2% 99.9% 7.7%
9.3% 76.3% 99.5% 7.8%
6.9% 62.3% 94.6% 5.5%
7.4% 49.6% 67.1% 78.9%
2.4% 0.1%
6.4% 7.7%
9.7% 7.8%
16.5% 5.5%
65.1% 78.9%
Source: von Amsberg, Lanjouw, and Need (2000), Barros and Foguel (2000).
352
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
4. An econometric model of Brazil’s government expenditures and income distribution Now we move our attention to government expenditures. The latter can have distributional implications because as tax payments (which are used to finance publicly provided goods) rise with income, the income earner does not necessarily consume less of publicly provided goods and services. In addition, a large amount of the theoretical literature about income inequality and government expenditures on redistribution has been concerned about variations of the standard median voter model, according to which more income inequality is associated with higher government spending on redistribution. The subsequent empirical literature has focused on testing the sign of the correlation between inequality and redistributive government expenditures. Cross-country and cross-regions studies normally regress aggregate government spending on some measure of income distribution and control for other determinants of public expenditure. De Mello and Tiongson (2006) proposed a model to study the association between inequality and redistributive government spending by regressing government-financed redistributive transfers to individuals/households as a proportion of the GDP on a variable that measures inequality (the Gini coefficient), and a vector of control variables. De Mello and Tiongson (2006) also allow for nonlinearities and instrument inequality with different measures of capital market imperfection. In order to analyze the empirical relationship between income inequality and redistributive expenditures for the Brazilian case, we use municipality data for the year of 2000. The data were obtained from a survey undertaken by IPEAData.7 These measures are the available data from this special survey for 5508 municipalities. However, since some municipalities do not present data for all considered variables we need to discard some data. Therefore, the net number of municipalities totalized 4617 for that year. Two different types of data on government transfers are available from the survey. First, the “Quotas from the Municipal Participation Fund”, which consists of the transfer of funds from the federal government to individual municipalities, the individual amounts depending on the population size of each municipality. Second, the “Total Current Municipal Expenditures”, which includes payroll of government employees, transfers, and capital expenditures. However, it should be noted that not all transfers are redistributive in nature. As a measure of income inequality we use the Gini index. For control variables we used per capita income, the percentage of households that are home owners, and the percentage of households of persons aged 25 years or more, with more than 11 years of schooling, and also per capita income squared. Since government expenditures could have heterogeneity in the potential effects, we use a structural quantile regression (QR) approach, where income distribution for each municipality is the dependent variable of primary interest and government expenditures is the explanatory variable of interest. QR estimation is fully described in Koenker (2005). Quantile regression method offers a more complete characterization of the stochastic relationship among variables and provide a more robust, and consequently more efficient, estimates in some non-Gaussian settings. In the case analyzed in this paper, this class of estimator is suitable, since it is important to analyze the behavior of government expenditures in each quantile of the conditional income inequality distribution. QR is not only concerned with the income distribution effect on the average individual, but allows one to estimate the marginal effect of a given government expenditure on individuals at different points in the conditional achievement distribution.
7
www.ipeadata.gov.br.
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
353
A major concern for the empirical study of government spending effects is the potential endogeneity. Thus, given the possibility of biased estimates, we attempt to control for this problem, following Ma and Koenker (2006).8 Consider the following simple linear location-scale model with two endogenous variables: Yi = α0 + α1 Si + α2 xi + Si (λνi + ui )
(1)
Si = β0 + β1 xi + β2 zi + νi ,
(2)
where Y is the dependent variable of primary interest; S the endogenous variable; x denotes individual specific characteristics; z is a vector of valid “instrumental variables”; and u and v are random errors terms. We assume that the pairs (ui , vi ) are i.i.d. and independent of the pairs (xi , zi ). To briefly describe this control variate approach, proposed by Ma and Koenker (2006), consider the model with conditional quantile functions: QYi (τ1 |Si , xi , zi , νi (τ2 )) = α0 + α1 Si + α2 xi + α3 Si νi (τ2 ) + Si Fu−1 (τ1 )
(3)
QSi (τ2 |xi , zi νi (τ2 )) = β0 + β1 xi + β2 zi + Fv−1 (τ2 ),
(4)
and
where Fu−1 and Fv−1 are the distributions functions of the errors and τ 1 and τ 2 are the corresponding quantiles. Estimating ˆ 2 ) = arg minβ β(τ
n
ρτ2 (Si − β0 − β1 xi − β2 zi ),
i=1
ˆ Si (τ2 |xi , zi ) and compute we obtain νˆ i (τ2 ) = Q ˆ 1 , τ2 ) = argminα α(τ
n
ρτ1 (Yi − α0 − α1 Si − α2 xi − α3 Si νˆ i (τ2 ))
i=1
where the function ρτ (u) = u(τ − I)(u < 0)) is as in Koenker and Bassett (1978). ˆ 1 , τ2 ) is consistent and asympMa and Koenker (2006) show that the structural estimator α(τ totically normal.9 We propose estimating the following model Ii = α0 + α1
Ti Ti + α2 yi + (λνi + ui ) Yi Yi
Ti = β0 + β1 yi + β2 Zi + ui Yi
(5) (6)
8 Amemiya (1982) proposed a two-stage median regressor estimator. Abadie et al. (2002) considered quantile regression methods for estimating endogenous treatment effects focusing on the binary case. Chesher (2003, 2004) has considerably expanded the scope of quantile regression methods for structural models. Ma and Koenker (2006) have considered estimators based on the Chesher identification strategy. 9 Asymptotic distribution for the estimator is carefully described in Ma and Koenker (2006). We use the asymptotic results for control variate estimators.
354
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Table 8 Estimates of the impact of government expenditures on income inequality using the quotas from the municipal participation fund as dependent variable τ
(a)
(b)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
−0.0037 (0.0061) −0.0105** (0.0047) −0.0163* (0.0024) −0.0213* (0.0040) −0.0258* (0.0042) −0.0329* (0.0032) −0.0409* (0.0043) −0.0489* (0.0071) −0.0590* (0.0075)
−0.0056 (0.0052) −0.0112* (0.0039) −0.0163* (0.0030) −0.0190* (0.0029) −0.0221* (0.0027) −0.0265* (0.0018) −0.0323* (0.0030) −0.0378* (0.0040) −0.0417* (0.0052)
* **
Asymptotic significant at the 1% level. Asymptotic significant at the 5% level.
where T denotes government expenditures, Y is income, I a variable that measures inequality, y per capita income which is an exogenous control variable, Z the matrix of instruments, the vector (u, v) is an error term, and i identifies the municipality in the sample. For T we use two different types of government expenditures—quotas from the Municipal Participation Fund and Total Current Municipal Government Expenditures. Y is the total income of the municipality, and y denotes the per capita income. I is the Gini coefficient index. In matrix Z we have three variables, that is, the percentage of households that are home owners, the percentage of households of persons of 25 years of age and above with more than 11 years of schooling, and the income per capita squared. We set the quantiles of interest τ 1 = τ 2 = τ for all estimates, and assume that the pairs (ui , vi ) are i.i.d. and independent of the pairs (yi , Zi ), and that λ > 0. Results for estimates of the effects of government expenditures on Gini index, described in equations (5) and (6), are presented in Table 8 for selected quantiles (τ). The variable used as a measure of government expenditures is the Quotas from the Municipal Participation Fund. Column (a) of Table 8 presents the estimated effects of government expenditures on Gini coefficients using the percentage of home-owned households, and the percentage of households of persons of 25 years of age and above with more than 11 years of schooling as instruments. Column (b) presents the same estimates using the same instruments in addition to the per capita income squared. The numbers inside the parenthesis are the estimated standard deviations. Table 8 indicates that there is strong evidence that the effect of government expenditures is not constant, but varies among the various quantiles. All the estimates have a negative sign, meaning that when government expenditures increase, the Gini index decreases, and so does income inequality. In addition, it is important to note that the absolute value of the coefficients is larger for upper deciles. This shows that government is relatively inefficient in decreasing income inequality in municipalities which already have a more equitable income distribution. Therefore, for municipalities in the low deciles of the conditional income inequality distribution, where the Gini coefficient is low and the income inequality is low, an increase in the government expenditures has little impact in improving the distribution of the income. On the other hand, for the upper deciles of the conditional inequality distribution, where the income is very concentrated, an increase in the government expenditure decreases income inequality more effectively. Table 9 shows the results for estimates of the effects of government expenditures on the Gini index for selected quantiles (τ), using Total Current Municipal Government Expenditures as a
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
355
Table 9 Estimates of the impact of government expenditures on income inequality using current government expenditures as dependent variable τ
(a)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
−0.0089*
*
−0.0140* −0.0187* −0.0214* −0.0249* −0.0308* −0.0304* −0.0311* −0.0395*
(b) (0.0035) (0.0032) (0.0029) (0.0035) (0.0027) (0.0042) (0.0048) (0.0043) (0.0067)
−0.0082* −0.0126* −0.0168* −0.0191* −0.0219* −0.0247* −0.0257* −0.0257* −0.0305*
(0.0032) (0.0029) (0.0012) (0.0024) (0.0020) (0.0023) (0.0027) (0.0037) (0.0041)
Asymptotic significant at the 1% level.
measure for general government expenditures. As in Table 8, Column (a) of Table 9 presents the estimated effects of governments expenditures on Gini coefficients, using percentage of households that are home-owners, and the percentage of households of persons of 25 years of age and above with more than 11 years of schooling as instruments, and Column (b) presents the same estimates using the same instruments in addition to the per capita income squared. Again, overall, the estimations show the relative inefficiency of government expenditures in reducing income inequality for municipalities at the bottom of the distribution vis-a-vis municipalities at the top of the conditional income inequality distribution in Brazil. In addition, the absolute value of the coefficients are larger for upper quantiles, giving further evidence that the government is more efficient in decreasing income inequality for municipalities where such inequality is higher. These results are not surprising. They confirm the intuition that to in order to be more efficient in decreasing income inequality, government should transfer to municipalities where the income inequality is higher. In fact, over the 1990s, and especially in the first years of the 21st century, the Brazilian government tried to increase the type of expenditures which would favor municipalities where such inequality is high. It created a number of social programs (such as the “bolsa escola”, which subsidized families whose children were in school). However, it seems that though well intentioned, these programs made up only a very small part of government expenditures. As was shown in section 3 above, most of government expenditures occurred in programs which favored the higher income groups. Until 2005, social programs had relatively little impact.10 It is our contention, and we shown this in the beginning of this article, that government expenditures generally favor the higher income segments of Brazil’s society. Thus, increases of public expenditures result in a relatively low effect in decreasing the income inequality. 5. Conclusion In this article we have shown that Brazil does not conform to the characteristics of most countries in terms of the distribution of income, the tax burden and expenditure patterns of the government. An examination of descriptive statistics on taxes and government expenditures give
10
See Amann and Baer (2006).
356
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
strong evidence that both the tax burden and government expenditures favor the higher income classes, which means that the country’s fiscal system has a relatively low redistributional impact. The results of estimates of the effects of government expenditures on income inequality, presents evidence that in order to reduce income inequality, the government would have to emphasize in its expenditure patterns those programs which benefit more municipalities with high income inequality. In addition, the results showed the relative inefficiency of government expenditures in reducing the income inequality for municipalities at the bottom of the income distribution, where income is more equitable, vis-a-vis municipalities at the top of the conditional income inequality distribution, where income is less equitable, in Brazil. Thus, the seeming paradox of a country with a high tax burden and a continually concentrated distribution of income can in great part be explained by the structure of the country’s fiscal system. Whereas in many advanced industrial countries the fiscal system has been used to make the distribution of income more equitable, the contrary was the case with Brazil. Not only is Brazil’s tax system regressive in nature, but also the distribution of the state’s expenditures favors the higher income classes. The implication of these findings is that one important means for improving Brazil’s income distribution is to drastically change not only the tax structure, but also the spending behavior of the state. Acknowledgment We would like to thank Odilon Camara, Enlinson Carvalho, Tiago Cavalcanti, Gabriel Montes Rojas and Peri Silva for many helpful comments. Of course, only we are responsible for the final product. Appendix A
List of countries used in Fig. 1 Argentina Australia Austria Belarus Belgium Bolivia Brazil Bulgaria Canada Chile Croatia Czech Republic Denmark Estonia Finland France Germany Greece Hungary India Ireland Israel
ARG AUS AUT BER BEL BOL BRA BUL CAN CHL CRO CZE DEN EST FIN FRA GER GRE HUN IND IRE ISR
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
357
Appendix A (Continued ) Italy Kazakhstan Kyrgyz Republic Latvia Lithuania Luxembourg Malaysia Mexico Moldova Mongolia Netherlands New Zealand Norway Peru Poland Portugal Romania Russia Slovak Republic Slovenia South Africa Spain Sweden Switzerland Thailand Ukraine United Kingdom United States
ITA KAZ KYR LAT LIT LUX MAL MEX MOL MON NZH NZE NOR PER POL POR ROM RUS SLK SLO SAF SPA SWE SWI THA UKR UK USA
References Abadie, A., Angrist, J., & Imbens, G. (2002). Instrumental variables estimation of the effects subsidized training on the quantiles of trainee earnings. Econometrica, 70, 91–117. Amann, E., & Baer, W. (2006). Economic orthodoxy versus social development? The dilemmas facing Brazil’s Labour Government. Oxford Development Studies, 34(2), 219–241. Amemiya, T. (1982). Two stage least absolute deviations estimators. Econometrica, 50, 689–711. Barros, R. P., & Foguel, M. (2000). Focalizac¸a˜ o dos Gastos P´ublicos Sociais e Erradicac¸a˜ o da Pobreza no Brasil. In Ricardo Henriques (Ed.), Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA. Camargo, J. M., & Ferreira, F. H. G. (2002, Winter). Missing the Target: Assessing Social Expenditures in Brazil. The Brown Journal of World Affairs, VIII(Issue 2). Carneiro, F. G., De Mello, L., & Tiongson, E. (2002). Income inequality and redistributive government spending: Theory and panel data evidence. In A. Levy & Jo˜ao Ricardo Faria (Eds.), Economic growth, inequality, and migration, Cheltenham, UK. Chesher, A. (2003). Identification in nonseparable models. Econometrica, 71, 1405–1441. Chesher, A. (2004). Nonparametric identification under discrete variation. CEMMAP, Working Paper. Clements, B. (1997). Income distribution and social expenditure in Brazil (IMF Working Paper WP/97/120). De Mello, L., & Tiongson, E. (2006). Income inequality and redistributive government spending. Public Finance Review, 34, 282–305. Hoffmann, Rodolfo. (2003). Inequality in Brazil: The contribution of pensions. Revista Brasileira de Economia (Dezembro), 57(4), 755–773. Koenker. (2005). Quantile regression. Cambridge University Press. Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica, 46, 33–49.
358
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Ma, L., & Koenker, R. (2006). Quantile regression methods for recursive structural equation models. Journal of Econometrics, 134(2), 471–506. Musgrave, R. A. (1969). Fiscal systems. New Haven: Yale University Press. Perkins, Dwight H., Radelet, Steven, Snodgrass, Donald R., Gillis, Malcolm, & Roemer, Michael. (2001). Economics of development (5th ed.). New York: W.W. Norton & Company. Ramos, C. A. (2000). Impacto Distributivo do Gasto Publico: Uma Analise a partir da PCV/1998. Texto Para Discuss˜ao, No. 732, IPEA, Rio de Janeiro. Rezende, Fernando, & Cunha, Armando. (2002). Contribuintes e Cidad˜aos: Compreendendo o Or¸camento Federal. Rio de Janeiro: Fundac¸a˜ o Getulio Vargas. Siqueira, R. B., Nogueira, J. R., & Levy, H. (2002, July). The effects of federal taxes and benefits on household income in Brazil. Paper prepared for evaluating the impact of policy on poverty and distribution: Models and experiences of Brazil and the World Bank seminar, organized by the World Bank and FIPE/USP, S˜ao Paulo. von Amsberg, J. et al. (2000). A focaliza¸ca˜ o do Gasto Social Sobre a Pobreza no Brasil. In R. Henriques (org.), Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA. von Amsberg, J., Lanjouw, P., & Need, K. (2000). A Focalizac¸a˜ o do Gasto Social sobre a Pobreza no Brasil. In Ricardo Henriques (Ed.), Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA.
Tax burden, government expenditures and income distribution in Brazil Werner Baer ∗ , Antonio Fialho Galv˜ao Jr. University of Illinois at Urbana-Champaign, United States Received 13 December 2006; received in revised form 13 December 2006; accepted 13 December 2006 Available online 25 March 2007
Abstract This article attempts to explain the seeming paradox of a country with a high tax burden and a continually concentrated distribution of income. By means of a structural quantile regression model we analyze the distributional impact of government expenditures on the Gini index, and it is shown that Brazil’s redistribution expenditures has a relatively smaller impact for low quantiles of the conditional distribution of income inequality. It is also noted that both the country’s tax and expenditure structure are, in part, responsible for the country’s continuous concentration of income. © 2007 Board of Trustees of the University of Illinois. All rights reserved. Keywords: Income distribution; Taxes; Government expenditures; Brazil
1. Introduction Toward the end of the 19th century the German political theorist Adolph Wagner devised his law of expanding state activity, also known as Wagner’s Law. This law stated that the size of the public sector in the economy grows as per capita income rises. Although this “law” was somewhat controversial, the data show that there is a tendency for government expenditures as a share of GDP to be larger for rich than for poor countries. For instance the World Bank’s World Development Report of 1994 showed that central government expenditures as a share of GDP was 17% for low-income countries and 32% for high-income countries. Analysts have also noted that wealthy countries “. . . have larger social welfare programs that cause subsidies and other transfers to be a much higher share of expenditures than in most lower-income countries.”1 ∗ 1
Corresponding author. Tel.: +1 217 333 8388; fax: +1 217 333 1398. E-mail address: [email protected] (W. Baer). Perkins et al. (2001), p. 423.
1062-9769/$ – see front matter © 2007 Board of Trustees of the University of Illinois. All rights reserved.
doi:10.1016/j.qref.2006.12.012
346
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Similarly, in his review of world fiscal systems Musgrave noted that “. . . the rise in the overall expenditure to GNP ratio has been due primarily to the growth of social services. This reflects in particular a growing re-distributional concern, rather than a rising demand for social goods as such. It appears that the underlying forces of social and political change greatly outweighed any tendency . . . for the transfer share to fall with rising per capita income.”2 Furthermore, he found that the “. . . rise in the relative importance of social services, which make heavy use of transfer payments, suggests a considerable increase in the importance of transfer payments relative to public purchases.”3 One would thus expect to find the tax burden of most low income countries to be lighter than the tax burden of wealthy countries, and at the same time that the distribution of income to be more equitable in higher than in lower income countries. It is thus striking to find that in Brazil, one of the major emerging countries, the tax burden is similar to that of many advanced industrial countries, its income distribution is among the most concentrated in the world. It is the purpose of this article to throw some light on this seeming paradox. Some previous studies focused on some of the elements which might lead to an explanation of the determinants of inequality linked to the fiscal system, especially its social expenditures. For instance, De Mello and Tiongson (2006) cross-section study found that “. . . re-distributive spending may be inefficient as an instrument to reduce poverty and to improve income distribution because the benefits of public spending may be captured by the non-poor.” (p. 303). Carneiro, De Mello, and Tiongson (2002) made a study of 37 industrial and developing countries in 1972–1987 and found that “. . . the countries where redistributive public spending is more needed were found to be the ones that are less likely to redistribute income through public policies.” (p. 103). Another recent study by Siqueira, Nogueira and Levy (2002) attempted to identify the types of households which benefit from federal government expenditures in Brazil. They came to the conclusion that Brazil’s tax and benefits system is deceptively small given the substantial amount of resources involved. Clements (1997) found that government social expenditures in Brazil have exacerbated income inequalities. Another set of studies (Barros & Foguel, 2000; Hoffmann, 2003; Ramos, 2000; von Amsberg et al., 2000) showed that one of the reasons that the Brazilian government did not succeed in eliminating poverty was due to the inadequate targeting of public expenditures. Within this framework we shall expand the study of the relation of the fiscal structure to the distribution of income. We examine some evidence with regard to the distribution of the tax burden4 of the tax system in Brazil. In addition, we estimate the impact of government expenditures on the country’s income distribution5 using a structural quantile regression model. An examination of descriptive statistics on taxes and government expenditures give strong evidence that both the tax burden and government expenditures favor the higher income classes, which means that the country’s fiscal system has a relatively low redistributional impact. In addition, the results of estimates of the effects of government expenditures on income inequality, using municipality data, present evidence that in order to reduce income inequality, the government would have to emphasize in its expenditure patterns those programs which benefit more municipalities with high-income inequality.
2
Musgrave (1969), pp. 93–96. Musgrave (1969) p. 96. 4 By “tax burden” is meant the ratio of total taxes paid to the GDP. 5 We are aware that the country’s tax structure may also contribute to the concentration of income, which we discuss below. 3
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
347
Fig. 1. Tax burden vs. Gini coefficient.
The article is organized as follows. Section 2 presents an overview of tax burden and income inequality around the world. Section 3 presents some evidence for Brazil’s tax burden and government expenditure. Section 4 proposes an econometric model of Brazil’s government expenditures and income distribution. Finally, Section 5 contains a general conclusion. 2. Tax burden and income distribution: world trends In this section we shall take a brief look at the income distribution and tax burdens prevailing in a large sample of countries around the globe. It will be noted that Brazil does not fall into the generally observed pattern, that is, the higher the tax burden, the more equitable is the distribution of income. We first plot the Gini index and the tax burden for 50 countries.6 Data for Gini coefficient is of a cross-section nature, which are available in the Human Development Report 2000. The International Monetary Fund provides data for the tax burden. We use tax burden data for 1998. The tax burden is simply total tax revenues as a share of GDP. Fig. 1 shows the scatter diagram of Gini coefficients and tax burdens. It will be noted that Brazil’s income distribution was one of the most concentrated in the world with a relatively high tax burden. This becomes even clearer in Table 1, which describes the gap between the rich and poor for selected countries. Thus the evidence presented shows Brazil as being an important case to be studied, given that it appears to have a very concentrated income and relatively high tax burden.
6
A list of the countries examined can be found in Appendix A.
348
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Table 1 Gap between rich and poor for selected countries (% share of income)
Sierra Leone Brazil Mexico United States Germany Sweden
Lowest 20% of income groups
Highest 20% of income groups
1.1 2.2 3.5 4.2 8.2 9.6
63.4 64.1 57.4 46.9 38.5 34.5
Source: http://www.infoplease.com/ipa/A0908770.htm1. Table 2 Brazil’s tax burden as % of GDP 1988
1994
2000
2002
2003
Federal State Municipal Social security
16.1 5.7 0.6
13.4 7.8 1.0 5.63
14.8 8.35 0.84 7.12
17.1 8.84 1.11 7.45
16.4 8.61 1.20 7.41
Total
22.4
27.9
31.6
34.9
34.0
2004
2006
35.9
38.0
Source: IPEA, Secretaria da Receita Federal.
3. Some evidence for Brazil’s government expenditure In this section we present some evidence of how Brazil’s fiscal system contributes to the continued concentration of income. Let us begin by examining the country’s tax burden. As is clear from Table 2, Brazil’s tax burden has grown substantially. It amounted to 22.4% in 1988 and reached 38% in 2006. This burden was disproportionately born by the country’s lower income groups, which becomes clear when looking at the origins of these taxes. It will be noted in Tables 3 and 4 that the tax system was quite regressive. In 2002 direct taxes made up only 15.7% of the federal government’s revenues, whereas indirect taxes accounted for 42.1%. Table 4 (part (a)) shows that whereas 86% of direct taxes were paid by the top three deciles, these only paid 58% of indirect taxes. In part (b) of that table, it will be noted that the indirect Table 3 Federal taxes as share of total government receipts in 2002 Total federal taxes Direct taxes Personal income tax Social security contributions (employees) Indirect taxes Tax on industrial products (IPI) Contributions to the financing of social securitya Social integration program (PIS) and program for public servants (PASEP) Education salary Social security contributions of firms Source: Lei Orc¸amentaria (2002) and Rezende and Cunha (2002). a Confins.
57.8% 15.7% 7.7% 8.0% 42.1% 7.4% 16.7% 4.1% 1.2% 12.7%
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
349
Table 4 1
2
3
4
5
7
8
9
10
Total
3 0 6 7
5 0 10 10
7 0 14 13
13 3 22 16
66 97 38 29
100 100 100 100
5
6
9
11
16
40
100
(b) Taxes as a % of gross income by income groups (deciles) Direct taxes 0 1 2 3 4 Indirect taxes 22 17 16 14 14
4 13
5 13
5 12
6 11
10 8
(a) Distribution of tax receipts by income groups (deciles) (%) Direct taxes 0 1 1 2 2 IRPF 0 0 0 0 0 Soca 0 1 2 3 4 Indirect taxes 3 4 5 6 7 Total
2
3
4
4
6
Source: Rezende and Cunha (2002), p. 104 (for part (a)) and Rezende and Cunha (2002), p. 106 (for part (b)). a Social security contributions of employees.
tax burden falls more heavily on the lower than on the upper income groups. Thus, though the general tax burden of Brazil was high, the proportion of taxes born by the lower income groups was even higher. Taxes are only part of the story, as the beneficiaries of government expenditures also need to be examined. How do these expenditures affect the distribution of income? It will be noted in Table 5 (part (a)) that in 2005 over half of government expenditures were for amortization of the debt and debt servicing, which are not of a redistributive nature. Social security expenditures and “Other Current Expenditures” category, which includes health, education, etc. and which might be thought of as having a redistributive function, constituted about 18–20% of government expenditures in the years 2002–2005. However, even this category, when examined more carefully, will show that its redistributive function is extremely limited in the case of Brazil. For instance, another way of examining government expenditures is presented in Table 5 (part (b)). It shows in a more disaggregate way for the year 2002 the distribution of federal benefits as a percent of total public sector income. It is clear that most of the monetary benefits go towards pensions, while the non-monetary benefits going towards health care are twice as high as those going to education. Let us examine the manner in which the social security and other social expenditures were distributed among different income groups. Part (a) of Table 6 presents the distribution of monetary benefits according to 10 income distribution deciles. It is clear that most of the monetary benefits went to the top two income deciles, and in the case of pensions, which make up most of these benefits, over half went to the top income decile. Finally, 70% of non-monetary health benefits went to the lower 5 deciles (which made up only 7.1% of total public sector income), while 66% of educational non-monetary benefits went to the top four deciles. Table 7 offers some further direct and indirect evidence about the distributional impact of government expenditures. It will be noted that the non-attendance at publicly supported educational institutions is much higher among the lower than the upper quintiles of the population. This means that public expenditures on various levels of education favor the higher income groups. This is even more dramatically the case with respect to the distribution of government scholarships. Finally, the table also shows how 65% of government pensions go to the upper fifth income quintile of the population. This has led Camargo and Ferreira (2002) to conclude that the “. . . highly regressive pattern of incidence of ‘social expenditures’, which on the whole are dispropor-
350
Salaries
Debt servicing
Transfers to states and municipalities
(a) Brazil: federal government expenditures by major categories (%) 2002 11 8 11 2003 9 8 9 2004 10 8 10 2005a 9 8 10
Social security
Other current expenditures
Capital expenditures
Amortiz. and refinancing of debt
Total
13 12 13 12
7 6 7 6
15 13 12 7
35 43 40 48
100 100 100 100
(b) Federal benefits as a % total public sector income Benefits Monetary benefits Pensions Old age benefits Unemployment insurance Salary supplement Family wage supplement Family school scholarship Food scholarship Special children scholarship Non-monetary benefits Health Education Source: Ministerio da Fazenda, Secretaria do Tesouro Nacional (for part (a)) and IPEA and Rezende and Cunha (2002), p. 93 (for part (b)). a January–July 2005.
49.1 38.2 34.7 0.6 1.7 0.3 0.1 0.6 0.1 0.1 10.9 7.1 3.8
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Table 5
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
351
Table 6 1
2
3
4
5
6
7
Brazil: distribution of government monetary benefits by income groups (deciles—%) Total benefits 2 3 3 3 5 7 7 Pensions 0 1 2 3 4 7 7 Unemployment insurance 6 6 8 12 12 9 13 2 8 11 13 13 12 10 Family supporta 7 12 28 14 39 0 0 Old age supportb 35 38 19 7 1 0 0 Children supportc 1
2
3
4
5
6
7
Brazil: distribution of non-monetary benefits by income groups (deciles—%) Health 17 16 14 12 11 9 7 Education 6 6 6 6 5 5 10
8
9
10
Total
9 9 11 10 0 0
15 16 14 9 0 0
46 51 9 12 0 0
100 100 100 100 100 100
8
9
10
Total
6 12
5 17
3 27
100 100
Source: Rezende and Cunha (2002), p. 95. a Abono salarial de salario-familia. b Amparo ao idoso. c Bolsas-escola, alimentac ¸ a˜ o e crianc¸a cidad˜ao.
tionately appropriated by the middle-classes and the rich, is partially responsible for the failure to reduce poverty significantly.” (p. 97) One should also consider the distributional impact of the government debt. Over the last two decades, Brazil’s government debt increased substantially, reaching 53% of the GDP. Although the foreign debt has declined somewhat, a substantial portion of the domestic debt is also foreignowned. Such a high debt represents an expenditure burden on the government. This burden increased substantially since the devaluation of 1999, as the central bank has adopted a policy of extremely high interest rates, not only to control inflation, but also to attract foreign investors. The high interest rate has substantially increased the debt-servicing burden on the government (it will be noted in Table 5 (part (a))) that debt servicing amounted to 8% of total federal government expenditures. If one considers the fact that the owners of the government debt consist in large part of banks, investment funds (where individuals in the upper income brackets have the great representation), then the government’s debt servicing will contribute to increase the redistributing function of the fiscal system in favor of the higher income groups (taxes being collected in a regressive manner, while debt servicing favors the higher income groups).
Table 7
Not attending public primary school Not attending secondary school Not attending higher education Distribution of government scholarships Distribution of government pensions Distribution of government expenditures on scholarships
First quintile
Second quintile
Third quintile
Fourth quintile
Fifth quintile
31.2% 94.7% 100.0% 0.1%
12.9% 90.2% 99.9% 7.7%
9.3% 76.3% 99.5% 7.8%
6.9% 62.3% 94.6% 5.5%
7.4% 49.6% 67.1% 78.9%
2.4% 0.1%
6.4% 7.7%
9.7% 7.8%
16.5% 5.5%
65.1% 78.9%
Source: von Amsberg, Lanjouw, and Need (2000), Barros and Foguel (2000).
352
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
4. An econometric model of Brazil’s government expenditures and income distribution Now we move our attention to government expenditures. The latter can have distributional implications because as tax payments (which are used to finance publicly provided goods) rise with income, the income earner does not necessarily consume less of publicly provided goods and services. In addition, a large amount of the theoretical literature about income inequality and government expenditures on redistribution has been concerned about variations of the standard median voter model, according to which more income inequality is associated with higher government spending on redistribution. The subsequent empirical literature has focused on testing the sign of the correlation between inequality and redistributive government expenditures. Cross-country and cross-regions studies normally regress aggregate government spending on some measure of income distribution and control for other determinants of public expenditure. De Mello and Tiongson (2006) proposed a model to study the association between inequality and redistributive government spending by regressing government-financed redistributive transfers to individuals/households as a proportion of the GDP on a variable that measures inequality (the Gini coefficient), and a vector of control variables. De Mello and Tiongson (2006) also allow for nonlinearities and instrument inequality with different measures of capital market imperfection. In order to analyze the empirical relationship between income inequality and redistributive expenditures for the Brazilian case, we use municipality data for the year of 2000. The data were obtained from a survey undertaken by IPEAData.7 These measures are the available data from this special survey for 5508 municipalities. However, since some municipalities do not present data for all considered variables we need to discard some data. Therefore, the net number of municipalities totalized 4617 for that year. Two different types of data on government transfers are available from the survey. First, the “Quotas from the Municipal Participation Fund”, which consists of the transfer of funds from the federal government to individual municipalities, the individual amounts depending on the population size of each municipality. Second, the “Total Current Municipal Expenditures”, which includes payroll of government employees, transfers, and capital expenditures. However, it should be noted that not all transfers are redistributive in nature. As a measure of income inequality we use the Gini index. For control variables we used per capita income, the percentage of households that are home owners, and the percentage of households of persons aged 25 years or more, with more than 11 years of schooling, and also per capita income squared. Since government expenditures could have heterogeneity in the potential effects, we use a structural quantile regression (QR) approach, where income distribution for each municipality is the dependent variable of primary interest and government expenditures is the explanatory variable of interest. QR estimation is fully described in Koenker (2005). Quantile regression method offers a more complete characterization of the stochastic relationship among variables and provide a more robust, and consequently more efficient, estimates in some non-Gaussian settings. In the case analyzed in this paper, this class of estimator is suitable, since it is important to analyze the behavior of government expenditures in each quantile of the conditional income inequality distribution. QR is not only concerned with the income distribution effect on the average individual, but allows one to estimate the marginal effect of a given government expenditure on individuals at different points in the conditional achievement distribution.
7
www.ipeadata.gov.br.
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
353
A major concern for the empirical study of government spending effects is the potential endogeneity. Thus, given the possibility of biased estimates, we attempt to control for this problem, following Ma and Koenker (2006).8 Consider the following simple linear location-scale model with two endogenous variables: Yi = α0 + α1 Si + α2 xi + Si (λνi + ui )
(1)
Si = β0 + β1 xi + β2 zi + νi ,
(2)
where Y is the dependent variable of primary interest; S the endogenous variable; x denotes individual specific characteristics; z is a vector of valid “instrumental variables”; and u and v are random errors terms. We assume that the pairs (ui , vi ) are i.i.d. and independent of the pairs (xi , zi ). To briefly describe this control variate approach, proposed by Ma and Koenker (2006), consider the model with conditional quantile functions: QYi (τ1 |Si , xi , zi , νi (τ2 )) = α0 + α1 Si + α2 xi + α3 Si νi (τ2 ) + Si Fu−1 (τ1 )
(3)
QSi (τ2 |xi , zi νi (τ2 )) = β0 + β1 xi + β2 zi + Fv−1 (τ2 ),
(4)
and
where Fu−1 and Fv−1 are the distributions functions of the errors and τ 1 and τ 2 are the corresponding quantiles. Estimating ˆ 2 ) = arg minβ β(τ
n
ρτ2 (Si − β0 − β1 xi − β2 zi ),
i=1
ˆ Si (τ2 |xi , zi ) and compute we obtain νˆ i (τ2 ) = Q ˆ 1 , τ2 ) = argminα α(τ
n
ρτ1 (Yi − α0 − α1 Si − α2 xi − α3 Si νˆ i (τ2 ))
i=1
where the function ρτ (u) = u(τ − I)(u < 0)) is as in Koenker and Bassett (1978). ˆ 1 , τ2 ) is consistent and asympMa and Koenker (2006) show that the structural estimator α(τ totically normal.9 We propose estimating the following model Ii = α0 + α1
Ti Ti + α2 yi + (λνi + ui ) Yi Yi
Ti = β0 + β1 yi + β2 Zi + ui Yi
(5) (6)
8 Amemiya (1982) proposed a two-stage median regressor estimator. Abadie et al. (2002) considered quantile regression methods for estimating endogenous treatment effects focusing on the binary case. Chesher (2003, 2004) has considerably expanded the scope of quantile regression methods for structural models. Ma and Koenker (2006) have considered estimators based on the Chesher identification strategy. 9 Asymptotic distribution for the estimator is carefully described in Ma and Koenker (2006). We use the asymptotic results for control variate estimators.
354
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Table 8 Estimates of the impact of government expenditures on income inequality using the quotas from the municipal participation fund as dependent variable τ
(a)
(b)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
−0.0037 (0.0061) −0.0105** (0.0047) −0.0163* (0.0024) −0.0213* (0.0040) −0.0258* (0.0042) −0.0329* (0.0032) −0.0409* (0.0043) −0.0489* (0.0071) −0.0590* (0.0075)
−0.0056 (0.0052) −0.0112* (0.0039) −0.0163* (0.0030) −0.0190* (0.0029) −0.0221* (0.0027) −0.0265* (0.0018) −0.0323* (0.0030) −0.0378* (0.0040) −0.0417* (0.0052)
* **
Asymptotic significant at the 1% level. Asymptotic significant at the 5% level.
where T denotes government expenditures, Y is income, I a variable that measures inequality, y per capita income which is an exogenous control variable, Z the matrix of instruments, the vector (u, v) is an error term, and i identifies the municipality in the sample. For T we use two different types of government expenditures—quotas from the Municipal Participation Fund and Total Current Municipal Government Expenditures. Y is the total income of the municipality, and y denotes the per capita income. I is the Gini coefficient index. In matrix Z we have three variables, that is, the percentage of households that are home owners, the percentage of households of persons of 25 years of age and above with more than 11 years of schooling, and the income per capita squared. We set the quantiles of interest τ 1 = τ 2 = τ for all estimates, and assume that the pairs (ui , vi ) are i.i.d. and independent of the pairs (yi , Zi ), and that λ > 0. Results for estimates of the effects of government expenditures on Gini index, described in equations (5) and (6), are presented in Table 8 for selected quantiles (τ). The variable used as a measure of government expenditures is the Quotas from the Municipal Participation Fund. Column (a) of Table 8 presents the estimated effects of government expenditures on Gini coefficients using the percentage of home-owned households, and the percentage of households of persons of 25 years of age and above with more than 11 years of schooling as instruments. Column (b) presents the same estimates using the same instruments in addition to the per capita income squared. The numbers inside the parenthesis are the estimated standard deviations. Table 8 indicates that there is strong evidence that the effect of government expenditures is not constant, but varies among the various quantiles. All the estimates have a negative sign, meaning that when government expenditures increase, the Gini index decreases, and so does income inequality. In addition, it is important to note that the absolute value of the coefficients is larger for upper deciles. This shows that government is relatively inefficient in decreasing income inequality in municipalities which already have a more equitable income distribution. Therefore, for municipalities in the low deciles of the conditional income inequality distribution, where the Gini coefficient is low and the income inequality is low, an increase in the government expenditures has little impact in improving the distribution of the income. On the other hand, for the upper deciles of the conditional inequality distribution, where the income is very concentrated, an increase in the government expenditure decreases income inequality more effectively. Table 9 shows the results for estimates of the effects of government expenditures on the Gini index for selected quantiles (τ), using Total Current Municipal Government Expenditures as a
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
355
Table 9 Estimates of the impact of government expenditures on income inequality using current government expenditures as dependent variable τ
(a)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
−0.0089*
*
−0.0140* −0.0187* −0.0214* −0.0249* −0.0308* −0.0304* −0.0311* −0.0395*
(b) (0.0035) (0.0032) (0.0029) (0.0035) (0.0027) (0.0042) (0.0048) (0.0043) (0.0067)
−0.0082* −0.0126* −0.0168* −0.0191* −0.0219* −0.0247* −0.0257* −0.0257* −0.0305*
(0.0032) (0.0029) (0.0012) (0.0024) (0.0020) (0.0023) (0.0027) (0.0037) (0.0041)
Asymptotic significant at the 1% level.
measure for general government expenditures. As in Table 8, Column (a) of Table 9 presents the estimated effects of governments expenditures on Gini coefficients, using percentage of households that are home-owners, and the percentage of households of persons of 25 years of age and above with more than 11 years of schooling as instruments, and Column (b) presents the same estimates using the same instruments in addition to the per capita income squared. Again, overall, the estimations show the relative inefficiency of government expenditures in reducing income inequality for municipalities at the bottom of the distribution vis-a-vis municipalities at the top of the conditional income inequality distribution in Brazil. In addition, the absolute value of the coefficients are larger for upper quantiles, giving further evidence that the government is more efficient in decreasing income inequality for municipalities where such inequality is higher. These results are not surprising. They confirm the intuition that to in order to be more efficient in decreasing income inequality, government should transfer to municipalities where the income inequality is higher. In fact, over the 1990s, and especially in the first years of the 21st century, the Brazilian government tried to increase the type of expenditures which would favor municipalities where such inequality is high. It created a number of social programs (such as the “bolsa escola”, which subsidized families whose children were in school). However, it seems that though well intentioned, these programs made up only a very small part of government expenditures. As was shown in section 3 above, most of government expenditures occurred in programs which favored the higher income groups. Until 2005, social programs had relatively little impact.10 It is our contention, and we shown this in the beginning of this article, that government expenditures generally favor the higher income segments of Brazil’s society. Thus, increases of public expenditures result in a relatively low effect in decreasing the income inequality. 5. Conclusion In this article we have shown that Brazil does not conform to the characteristics of most countries in terms of the distribution of income, the tax burden and expenditure patterns of the government. An examination of descriptive statistics on taxes and government expenditures give
10
See Amann and Baer (2006).
356
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
strong evidence that both the tax burden and government expenditures favor the higher income classes, which means that the country’s fiscal system has a relatively low redistributional impact. The results of estimates of the effects of government expenditures on income inequality, presents evidence that in order to reduce income inequality, the government would have to emphasize in its expenditure patterns those programs which benefit more municipalities with high income inequality. In addition, the results showed the relative inefficiency of government expenditures in reducing the income inequality for municipalities at the bottom of the income distribution, where income is more equitable, vis-a-vis municipalities at the top of the conditional income inequality distribution, where income is less equitable, in Brazil. Thus, the seeming paradox of a country with a high tax burden and a continually concentrated distribution of income can in great part be explained by the structure of the country’s fiscal system. Whereas in many advanced industrial countries the fiscal system has been used to make the distribution of income more equitable, the contrary was the case with Brazil. Not only is Brazil’s tax system regressive in nature, but also the distribution of the state’s expenditures favors the higher income classes. The implication of these findings is that one important means for improving Brazil’s income distribution is to drastically change not only the tax structure, but also the spending behavior of the state. Acknowledgment We would like to thank Odilon Camara, Enlinson Carvalho, Tiago Cavalcanti, Gabriel Montes Rojas and Peri Silva for many helpful comments. Of course, only we are responsible for the final product. Appendix A
List of countries used in Fig. 1 Argentina Australia Austria Belarus Belgium Bolivia Brazil Bulgaria Canada Chile Croatia Czech Republic Denmark Estonia Finland France Germany Greece Hungary India Ireland Israel
ARG AUS AUT BER BEL BOL BRA BUL CAN CHL CRO CZE DEN EST FIN FRA GER GRE HUN IND IRE ISR
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
357
Appendix A (Continued ) Italy Kazakhstan Kyrgyz Republic Latvia Lithuania Luxembourg Malaysia Mexico Moldova Mongolia Netherlands New Zealand Norway Peru Poland Portugal Romania Russia Slovak Republic Slovenia South Africa Spain Sweden Switzerland Thailand Ukraine United Kingdom United States
ITA KAZ KYR LAT LIT LUX MAL MEX MOL MON NZH NZE NOR PER POL POR ROM RUS SLK SLO SAF SPA SWE SWI THA UKR UK USA
References Abadie, A., Angrist, J., & Imbens, G. (2002). Instrumental variables estimation of the effects subsidized training on the quantiles of trainee earnings. Econometrica, 70, 91–117. Amann, E., & Baer, W. (2006). Economic orthodoxy versus social development? The dilemmas facing Brazil’s Labour Government. Oxford Development Studies, 34(2), 219–241. Amemiya, T. (1982). Two stage least absolute deviations estimators. Econometrica, 50, 689–711. Barros, R. P., & Foguel, M. (2000). Focalizac¸a˜ o dos Gastos P´ublicos Sociais e Erradicac¸a˜ o da Pobreza no Brasil. In Ricardo Henriques (Ed.), Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA. Camargo, J. M., & Ferreira, F. H. G. (2002, Winter). Missing the Target: Assessing Social Expenditures in Brazil. The Brown Journal of World Affairs, VIII(Issue 2). Carneiro, F. G., De Mello, L., & Tiongson, E. (2002). Income inequality and redistributive government spending: Theory and panel data evidence. In A. Levy & Jo˜ao Ricardo Faria (Eds.), Economic growth, inequality, and migration, Cheltenham, UK. Chesher, A. (2003). Identification in nonseparable models. Econometrica, 71, 1405–1441. Chesher, A. (2004). Nonparametric identification under discrete variation. CEMMAP, Working Paper. Clements, B. (1997). Income distribution and social expenditure in Brazil (IMF Working Paper WP/97/120). De Mello, L., & Tiongson, E. (2006). Income inequality and redistributive government spending. Public Finance Review, 34, 282–305. Hoffmann, Rodolfo. (2003). Inequality in Brazil: The contribution of pensions. Revista Brasileira de Economia (Dezembro), 57(4), 755–773. Koenker. (2005). Quantile regression. Cambridge University Press. Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica, 46, 33–49.
358
W. Baer, A.F. Galv˜ao Jr. / The Quarterly Review of Economics and Finance 48 (2008) 345–358
Ma, L., & Koenker, R. (2006). Quantile regression methods for recursive structural equation models. Journal of Econometrics, 134(2), 471–506. Musgrave, R. A. (1969). Fiscal systems. New Haven: Yale University Press. Perkins, Dwight H., Radelet, Steven, Snodgrass, Donald R., Gillis, Malcolm, & Roemer, Michael. (2001). Economics of development (5th ed.). New York: W.W. Norton & Company. Ramos, C. A. (2000). Impacto Distributivo do Gasto Publico: Uma Analise a partir da PCV/1998. Texto Para Discuss˜ao, No. 732, IPEA, Rio de Janeiro. Rezende, Fernando, & Cunha, Armando. (2002). Contribuintes e Cidad˜aos: Compreendendo o Or¸camento Federal. Rio de Janeiro: Fundac¸a˜ o Getulio Vargas. Siqueira, R. B., Nogueira, J. R., & Levy, H. (2002, July). The effects of federal taxes and benefits on household income in Brazil. Paper prepared for evaluating the impact of policy on poverty and distribution: Models and experiences of Brazil and the World Bank seminar, organized by the World Bank and FIPE/USP, S˜ao Paulo. von Amsberg, J. et al. (2000). A focaliza¸ca˜ o do Gasto Social Sobre a Pobreza no Brasil. In R. Henriques (org.), Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA. von Amsberg, J., Lanjouw, P., & Need, K. (2000). A Focalizac¸a˜ o do Gasto Social sobre a Pobreza no Brasil. In Ricardo Henriques (Ed.), Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA.