## 1. The Cost of Living Adjustment

## Source details

**Canonical URL:** [1. The Cost of Living Adjustment](https://www.imf.org/-/media/files/publications/wp/2017/wp17225.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17225.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17225.pdf.json)

---

### Overview: recent social and inequality trends in Brazil
- Poverty (World Bank international poverty line): from 25 percent of the population in 2004 to 8.5 percent in 2014.
- Extreme poverty: from 12 percent in 2004 to 4 percent in 2014.
- World Bank Gini coefficient: from 0.60 in 1990 to 0.51 in 2014.
- Distribution of labor income (PNAD 2014): top decile = 40 percent of labor income of all Brazilian families; top 1 percent ≈ 12 percent; top 0.1 percent ≈ 2.5 percent; "Half percent of all labor income is concentrated in the top 0.01 percent."
- Recession starting in 2014 affected progress: unemployment reached 13 percent in 2017; by early-2017, more than one in four young adults in Brazil were unemployed.
- Fiscal context: post-recession government faces a long period of fiscal consolidation; observing the cap on federal government non-interest expenditures will require restraint across all categories of spending in the medium term.
- Main drivers of the past decline in inequality (paper’s regression findings): income growth, higher schooling levels, labor formalization; Bolsa Família contributed to income convergence; civil servants’ wage growth slowed gains in equality.
- Policy implications highlighted: ensure fiscal sustainability while improving spending efficiency and avoiding adverse distributional effects; better targeting of social benefits, rationalizing the tax system, and moderating civil servants’ wages are key as labor formalization and income growth slow.

### Data and methodological innovation: spatial price adjustment (Box 1)
- Rationale: living standards (price levels) differ across Brazilian regions; adjusting incomes for spatial price differences is necessary to distinguish nominal and real differences in incomes and to facilitate comparisons across states.
- Proxy used: declared household rent prices from PNAD and dwelling characteristics (number of rooms, area in square meters) because consumer price indices are available for only 12 metropolitan areas.
- Two-step procedure (notation preserved):
  - For each sub-region k = [1, 2,...,7] of each state s = [1, 2,...,27] and each year t = [2004,...,2015], construct a rental spatial price difference index:
    r_{s,k,t} = ( m_{s,k,t} / n_{s,k,t} ) ( m_{t}^* n_{t}^* )^{-1}
    where m is the average monthly rent price for the cluster s,k, while n is the average number of rooms per household for the cluster, and the stars denote national averages.
  - Given that overall spatial price differences can be well approximated by a linear function of housing spatial price differences, use the parameter from Azzoni and Almeida (2016), assumed homogeneous across regions, and the heterogeneous rental spatial-price difference index to fit an overall spatial price difference index p̂_{s,k,t} = φ r_{s,k,t}.
  - Use p̂_{s,k,t} to obtain adjusted household incomes for analysis of income distributions and trends.
- Empirical checks:
  - The authors’ overall Gini estimates are nearly perfectly correlated with IBGE official estimates.
  - Higher household income per capita regions tend to face price levels above the national average; lower income regions face price levels below the national average, so spatial adjustment compresses nominal income differences and decreases the overall inequality indicator.

### Historical trends in regional inequality (2004–14)
- IBGE Gini (2004–2014): fell from 0.54 in 2004 to 0.49 in 2014.
- Authors' adjusted Gini (PNAD microdata, spatial-price adjusted): declined from 0.55 to 0.50 over 2004–2014.
- Between-state inequality:
  - Between-states inequality decreased as a share of total inequality because incomes grew faster in poorer regions (North, Northeast, Midwest).
  - GE and Atkinson indices used as consistency checks and are perfectly decomposable into within and between components.
- Within-state inequality:
  - Declined across most states; decline driven primarily by substantially higher income growth rates for lower-income households in nearly all states.
  - Convergence in within-state inequality was stronger in states with higher initial inequality in 2004 (especially after excluding outliers SC and DF).
  - Distributional dispersion: Gini of the most unequal state in 2014 was 18 percent higher than the national Gini; the least unequal state’s Gini was almost 20 percent lower than the national ratio.
  - Standard deviation of state Gini coefficients declined from 0.035 to 0.033 between 2004 and 2014.
- Income-position mapping:
  - Households in the lowest and highest deciles of state income distributions correspond to the lowest/highest deciles of the national distribution.
  - State medians could fall anywhere between the 30th and the 70th percentile of the national distribution; these differences shrank between 2004 and 2014 (convergence around the median).

### 2015 snapshot and 2015–2016 outlook
- 2015 snapshot:
  - No evidence of reversal of equality gains in PNAD 2015: with a drop in employed population, real gross household earnings contracted in 2015 across all professions for the first time in 11 years.
  - Real income decline touched the entire income distribution, but it was more severe in higher income brackets, leading to a slight fall in inequality.
  - Official Gini for all income sources fell from 0.497 in 2014 to 0.491 in 2015.
  - The Gini calculated for labor income fell from 0.490 to 0.485 (period implied in source).
  - Gini for household income fell from 0.494 to 0.493.
- 2016 outlook:
  - Continuation of the recession through 2016 may have dented equality gains.
  - Preliminary 2016 inequality estimates from FGV Social suggest that inequality widened slightly for the first time in 22 years.
  - World Bank (2017) estimates:
    - Number of poor in Brazil will likely increase by 2.5‒3.6 million by 2017.
    - Gini index will increase from 0.51 to 0.52‒0.54 by 2017.
  - Among the “new poor”, young, skilled workers in the service sector will represent the higher share of those falling below the poverty line due to the crisis.

### Tax policy and distributional incidence
- Brazil’s overall tax system relies relatively more on indirect taxes, which are regressive.
- Ratio of direct to indirect taxes at the general government level: 45 percent in 2016.
- Effective personal income tax (PIT) rates:
  - Effective PIT rates that take into account admissible deductions do not seem progressive (green line in chart referenced).
  - When accounting for taxation of dividends at corporate level (adding corporate taxes imputed to individuals), the system’s progressivity appears to be restored (red line in chart referenced).
- Empirical finding: Lustig and others (2014) find personal income taxes in Brazil are progressive and redistributive, contributing to reducing the Gini of after-tax incomes by 1.9 percent in 2009.
- Tax burden shares reported by Amaral and others (2016):
  - Average Brazilian worker pays 15 percent of his gross income in income taxes.
  - 3 percent in asset taxes.
  - 24 percent in consumption taxes.
  - Those making up to R$3,000 per month pay 24 percent of their gross income in consumption taxes.
  - Those making more than 10,000 pay 17 percent in consumption taxes.
- Conclusion: While PIT is progressive, excessive reliance on consumption taxes makes the overall system regressive.

### Minimum wage, public-sector wages, and labor market effects
- Minimum wage policy:
  - Supported upward social mobility for lower classes during strong growth.
  - Ambiguous effect on inequality due to potential offsets on employment and inflation (Jaumotte and Osorio Buitron, 2015).
  - Maurizio (2014): increases in minimum wage led to wage compression, reducing inequality among wage earners.
  - Average hourly wage for a worker with a given level of education rose much faster among the poor than for the rest of the population in the last decade mainly because of the minimum wage policy.
  - Increases in real minimum wage above productivity coincided with declines in unemployment supported by strong output growth pre-2014.
  - Brazil’s contraction in investment and growth after 2014 reduced labor demand; further minimum wage increases above productivity may affect employment negatively, more for unskilled workers (IMF, 2015; Jaumotte and Osorio Buitron, 2015). This would lead to higher before-tax (gross) inequality.
- Public-sector wage premium:
  - Public sector wage increases were systematically above private sector wage growth.
  - Estimated median premium on public sector jobs across comparable professions was about 50 percent up to the secondary education level (2014 estimate referenced).
  - Mincer regressions (~150,000 observations, 50+ controls) find predicted earnings increasing with years of schooling in both sectors, but public sector earnings consistently higher.
  - Up to secondary education level, the 25 percent lowest predicted earnings in the public sector are higher than the median earnings in the private sector.
  - At least 75 percent of workers would benefit by moving from private to public sector in comparable jobs at all education levels (based on observed characteristics).
  - Caveat: unobservable attributes (risk aversion, productivity) may bias premium estimates; alternative modelling (Emilio and others, 2012) suggests premia may be smaller or insignificant under certain assumptions.

### Bolsa Família and targeted transfers
- Coverage and fiscal size:
  - Beneficiary coverage increased from about 6.5 million households in 2004 to over 14 million in 2014 (56 million people).
  - Budgetary appropriations increased from about 0.3 percent of GDP to 0.6 percent of GDP over the same period.
- Impact estimates:
  - World Bank (2017) estimates that 58 percent of the decline in extreme poverty in Brazil over 2004–14 was due to Bolsa Família transfers.
  - Soares and others (2006): in 2005 about 80 percent of Bolsa Família and other cash-targeted programs went to families below the poverty line (half of the minimum wage per capita); the program was responsible for 21 percent of the decline in the Gini coefficient between 1995 and 2005.
  - Neri, Vaz and Ferreira de Souza (2013) estimate the multiplier effect of Bolsa Família to be 1.78.
- Distributional concern: Bolsa Família had a higher impact on bottom-quartile incomes, but effects on the top quartile also observed, possibly due to high estimated multiplier.

### Education spending, targeting, and inequality of opportunity
- Public universities are tuition-free and more accessible to children of wealthier parents.
- Public university enrollment and household income (PNAD 2014):
  - Nearly half of the public university student population belongs to households in the top quartile of the income distribution.
  - Only 9 percent of university students come from families in the bottom quartile.
- Probability of attending public university (logit estimates controlling for age, gender, race, region):
  - A student in the 25th percentile of the income distribution has a 2 percent probability of attending a public university.
  - A student in the 99th percentile has more than 30 percent probability of attending a public university.
- Fiscal cost and targeting:
  - Funding a student at higher education level costs about four times as much as funding a student at the secondary education level in Brazil.
  - OECD average ratio: 150 percent.
- Policy implication: Redirecting spending from tertiary to primary and secondary education would improve overall welfare and equality; targeting education spending on the poor and cutting subsidies to the rich can generate fiscal savings while improving equality of opportunities.

### Regression analysis: stylized facts and drivers of inequality (state-level panel)
- Data and setup:
  - State-level inequality indicators constructed from PNAD aggregated to states (27 states panel).
  - State-level regressors include: average income, civil servants’ income, tax revenues (personal and corporate income taxes collected at federal level as share of state GDP), share of formal sector workers, share of civil servants in employment, schooling, per capita Bolsa Família budget, employment rate, formalization rate, average schooling for bottom and top quartiles.
- First set of regressions (logs of spatially price-adjusted incomes for Q1 and Q4):
  - Bottom quartile incomes have been more responsive to overall income growth than top quartiles.
  - Increased schooling significantly raised incomes of the poor but did not affect incomes of the top quartile.
  - Bolsa Família had a higher impact on the bottom quartile incomes, but also appears to have increased the income of the top quartile, likely due to the program's high estimated multiplier effect.
  - Labor formalization and civil servants’ incomes have had opposite effects on growth of top and bottom quartile income.
- Second regression (within-state Gini g_s,t on same regressors):
  - Findings: employment, labor formalization, income growth, Bolsa Família budgets and schooling contribute to explaining inequality (coefficients reported in Table 5).

### Findings on drivers of inequality (2004–14 / 2004−15)
- Decline in inequality observed both between and within Brazil’s 27 states from 2004 to 2014.
- Falling inequality attributed to convergence in households’ incomes in the proximity of the mid-point of the state-wide distribution, stronger in more unequal states.
- Increased schooling and labor formalization explain the largest share of the decline in the Gini.
- Growth of average incomes and Bolsa Família also contributed to lowering inequality.
- Growth of incomes of civil servants has affected equality negatively.
- Income taxes are not a significant determinant of inequality, possibly because the PNAD may be underestimating the income of the top 1 percent of the population.
- Findings on Bolsa Família align with previous literature underlining the program’s redistributive power.
- Figure note: Population-weighted averages from panel regression of 27 states. Percent changes are slightly overstated due to log-linear approximation.

### Data limitations and interpretive caveats
- PNAD may underreport income received by the top 1 percent of the income distribution in Brazil (Souza, 2013).
- If income of the richest segment not captured in PNAD has increased over time, inequality statistics could overstate the decline in inequality.
- Medeiros and others (2015) combining DIRPF with PNAD conclude that a growing share of income was received by the top 1 percent between 2006 and 2012, causing overall inequality to stagnate over that period; DIRPF adjustment was not possible in this study because DIRPF data is not available for the entire period under consideration.
- Implicit assumption: estimated parameters are homogeneous and linear across states; extrapolating results to the future or to specific states must be done cautiously.

### Data sources, coverage, and variable definitions
- Main source: PNAD (Pesquisa Nacional por Amostra de Domicílios), a National Household Survey conducted yearly.
- PNAD collects data on nearly 360,000 individuals distributed through about 140,000 households.
- To build time series for each state between 2004−15, around 6 million data points were used.
- PNAD has two annual datasets: one with collective household characteristics and one with individual characteristics.
- Survey design incorporated by using PNAD-provided weights and adjusting estimates and reported errors by double clustering at the state and household levels.
- Constructed series for each state: inequality indices (Gini, General Entropy and Atkinson) based on household income per capita; indicators including average household income per capita; average wage per active worker; share of formal workers in total population; share of public sector employment in total employment; average years of schooling for adult population (above 16); share of population between 17 and 30 years old in private universities; share of population between 17 and 30 years old in public universities; per capita Bolsa Família budget.
- Bolsa Família variable:
  - Data from the Brazilian Ministry of Social Development.
  - Constructed as yearly real per capita outlays: total Bolsa Família state budget divided by state population after adjusting for spatial price differences and inflation.
- Tax revenue variable:
  - Federal personal and corporate income tax revenues collected by the states (reported by Receita Federal) divided by state GDP from IBGE.
- Formal work variable:
  - Share of formal workers in total for each state, constructed from PNAD answers with explicit inclusion/exclusion rules listed in source.
- Employment rate variable:
  - Share of persons who were working in the reference week in the total labor force.
- Average income variable:
  - Average household per capita income for the state aggregated from household income indicator and adjusted for spatial price differences.
  - Estimates of inequality based on after-tax per capita income as reported in PNAD, which includes labor income, retirement benefits, disability and survivors’ pensions, social transfers and income from financial and real assets.
- Civil servant income:
  - Equal to household income per capita for households whose reference person is a civil servant, adjusted for spatial price differences.
- Schooling variables:
  - Average number of years of schooling for persons in the lower and top quartiles of the national household income per capita distribution.

### Policy conclusions and recommendations
- Preserve equality gains while striking a balance between fiscal sustainability and income equality amid a sharp recession eroding incomes of the poor.
- Reform priorities highlighted: social security, labor market and tax reforms.
- Policy measures to preserve inclusiveness without further increasing spending:
  - Moderate civil servants’ wage growth.
  - Use direct instruments to provide benefits based on need, such as Bolsa Família.
  - Improve access to education and educational attainment for lower-income families by redirecting resources currently funding universal, tuition-free access to tertiary education, while continuing to support university students based on need.
  - Tax reform: rely more on direct taxation and less on indirect taxes.
  - Minimum wage policy: provide appropriate remuneration for the poor without discouraging formal employment.

### Appendix II — Selected econometric results and tables (high-level highlights)
- Table 2 (2014 percentile incomes, spatial-price adjusted): selected 100th-percentile state values include DF: 12,938; SP: 8,192; MG: 7,702; RS: 7,768; RJ: 7,594; range (Max - Min) by percentile reported: 49, 71, 84, 97, 110, 121, 128, 138, 139, 146, 155, 165, 193, 230, 282, 356, 464, 604, 854, 1,637.
- Table 3 (Spatial price adjustment factors, 2004–2014): selected state-year values include RO: 2004=0.875 ... 2014=0.960; PI: 2004=0.650 ... 2014=0.668; RJ: 2004=1.245 ... 2014=1.201; SP: 2004=1.133 ... 2014=1.177; DF: 2004=1.286 ... 2014=1.241.
- Table 4 (Panel regressions, drivers of top and bottom quartile incomes, 2004–14): reported coefficients (examples) include 1.402*** (0.099); 0.994*** (0.108); 0.722*** (0.079); 0.734*** (0.072); and diagnostics: Observations 297; Number of states 27; R2 Overall across specs: 0.927, 0.952, 0.966, 0.966, 0.949, 0.950, 0.958, 0.958; R2 Within across specs: 0.905, 0.938, 0.956, 0.956, 0.876, 0.880, 0.898, 0.899.
- Table 5 (Drivers of within-state Gini, 2004–14): selected coefficients (examples) include -0.158*** (0.021); -0.092*** (0.030); -0.151*** (0.016); 0.071*** (0.020); -0.034*** (0.011); -0.020*** (0.006); -0.027*** (0.007); Constants examples: 1.017*** (0.084), 0.708*** (0.127). Diagnostics: Observations 297; Number of states 27; R2 Overall across specs: 0.776, 0.801, 0.842, 0.845, 0.276, 0.407, 0.675, 0.670; R2 Within across specs: 0.445, 0.506, 0.607, 0.617, 0.440, 0.490, 0.598, 0.602.
- Table 6 (Logistic regression: determinants of attending public university, 2014): selected marginal effects include LN of HH income per capita: 0.025*** (0.001) in spec (1); 0.022*** (0.001) in spec (2); Age: 0.001*** (0.000); Male: -0.004*** (0.001); Black: -0.014*** (0.003); Northeast Dummy: 0.021*** (0.001); Pseudo-R2: 0.0716 (spec 1), 0.0953 (spec 2); Observations: 40,385.

*Source: IMF Working Paper (wp17225), "1. The Cost of Living Adjustment" (PDF chapter).*

### 1. The Cost of Living Adjustment  ______________________________________________ 6

### 1. The Cost of Living Adjustment

### Overview: recent social and inequality trends in Brazil
- Poverty (World Bank international poverty line): from 25 percent of the population in 2004 to 8.5 percent in 2014.
- Extreme poverty: from 12 percent in 2004 to 4 percent in 2014.
- World Bank Gini coefficient: from 0.60 in 1990 to 0.51 in 2014.
- Distribution of labor income (PNAD 2014): top decile = 40 percent of labor income of all Brazilian families; top 1 percent ≈ 12 percent; top 0.1 percent ≈ 2.5 percent; "Half percent of all labor income is concentrated in the top 0.01 percent."
- Recession starting in 2014 affected progress: unemployment reached 13 percent in 2017; by early-2017, more than one in four young adults in Brazil were unemployed.
- Fiscal context: post-recession government faces a long period of fiscal consolidation; observing the cap on federal government non-interest expenditures will require restraint across all categories of spending in the medium term.
- Main drivers of the past decline in inequality (paper’s regression findings): income growth, higher schooling levels, labor formalization; Bolsa Família contributed to income convergence; civil servants’ wage growth slowed gains in equality.
- Policy implications highlighted: ensure fiscal sustainability while improving spending efficiency and avoiding adverse distributional effects; better targeting of social benefits, rationalizing the tax system, and moderating civil servants’ wages are key as labor formalization and income growth slow.

### Data and methodological innovation: spatial price adjustment (Box 1)
- Rationale: living standards (price levels) differ across Brazilian regions; adjusting incomes for spatial price differences is necessary to distinguish nominal and real differences in incomes and to facilitate comparisons across states.
- Proxy used: declared household rent prices from PNAD and dwelling characteristics (number of rooms, area in square meters) because consumer price indices are available for only 12 metropolitan areas.
- Two-step procedure (summary of steps and notation preserved):
  - For each sub-region k = [1, 2,...,7] of each state s = [1, 2,...,27] and each year t = [2004,...,2015], construct a rental spatial price difference index:
    r_{s,k,t} = ( m_{s,k,t} / n_{s,k,t} ) ( m_{t}^* n_{t}^* )^{-1}
    where m is the average monthly rent price for the cluster s,k, while n is the average number of rooms per household for the cluster, and the stars denote national averages.
  - Given that overall spatial price differences can be well approximated by a linear function of housing spatial price differences, use the parameter from Azzoni and Almeida (2016), assumed homogeneous across regions, and the heterogeneous rental spatial-price difference index to fit an overall spatial price difference index p̂_{s,k,t} = φ r_{s,k,t}.
  - Use p̂_{s,k,t} to obtain adjusted household incomes for analysis of income distributions and trends.
- Empirical checks:
  - The authors’ overall Gini estimates are nearly perfectly correlated with IBGE official estimates.
  - Higher household income per capita regions tend to face price levels above the national average; lower income regions face price levels below the national average, so spatial adjustment compresses nominal income differences and decreases the overall inequality indicator.

### Historical trends in regional inequality (2004–14)
- IBGE Gini (2004–2014): fell from 0.54 in 2004 to 0.49 in 2014.
- Authors' adjusted Gini (PNAD microdata, spatial-price adjusted): declined from 0.55 to 0.50 over 2004–2014.
- Between-state inequality:
  - Between-states inequality decreased as a share of total inequality because incomes grew faster in poorer regions (North, Northeast, Midwest).
  - GE and Atkinson indices used as consistency checks and are perfectly decomposable into within and between components.
- Within-state inequality:
  - Declined across most states; decline driven primarily by substantially higher income growth rates for lower-income households in nearly all states.
  - Convergence in within-state inequality was stronger in states with higher initial inequality in 2004 (especially after excluding outliers SC and DF).
  - Distributional dispersion: Gini of the most unequal state in 2014 was 18 percent higher than the national Gini; the least unequal state’s Gini was almost 20 percent lower than the national ratio.
  - Standard deviation of state Gini coefficients declined from 0.035 to 0.033 between 2004 and 2014.
- Income-position mapping:
  - Households in the lowest and highest deciles of state income distributions correspond to the lowest/highest deciles of the national distribution.
  - State medians could fall anywhere between the 30th and the 70th percentile of the national distribution; these differences shrank between 2004 and 2014 (convergence around the median).
- 2015 snapshot:
  - No evidence of reversal of equality gains in PNAD 2015: with a drop in employed population, real gross household earnings contracted in 2015 across all professions for the first time in 11 years.
  - Real income decline touched the entire income distribution, but it was more severe in higher income brackets, leading to a slight fall in inequality.
  - Official Gini for all income sources fell from 0.497 in 2014 to (value truncated in source).

### Key figures and statistical notes (preserve original numeric expressions)
- Poverty: 25 percent → 8.5 percent (2004 → 2014).
- Extreme poverty: 12 percent → 4 percent (2004 → 2014).
- World Bank Gini: 0.60 (1990) → 0.51 (2014).
- Labor income shares: top decile = 40 percent; top 1 percent ≈ 12 percent; top 0.1 percent ≈ 2.5 percent; "Half percent of all labor income is concentrated in the top 0.01 percent."
- Unemployment: 13 percent in 2017; more than one in four young adults unemployed by early-2017.
- Authors’ adjusted Gini: 0.55 → 0.50 (2004 → 2014).
- IBGE Gini: 0.54 → 0.49 (2004 → 2014).
- Standard deviation of state Gini coefficients: 0.035 → 0.033 (2004 → 2014).

*Source: IMF Working Paper (wp17225), "1. The Cost of Living Adjustment" (PDF chapter).*

### 0.491 in 2015. The Gini calculated for labor income fell from 0.490 to 0.485 and, in the case of

### wp17225 - 0.491 in 2015. The Gini calculated for labor income fell from 0.490 to 0.485 and, in the case of

### Inequality trends and 2015–2016 outlook
- Gini overall: 0.491 in 2015.
- Gini for labor income: fell from 0.490 to 0.485 (period implied in source).
- Gini for household income: fell from 0.494 to 0.493.
- Continuation of the recession through 2016 may have dented equality gains.
- Preliminary 2016 inequality estimates from FGV Social suggest that inequality widened slightly for the first time in 22 years.
- World Bank (2017) estimates:
  - Number of poor in Brazil will likely increase by 2.5‒3.6 million by 2017.
  - Gini index will increase from 0.51 to 0.52‒0.54 by 2017.
- Among the “new poor”, young, skilled workers in the service sector will represent the higher share of those falling below the poverty line due to the crisis.

### Tax policy and distributional incidence
- Brazil’s overall tax system relies relatively more on indirect taxes, which are regressive.
- Ratio of direct to indirect taxes at the general government level: 45 percent in 2016.
- Effective personal income tax (PIT) rates:
  - Effective PIT rates that take into account admissible deductions do not seem progressive (green line in chart referenced).
  - When accounting for taxation of dividends at corporate level (adding corporate taxes imputed to individuals), the system’s progressivity appears to be restored (red line in chart referenced).
- Empirical finding: Lustig and others (2014) find personal income taxes in Brazil are progressive and redistributive, contributing to reducing the Gini of after-tax incomes by 1.9 percent in 2009.
- Tax burden shares reported by Amaral and others (2016):
  - Average Brazilian worker pays 15 percent of his gross income in income taxes.
  - 3 percent in asset taxes.
  - 24 percent in consumption taxes.
  - Those making up to R$3,000 per month pay 24 percent of their gross income in consumption taxes.
  - Those making more than 10,000 pay 17 percent in consumption taxes.
- Conclusion: While PIT is progressive, excessive reliance on consumption taxes makes the overall system regressive.

### Minimum wage, public-sector wages, and labor market effects
- Minimum wage policy:
  - Supported upward social mobility for lower classes during strong growth.
  - Ambiguous effect on inequality due to potential offsets on employment and inflation (Jaumotte and Osorio Buitron, 2015).
  - Maurizio (2014): increases in minimum wage led to wage compression, reducing inequality among wage earners.
  - Average hourly wage for a worker with a given level of education rose much faster among the poor than for the rest of the population in the last decade mainly because of the minimum wage policy.
  - Increases in real minimum wage above productivity coincided with declines in unemployment supported by strong output growth pre-2014.
  - Brazil’s contraction in investment and growth after 2014 reduced labor demand; further minimum wage increases above productivity may affect employment negatively, more for unskilled workers (IMF, 2015; Jaumotte and Osorio Buitron, 2015). This would lead to higher before-tax (gross) inequality.
- Public-sector wage premium:
  - Public sector wage increases were systematically above private sector wage growth.
  - Estimated median premium on public sector jobs across comparable professions was about 50 percent up to the secondary education level (2014 estimate referenced).
  - Empirical methods (Mincer regressions with ~150,000 observations, 50+ controls) find predicted earnings increasing with years of schooling in both sectors, but public sector earnings consistently higher.
  - Up to secondary education level, the 25 percent lowest predicted earnings in the public sector are higher than the median earnings in the private sector.
  - At least 75 percent of workers would benefit by moving from private to public sector in comparable jobs at all education levels (based on observed characteristics).
  - Some evidence of compression of the premium at higher educational levels, but premium remains high across all years of schooling.
  - Caveat: unobservable attributes (risk aversion, productivity) may bias premium estimates; alternative modelling (Emilio and others, 2012) suggests premia may be smaller or insignificant under certain assumptions.

### Bolsa Família and targeted transfers
- Bolsa Família coverage:
  - Beneficiary coverage increased from about 6.5 million households in 2004 to over 14 million in 2014 (56 million people).
  - Budgetary appropriations increased from about 0.3 percent of GDP to 0.6 percent of GDP over the same period.
- Impact estimates:
  - World Bank (2017) estimates that 58 percent of the decline in extreme poverty in Brazil over 2004–14 was due to Bolsa Família transfers.
  - Soares and others (2006): in 2005 about 80 percent of Bolsa Família and other cash-targeted programs went to families below the poverty line (half of the minimum wage per capita); the program was responsible for 21 percent of the decline in the Gini coefficient between 1995 and 2005.
  - Neri, Vaz and Ferreira de Souza (2013) estimate the multiplier effect of Bolsa Família to be 1.78.
- Distributional concern: Bolsa Família had a higher impact on bottom-quartile incomes, but effects on the top quartile also observed, possibly due to high estimated multiplier.

### Education spending, targeting, and inequality of opportunity
- Public universities are tuition-free and more accessible to children of wealthier parents.
- Public university enrollment and household income (PNAD 2014):
  - Nearly half of the public university student population belongs to households in the top quartile of the income distribution.
  - Only 9 percent of university students come from families in the bottom quartile.
- Probability of attending public university (logit estimates controlling for age, gender, race, region):
  - A student in the 25th percentile of the income distribution has a 2 percent probability of attending a public university.
  - A student in the 99th percentile has more than 30 percent probability of attending a public university.
- Fiscal cost and targeting:
  - Funding a student at higher education level costs about four times as much as funding a student at the secondary education level in Brazil.
  - OECD average ratio: 150 percent (i.e., funding higher education costs 1.5 times secondary on average).
- Policy implication: Redirecting spending from tertiary to primary and secondary education would improve overall welfare and equality; targeting education spending on the poor and cutting subsidies to the rich can generate fiscal savings while improving equality of opportunities.

### Regression analysis: stylized facts and drivers of inequality (state-level panel)
- Data and setup:
  - State-level inequality indicators constructed from PNAD aggregated to states (~27 states panel).
  - State-level regressors include: average income, civil servants’ income, tax revenues (personal and corporate income taxes collected at federal level as share of state GDP), share of formal sector workers, share of civil servants in employment, schooling, per capita Bolsa Família budget, employment rate, formalization rate, average schooling for bottom and top quartiles.
- First set of regressions: natural logs of spatially price-adjusted average household income per capita for bottom (Q1) and top (Q4) quartiles regressed on:
  - log average income (y_s,t), log Bolsa Família per capita expenditures (b_s,t), tax revenues t_s,t, civil servants’ characteristics vector c′_s,t = [w_s,t, k_s,t] (w = log average income of households headed by civil servants; k = share of civil servants in workforce), labor market vector l′_s,t = [e_s,t, f_s,t] (e = employment rate; f = formalization rate), educational vector h′_s,t = [h_s,t_Q1, h_s,t_Q4] (average schooling years for bottom and top quartiles), state fixed effects and residuals (see equations (1) in source).
- Key stylized regression findings:
  - Bottom quartile incomes have been more responsive to overall income growth than top quartiles.
  - Increased schooling significantly raised incomes of the poor but did not affect incomes of the top quartile.
  - Bolsa Família had a higher impact on the bottom quartile incomes, but also appears to have increased the income of the top quartile, likely due to the program's high estimated multiplier effect.
  - Labor formalization and civil servants’ incomes have had opposite effects on growth of top and bottom quartile income.
- Second regression: within-state Gini (g_s,t) regressed on the same set of regressors (see equation (2) in source).
  - Findings: employment, labor formalization, income growth, Bolsa Família budgets and schooling contribute to explaining inequality (coefficients reported in Table 5 in source).

*Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17225.pdf*

### Appendix II) are mostly significant and bear the expected sign, and the trajectories over time of

### wp17225 - Appendix II) are mostly significant and bear the expected sign, and the trajectories over time of

### Findings on drivers of inequality (2004–14 / 2004−15)
- Decline in inequality observed both between and within Brazil’s 27 states from 2004 to 2014.
- Falling inequality attributed to convergence in households’ incomes in the proximity of the mid-point of the state-wide distribution, stronger in more unequal states.
- Increased schooling and labor formalization explain the largest share of the decline in the Gini.
- Growth of average incomes and Bolsa Família also contributed to lowering inequality.
- Growth of incomes of civil servants has affected equality negatively.
- Income taxes are not a significant determinant of inequality, possibly because the PNAD may be underestimating the income of the top 1 percent of the population.
- Findings on Bolsa Família are in line with previous literature which underlines the redistributive power of the program (Neri, 2010; Azzoni and Silveira-Neto, 2012).
- Figure note: Population-weighted averages from panel regression of 27 states. Percent changes are slightly overstated due to log-linear approximation.

### Data limitations and interpretive caveats
- PNAD may underreport income received by the top 1 percent of the income distribution in Brazil (Souza, 2013).
- If income of the richest segment not captured in PNAD has increased over time, inequality statistics could overstate the decline in inequality.
- Medeiros and others (2015) combining DIRPF with PNAD conclude that a growing share of income was received by the top 1 percent between 2006 and 2012, causing overall inequality to stagnate over that period; DIRPF adjustment was not possible in this study because DIRPF data is not available for the entire period under consideration.
- Implicit assumption: estimated parameters are homogeneous and linear across states. This limitation implies extrapolating results to draw conclusions for the future or for specific states must be done cautiously.

### Data sources, coverage, and variable definitions
- Main source: PNAD (Pesquisa Nacional por Amostra de Domicílios), a National Household Survey conducted yearly.
- PNAD collects data on nearly 360,000 individuals distributed through about 140,000 households.
- To build time series for each state between 2004−15, around 6 million data points were used.
- PNAD has two annual datasets: one with collective household characteristics and one with individual characteristics.
- Survey design incorporated by using PNAD-provided weights and adjusting estimates and reported errors by double clustering at the state and household levels.
- Constructed series for each state: inequality indices (Gini, General Entropy and Atkinson) based on household income per capita; indicators including average household income per capita; average wage per active worker; share of formal workers in total population; share of public sector employment in total employment; average years of schooling for adult population (above 16); share of population between 17 and 30 years old in private universities; share of population between 17 and 30 years old in public universities; per capita Bolsa Família budget.

- Bolsa Família variable:
  - Data from the Brazilian Ministry of Social Development.
  - Constructed as yearly real per capita outlays: total Bolsa Família state budget divided by state population after adjusting for spatial price differences and inflation.
- Tax revenue variable:
  - Federal personal and corporate income tax revenues collected by the states (reported by Receita Federal) divided by state GDP from IBGE.
- Formal work variable:
  - Expressed as the share of formal workers in total for each state, constructed from PNAD answers.
  - “formal worker” includes: Formal contract (carteira assinada); Military; Civil servant; Employer/entrepreneur; Domestic employee with a formal contract (doméstico com carteira assinada); Unpaid/Voluntary work (não remunerado); Self-employed as a manager/director; Self-employed as an artist.
  - “informal worker” includes: No formal contract (sem carteira assinada); Domestic with no formal contract (doméstico sem carteira assinada); Self-employed (except for manager/director and artist); Self-consumption worker in production; Self-consumption worker in construction.
- Employment rate variable:
  - Share of persons who were working in the reference week in the total labor force (persons economically active).
- Average income variable:
  - Average household per capita income for the state aggregated from household income indicator and adjusted for spatial price differences.
  - Estimates of inequality based on after-tax per capita income as reported in PNAD, which includes labor income, retirement benefits, disability and survivors’ pensions, social transfers and income from financial and real assets.
- Civil servant income:
  - Equal to household income per capita for households whose reference person is a civil servant, adjusted for spatial price differences.
- Schooling variables:
  - Average number of years of schooling for persons in the lower and top quartiles of the national household income per capita distribution.

### Policy conclusions and recommendations
- Preserve equality gains while striking a balance between fiscal sustainability and income equality amid a sharp recession eroding incomes of the poor.
- Reform priorities highlighted: social security, labor market and tax reforms.
- Policy measures to preserve inclusiveness without further increasing spending:
  - Moderate civil servants’ wage growth.
  - Use direct instruments to provide benefits based on need, such as Bolsa Família (IMF, 2014).
  - Improve access to education and educational attainment for lower-income families by redirecting resources currently funding universal, tuition-free access to tertiary education, while continuing to support university students based on need.
  - Tax reform: rely more on direct taxation and less on indirect taxes.
  - Minimum wage policy: provide appropriate remuneration for the poor without discouraging formal employment.

*Source: IMF staff estimates; PNAD microdata; IMF staff calculations.*

### Appendix II. Econometric Results and Tables

### Appendix II. Econometric Results and Tables

### Table 2 — Brazil: Household Income per Capita by State Household Income Percentile, 2014
(Percentile average, in current reais per capita, adjusted for spatial price differences)
- Dataset: Authors' estimates with dataset consolidated from PNAD microdata.
- Coverage: Percentile averages (5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, 100) reported for each state (examples below preserve values exactly as in the source).
- Selected state 100th-percentile values:
  - RO: 5,689
  - AC: 5,834
  - AP: 5,541
  - RR: 6,431
  - PA: 4,594
  - AM: 4,957
  - TO: 6,231
  - MA: 5,151
  - PI: 6,098
  - CE: 5,159
  - RN: 5,851
  - PB: 6,623
  - PE: 5,197
  - AL: 4,807
  - SE: 5,420
  - BA: 6,333
  - MG: 7,702
  - ES: 6,637
  - RJ: 7,594
  - SP: 8,192
  - PR: 7,675
  - SC: 6,423
  - RS: 7,768
  - MS: 7,660
  - MT: 7,037
  - GO: 6,355
  - DF: 12,938
- Range (Max - Min) by percentile (as reported):
  - 49, 71, 84, 97, 110, 121, 128, 138, 139, 146, 155, 165, 193, 230, 282, 356, 464, 604, 854, 1,637

### Table 3 — Brazil: Time Series of Spatial Price Adjustment Factors by State
(Average for the state, relative to national average; years 2004–2014)
- Dataset: Authors' estimates with dataset consolidated from PNAD microdata.
- States include RO, AC, AP, RR, PA, AM, TO, MA, PI, CE, RN, PB, PE, AL, SE, BA, MG, ES, RJ, SP, PR, SC, RS, MS, MT, GO, DF.
- Selected time-series values (state / year = value as reported):
  - RO: 2004=0.875 2005=0.903 2006=0.930 2007=0.877 2008=0.919 2009=0.959 2010=0.977 2011=0.995 2012=0.874 2013=0.847 2014=0.960
  - PI: 2004=0.650 2005=0.675 2006=0.763 2007=0.695 2008=0.780 2009=0.710 2010=0.679 2011=0.648 2012=0.638 2013=0.679 2014=0.668
  - RJ: 2004=1.245 2005=1.225 2006=1.206 2007=1.193 2008=1.213 2009=1.238 2010=1.235 2011=1.232 2012=1.210 2013=1.202 2014=1.201
  - SP: 2004=1.133 2005=1.111 2006=1.102 2007=1.094 2008=1.092 2009=1.103 2010=1.126 2011=1.148 2012=1.158 2013=1.180 2014=1.177
  - DF: 2004=1.286 2005=1.239 2006=1.257 2007=1.256 2008=1.274 2009=1.273 2010=1.300 2011=1.328 2012=1.276 2013=1.254 2014=1.241

### Table 4 — Brazil: Drivers of Top and Bottom Quartile Household Income per Capita (2004–14)
(Panel regressions; robust standard errors clustered at the state level; significance: *** p<0.01, ** p<0.05, * p<0.1)
- Data sources: PNAD microdata; Secretaria da Receita Federal data on taxes; Ministry of Social Development data on Bolsa Família.
- Dependent variables and model specifications include:
  - Spatial-price adjusted bottom quartile HH income per capita, log
  - Spatial-price adjusted top quartile HH income per capita, log
  - Average spatial-price adjusted real HH income per capita, log
  - Civil servants' spatial-price adjusted real HH income per capita, log
- Selected reported coefficients (coefficient (standard error)):
  - 1.402*** (0.099)
  - 0.994*** (0.108)
  - 0.722*** (0.079)
  - 0.734*** (0.072)
  - 0.573*** (0.042)
  - 0.496*** (0.053)
  - 0.594*** (0.058)
  - 0.618*** (0.067)
  - -0.141 (0.095)
  - -0.186** (0.073)
  - -0.131** (0.058)
  - -0.135** (0.056)
  - 0.182*** (0.041)
  - 0.173*** (0.039)
  - 0.153*** (0.032)
  - 0.145*** (0.034)
  - Other coefficients and diagnostics:
    - Observations: 297 (per specification)
    - Number of states: 27
    - R2 Overall across specs: 0.927, 0.952, 0.966, 0.966, 0.949, 0.950, 0.958, 0.958
    - R2 Within across specs: 0.905, 0.938, 0.956, 0.956, 0.876, 0.880, 0.898, 0.899
    - R2 Between: reported as 1 1 1 1 1 1 (in table source)

- Variable labels (as used in the regressions):
  - Spatial-price adjusted real Bolsa Familia expenditure per capita, log
  - Share of formal workers in workforce, pct
  - Employment Rate, pct
  - Average Formal Schooling Years for Adults, Bottom Quartile
  - Average Formal Schooling Years for Adults, Top Quartile
  - Federal Income Tax Revenue, share of state GDP
  - Share of civil servants in workforce, pct
  - Civil servants' spatial-price adjusted real HH income per capita, log

### Table 5 — Brazil: Drivers of Within-State Household Income Inequality in Brazil (2004–14)
(Dependent variable: State Gini Coefficient; models with State Fixed Effects and Random Effects; robust standard errors clustered at state level)
- Data sources: PNAD microdata; Secretaria da Receita Federal data on taxes; Ministry of Social Development data on Bolsa Família.
- Selected reported coefficients (coefficient (standard error)):
  - -0.158*** (0.021)
  - -0.092*** (0.030)
  - -0.036 (0.025)
  - -0.019 (0.019)
  - -0.151*** (0.016)
  - -0.142*** (0.016)
  - -0.042** (0.017)
  - -0.039** (0.017)
  - 0.071*** (0.020)
  - 0.075*** (0.018)
  - 0.065*** (0.015)
  - 0.059*** (0.015)
  - -0.034*** (0.011)
  - -0.010 (0.010)
  - -0.011 (0.009)
  - -0.020*** (0.006)
  - -0.002** (0.001)
  - -0.002** (0.001)
  - -0.002*** (0.001)
  - -0.002*** (0.001)
  - -0.027*** (0.007)
  - -0.028*** (0.007)
  - -0.024*** (0.006)
  - -0.023*** (0.005)
  - 0.010*** (0.004)
  - 0.012*** (0.004)
  - 0.013*** (0.004)
  - 0.014*** (0.004)
  - Constant examples: 1.017*** (0.084), 0.708*** (0.127), 0.513*** (0.167), 0.475*** (0.147)
- Model diagnostics and sample:
  - Observations: 297 (each specification)
  - Number of states: 27
  - R2 Overall across specs: 0.776, 0.801, 0.842, 0.845, 0.276, 0.407, 0.675, 0.670
  - R2 Within across specs: 0.445, 0.506, 0.607, 0.617, 0.440, 0.490, 0.598, 0.602
  - R2 Between reported variably (table)

- Regressors (as labeled in table):
  - Spatial-price adjusted real Bolsa Familia expenditure per capita, log
  - Share of formal workers in workforce, pct
  - Employment Rate, pct
  - Average Formal Schooling Years for Adults, Bottom Quartile
  - Average Formal Schooling Years for Adults, Top Quartile
  - Federal Personal Income Tax Revenue, share of state GDP
  - Average spatial-price adjusted real HH income per capita, log
  - Share of civil servants in workforce, pct
  - Civil servants' spatial-price adjusted real HH income per capita, log

### Table 6 — Brazil: Logistic Regression: Determinants of Attending Public University, 2014
(Coefficients denote marginal effects; continuous variables calibrated at medians; categorical variables calibrated to zero; robust standard errors in parentheses)
- Data sources: PNAD microdata; IMF staff calculations.
- Selected marginal effects (coefficient (standard error)) and model info:
  - LN of HH income per capita: 0.025*** (0.001) in specification (1); 0.022*** (0.001) in specification (2)
  - Age: 0.001*** (0.000)
  - Male: -0.004*** (0.001)
  - Black: -0.014*** (0.003)
  - Brown: -0.013*** (0.002)
  - Asian: 0.016** (0.007)
  - Native Brazilian: -0.024* (0.013)
  - North Dummy: 0.021*** (0.002)
  - Northeast Dummy: 0.021*** (0.001)
  - South Dummy: -0.000 (0.002)
  - Midwest Dummy: 0.013*** (0.002)
  - Pseudo-R2: 0.0716 (spec 1), 0.0953 (spec 2)
  - Observations: 40,385

*Source: Appendix II. Econometric Results and Tables (Authors' estimates; PNAD microdata; Secretaria da Receita Federal; Ministry of Social Development).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17225.pdf_