## wp17278

## Source details

**Canonical URL:** [wp17278](https://www.imf.org/-/media/files/publications/wp/2017/wp17278.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17278.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17278.pdf.json)

---

### Linked employer-employee data (RAIS) — dataset role, scope, and use
- RAIS: large administrative linked employer-employee dataset with information on hundreds of millions of job spells between 1988 and 2012.
- Central use: repeatedly estimate an additive worker and firm fixed effects model (AKM) to quantify contributions of firm and worker-specific factors to changes in Brazilian inequality.
- Analytical pipeline:
  - First stage: RAIS used to decompose earnings into worker and firm components.
  - Second stage: estimated worker and firm fixed effects linked to worker demographics and PIA firm financials for manufacturing and mining to study transmission from characteristics into pay.
  - Subpopulation analyses (e.g., large manufacturing and mining firms) use RAIS-linked PIA financials to assess how observable firm characteristics (productivity, size, exporter status) relate to the firm component of pay.

### RAIS and PIA data collection, variables, sample selection, and descriptive statistics
- RAIS features and construction:
  - Mandatory annual survey by the Brazilian Ministry of Labor and Employment (MTE); data collection initiated in 1986 and reached complete coverage of formal establishments in 1994.
  - Reported earnings are gross and include regular salary payments, holiday bonuses, performance-based and commission bonuses, tips, and profit-sharing agreements; they exclude capital income and in-kind transfers.
  - Total annual earnings from an employment relationship divided by job-spell duration to control for labor supply variation.
  - Weekly hours available only from 1994; analysis focuses on adult males age 18–49 to limit labor supply differences.
  - Age binned: less than 18, 18–24, 25–29, 30–39, 40–49, more than 49. Schooling groups: primary (4 years), middle (7 years), high school (12 years), college (12 or more years).
  - Occupation: pre-2003 CBO one-digit; sector: pre-2003 CNAE two-digit; 2003–2012 classifications made consistent via IBGE conversion tables.
  - Firm size = number of full-time equivalent workers = total worker-months / 12.
  - Sample cleaning drops <1 percent of original population; one observation per worker-year retained (highest-paying among longest spells); average number of jobs per year = 1.2.
  - RAIS subperiods used in AKM: 1988–1992, 1992–1996, 1996–2000, 2000–2004, 2004–2008, 2008–2012.
- Key RAIS panel statistics (selected entries, means and standard deviations as in source):
  - Panel A. ALL FORMAL SECTOR WORKERS (RAIS)
    - 1988–1992: # Worker-EarningsAgeSchooling years — 165.54 1.3 1.10 0.86 31.90 11.44 7.65 4.43
    - 1996–2000: 174.64 6.7 1.19 0.84 33.67 11.26 8.61 4.26
    - 2008–2012: 326.57 5.7 0.71 0.71 34.46 11.62 10.81 3.51
  - Panel B. ADULT MALE WORKERS (age 18–49)
    - 1996–2000: 83.92 6.5 1.26 0.83 31.54 8.17 7.94 4.07
    - 2008–2012: 135.03 9.6 0.76 0.68 31.76 8.21 10.24 3.46
  - Panel C. ADULT MALE WORKERS AT LARGE MANUFACTURING AND MINING FIRMS (PIA)
    - 1996–2000: 15.5 5.8 1.54 0.86 31.51 8.05 7.95 4.01
    - 2008–2012: 23.9 9.0 0.96 0.73 31.26 8.08 10.16 3.51
  - Notes: In 1996–2000 prime age adult males had 0.67 years less schooling than overall sample; gap falls to 0.57 years in last subperiod. Adult males earn 4–13 log points more than overall population.
- PIA (firm financials) features:
  - PIA: IBGE-constructed firm financial characteristics for manufacturing and mining, 1996–2012.
  - Mandatory survey for firms with >30 employees or above revenue threshold plus an annual random sample of smaller firms; anonymized unique firm identifier enables linkage to RAIS.
  - Variables: operational and non-operational revenues, costs, investment and capital sales, number of employees, payroll; nominal values converted to real via IBGE CPI.
  - Operational costs = raw materials, intermediate inputs, electricity and utilities; net revenues = gross sales net of returns, cancellations, inventory changes.
  - Value added = net revenues − intermediate inputs; value added per worker = value added / full-time equivalent workers (main productivity measure).
  - Productivity measure does not control for capital intensity due to lack of capital stock data; alternative variants adjusting for worker input quality yield similar results.
  - Sample focuses on relatively larger firms to exploit panel dimension and avoid excessive attrition.
- PIA descriptive statistics (means and standard deviations, as reported):
  - 1996–2000: # Firm-years # Unique firms — 110.5 34.8; Firm size Mean St.d. — 6.41 1.74; Value added p.w. Mean St.d. — 11.15 1.13
  - 2008–2012: 176.8 55.8; Firm size Mean St.d. — 6.69 1.98; Value added p.w. Mean St.d. — 11.30 1.31
  - Between first and last period: number of firms in PIA increased by 60 percent; average log firm size increased by 28 log points; average log real value added per worker grew by 15 log points.

### Macroeconomic and policy background (selected facts)
- Hyperinflation and wage indexation:
  - 1980–1989 yearly inflation averaged 355 percent.
  - 1990–1994 yearly inflation averaged 1,667 percent.
  - Minimum wage indexed to realized inflation; in 1994 Plano Real introduced gradual float, tightened monetary and fiscal policy, lowered inflation below two-digits.
- Trade liberalization and productivity:
  - Tariff and non-tariff trade barriers reduced to one tenth of their 1987 levels upon presidency of Fernando Henrique Cardoso in 1995.
  - Opening to trade over last 25 years frequently cited as major contributor to growth in total factor productivity.
- Social programs and minimum wage policy:
  - Social expenditure as fraction of GDP doubled following Workers’ Party election in 2003; public cash transfer Bolsa Famìlia reached 11 million families in 2006 (~25 percent of population).
  - Education spending increased to 5.5 percent of GDP in 2009 (from 3.5 percent in 2000).
  - Workers’ Party implemented a 20 percent minimum wage increase in 2003 and continued yearly increases averaging over 10 percent during next 10 years.
  - Minimum to median earnings of adult male workers rose from ~34 percent in 1996 to over 60 percent.
  - Engbom and Moser (2017a) argue minimum wage increases can explain a significant fraction of reduction in earnings inequality over 1996–2012.

### Evolution of earnings inequality (RAIS-based findings)
- Variance of log earnings (males age 18–49):
  - Between 1996 and 2012 variance decreased by 28 log points or 39 percent, from 0.73 to 0.44.
- Percentile ratio changes (log percentile ratios, 1996→2012):
  - log 90–50 falls by 26 log points.
  - log 50–10 decreases by 38 log points.
  - log 50–5 decreases by 53 log points.
  - Compression widespread through distribution, reaching as high as 75th percentile; less compression above 75th percentile.
- Decomposition between- vs within-firm:
  - Between-firm inequality decreased by 18 log points or 42 percent from 1988–2012.
  - Within-firm inequality dropped by 11 log points or 36 percent from 1988–2012.
  - Large between-firm pay compressions occurred in all sectors except government and across regions, firm sizes, and productivity terciles (less so at most productive firms).

### AKM decomposition — variances, shares, and trends (Table 4)
- Variance of log earnings by subperiod (with total and percent shares):
  - 1988–1992: 0.75 (100.0%)
  - 1996–2000: 0.69 (100.0%)
  - 2008–2012: 0.47 (100.0%)
  - Change 1996–2012: -0.23 (100.0%)
- Variance of worker effects:
  - 1988–1992: 0.38 (50.8%)
  - 1996–2000: 0.34 (48.8%)
  - 2008–2012: 0.27 (58.4%)
  - Change 1996–2012: -0.07 (29.0%)
- Variance of firm effects:
  - 1988–1992: 0.16 (21.1%)
  - 1996–2000: 0.16 (23.7%)
  - 2008–2012: 0.07 (16.0%)
  - Change 1996–2012: -0.09 (39.6%)
- Covariance and residual:
  - 2×Cov(worker, firm) ranges ~0.13 to 0.09; Change 1996–2012: -0.05 (23.2%)
  - Variance of residual ranges ~0.08 to 0.04; Change 1996–2012: -0.02 (8.7%)
- Sample and fit:
  - # worker years by period: 77.3, 81.5, 132.2 (in millions across selected periods)
  - # firms (millions): 0.96 → 2.16 across periods
  - R2 ranges 0.90–0.93
- Two principal AKM results:
  - Worker heterogeneity is the largest determinant of earnings inequality: worker fixed effects share rises from 49 percent in 1996–2000 to 58 percent in 2008–2012.
  - Firm effects fall more than proportionately: variance of firm effects falls from 16 to seven log points between 1996–2000 and 2008–2012; correlation between worker and firm effects stays around 0.3.

### Link between firm effects and firm characteristics (PIA results)
- PIA subpopulation coverage:
  - PIA subpopulation includes 2.6–2.9 percent of firms but covers 18–19 percent of employment.
  - Variance of log earnings in PIA close to or slightly above population estimate (e.g., 1996–2000: 0.74 (106.9%); 2008–2012: 0.53 (112.6%)).
- Regression findings (value added p.w. and firm size as predictors):
  - 1996–2000: one log point higher value added p.w. pays a given worker ~0.26 log points more (Panel A no controls); interpreted as one standard deviation more productive firm paying over 30 percent more to a given worker.
  - Return to firm size largely explained by correlation with productivity; controlling for value added p.w. reduces firm size effect to marginal.
  - Linear term in average log value added p.w. explains ~half the variation in firm component of pay; subsector controls push R2 > 0.6.
- Decline in pass-through over time (selected coefficients):
  - Value added p.w. coefficient (Panel A Table 6): 1996–2000 ≈ 0.256; 2008–2012 ≈ 0.140.
  - Period-specific value added p.w. coefficients controlling for firm size interactions (Table 7 baseline):
    - 1996–2000: 0.158
    - 2000–2004: 0.124
    - 2004–2008: 0.082
    - 2008–2012: 0.058
  - Value added p.w. × Firm size interaction: 0.012 (Baseline)
  - Worker-years in Table 7 regressions: 77.1 (millions). R2: 0.553 (Baseline), 0.627 (State), 0.648 (Sector).
- Quantitative contribution:
  - Between 1996–2000 and 2008–2012 the variance explained by observable firm characteristics decreases by six log points, accounting for roughly 30 percent of the overall decrease in the variance of log earnings; all of this decrease driven by changes in pass-through from productivity to pay.
- Heterogeneity:
  - Pass-through flattening occurs within firm size quartiles and for exporters and non-exporters; changes not driven solely by firm size distribution shifts.

### Link between worker effects and worker characteristics
- Regression of worker effects on age and education (Table 8):
  - Age (relative to 18–24), examples:
    - Age 25–29: ~0.194 to 0.158 across specifications and periods.
    - Age 30–39: ~0.399 to 0.299.
    - Age 40–49: ~0.533 to 0.436.
  - Education (relative to < middle school), examples:
    - Middle school: ~0.204 to 0.084.
    - High school: ~0.617 to 0.246.
    - College or more: ~1.235 to 1.030.
  - R2 for these regressions: 0.365 to 0.423.
- Contributions to variance decline:
  - Observable worker characteristics jointly explain 37–42 percent of variance in the worker component.
  - Education and age account for roughly 50 percent of decline in dispersion of the worker component between 1996–2000 and 2008–2012, driven entirely by changing returns (not distribution).
- Specific relative pay declines:
  - Workers aged 40–49 saw a decrease relative to 18–24 of eight log points (15 percent) between 1996–2000 and 2008–2012.
  - Middle school, high-school and college graduates saw decreases of five, 19 and nine log points, respectively, relative to those with only primary school over same period.

### Robustness checks and diagnostics supporting AKM
- Mobility and identification:
  - Subperiod mobility statistics (unique workers in millions, average # of jobs, % switchers):
    - 1988–1992: 23.1 ; 1.56 ; 0.37
    - 1996–2000: 25.6 ; 1.45 ; 0.32
    - 2008–2012: 38.6 ; 1.66 ; 0.42
  - Largest connected set covers 97–98 percent of workers in each subperiod.
- Event-study of switchers:
  - Gains from switching up similar to losses from switching down; consistent with AKM specification across subperiods.
- Residual patterns:
  - Some systematic residuals at extremes: lowest decile of worker effects shows positive residuals at lowest-paying firms; highest decile shows opposite; magnitudes modest beyond lowest deciles.
  - Bonhomme et al. (2016) caveat: models lacking log additivity may still satisfy switchers test.
- Additional robustness:
  - Longer subperiods (1988–1996, 1996–2004, 2004–2012) produce consistent results: worker effects explain most of level; firm effects explain disproportionate share of decline.
  - Hourly wages (last four subperiods where hours available) show inequality in hourly wages falls by four log points more than inequality in monthly income; larger fall in variance of worker component; firm component fall in line with earlier findings.
  - Balanced panel of firms and exporter controls yield similar declines in pass-through; exporting firms pay higher premium but difference largely explained by productivity; positive productivity-pay relationship flattens for both exporters and non-exporters.

### Potential explanations and policy-relevant mechanisms behind decreasing inequality
- Main identified mechanism:
  - Collapse in pass-through from worker ability and firm productivity to pay between 1996 and 2012 — producing lower worker skill pay premium and lower firm productivity pay premium.
  - Inequality decline occurred primarily within states and detailed industries and within firm size groups, exporters/non-exporters, and surviving firms.
- Promising candidates discussed:
  - Rapid expansion in educational attainment:
    - High school attainment: 16 percent in 1996 → 49 percent in 2012.
    - College attainment: 8 percent in 1996 → 10 percent in 2012.
    - Decline in returns: high school premium from 0.72 in 1996 to 0.22 in 2012; college premium from 1.32 to 1.23.
    - Caveat: most inequality decrease driven by firm-side forces, so education supply shifts alone may not explain firm pay compression.
  - Changes in bargaining over firm-specific surplus:
    - More centralized bargaining could reduce dependence of pay on firm-specific performance and increase baseline pay.
    - Detailed data on union bargaining agreements recommended for further study.
  - Federal minimum wage increases:
    - Federal minimum wage rose by 119 percent in real terms from 1996 to 2012.
    - Minimum wage is a promising candidate because it cuts across sectors and states and can alter surplus sharing without changing distributions of fundamentals.
    - Challenge: explaining spread of inequality decline across large parts of distribution requires quantifying spillovers (Engbom and Moser (2017a) pursue equilibrium search model with spillovers).
- Evidence against pure compositional explanations:
  - Decline in pass-through observed within narrow groups and surviving firms; measures of worker ability and firm productivity remained positively correlated with pay and did not compress sufficiently to explain observed fall in dispersion.
  - Therefore primary mechanism identified is weaker mapping (lower pass-through) from fundamentals to pay rather than compression in distributions of fundamentals.
- Theory and research implications:
  - Candidate theories should generate strong positive correlation between firm pay differences for identical workers and firm productivity, and explain compression in firm pay differences via weaker productivity→pay link rather than productivity distribution compression.
  - On worker side, successful theory should generate compression primarily via declining returns to ability.
  - Investigate wage-setting changes (minimum wage effects, bargaining centralization) and other labor market institutions; follow-up equilibrium modeling (e.g., Engbom and Moser, 2017a) is promising.

*Source: wp17278 — https://www.imf.org/-/media/files/publications/wp/2017/wp17278.pdf*

### 3.1 Linked employer-employee data (RAIS) .................................................................10

### 3.1 Linked employer-employee data (RAIS) .................................................................10

### Dataset role and scope
- The paper exploits a large administrative linked employer-employee dataset containing information on hundreds of millions of job spells between 1988 and 2012.
- This linked dataset is central to repeatedly estimating an additive worker and firm fixed effects model (AKM) to quantify contributions of firm and worker-specific factors to changes in Brazilian inequality.

### Analytical uses of RAIS in the paper
- RAIS is used in the first-stage decomposition that separates earnings into worker and firm components.
- The estimated worker and firm fixed effects from RAIS are linked, in a second stage, to a rich set of worker demographics and firm financials data for Brazilian manufacturing and mining firms to study transmission from worker and firm characteristics into pay.
- Subpopulation analyses (e.g., larger mining and manufacturing firms) use RAIS-linked firm financials to assess how observable firm characteristics (productivity, size, exporter status) relate to the firm component of pay.

### Key empirical findings tied to the RAIS-linked decomposition
- Firms explained 40 percent of the fall in the variance of log earnings between 1996 and 2012; compression in worker fixed effects explains an additional 29 percent.
- More than half of the decrease in the firm component is accounted for by observable firm characteristics; a weaker pass-through from firm characteristics to pay explains 30 percent of the overall fall in the variance of log earnings over the period.
- Age and education explain close to 40 percent of the variance of the worker component of pay in levels; lower returns to age and education explain 14 percent of the overall fall in the variance of log earnings over the 1996–2012 period.

### Implications for empirical strategy and robustness
- The richness and size of the RAIS-linked employer-employee data enable application of the AKM framework in overlapping subperiods to study changes in earnings components over time.
- RAIS-based estimates are paired with firm financials for robustness checks and to decompose the firm component into parts explained by observable firm characteristics and changing pass-through.

*Source: wp17278 - 3.1 Linked employer-employee data (RAIS) — https://www.imf.org/-/media/files/publications/wp/2017/wp17278.pdf*

### 2011. The overall labor force participation rate has remained roughly stable at 65–70 percent over

### wp17278 - 2011. The overall labor force participation rate has remained roughly stable at 65–70 percent over

### Hyperinflation and wage indexation
- From 1980 to 1989, yearly inflation averaged 355 percent.
- Between 1990 and 1994, yearly inflation averaged 1,667 percent.
- To keep up with inflation, the minimum wage was adjusted first annually and then on a monthly basis proportionately to the previous period’s realized inflation rate.
- In 1994, hyperinflation subsided with the introduction of the Plano Real, which:
  - introduced a gradual float of the local currency;
  - tightened monetary and fiscal policy;
  - lowered inflation below two-digits.

### Trade liberalization and productivity
- Trade liberalization progressed from initially high import tariffs (which had replaced import bans) through late 1980s bills that eliminated selected tariffs and eradicated quantitative import controls.
- Upon becoming president in 1995, Fernando Henrique Cardoso reduced tariff and non-tariff trade barriers to one tenth of their levels in 1987.
- The opening up to trade over the last 25 years is frequently cited as a major contributor to growth in total factor productivity.
- Helpman et al. (2016) argue trade reforms contributed to the rise in earnings inequality in the late 1980s and early 1990s.

### Expansion of social programs and education; minimum wage policy
- Health, education and other social programs began expanding during the late 1990s.
- Following the election of the Workers’ Party in 2003, social expenditure as a fraction of GDP was doubled.
- Although social spending remains less than one percent of GDP, it is portrayed as an important contributor to the reduction in household income inequality.
- The public cash transfer program Bolsa Famìlia reached 11 million families in 2006, comprising nearly 25 percent of the total population.
- Education spending increased to 5.5 percent of GDP in 2009 (compared to 3.5 percent in 2000 and 5.7 percent among G20).
- Brazil had the greatest increase in mathematics PISA scores among 65 countries since 2003.
- The Workers’ Party implemented a 20 percent increase of the minimum wage in 2003 and continued yearly increases averaging over 10 percent during the next 10 years.
- The minimum to median earnings of adult male workers in Brazil increased from around 34 percent in 1996 to over 60 percent (close to the level in France).
- Engbom and Moser (2017a) argue the large increase in the minimum wage can explain a significant fraction of the reduction in earnings inequality in Brazil over the 1996–2012 period.

### Data sources: RAIS (linked employer-employee) — collection, variables, and sample selection
- RAIS is a mandatory annual survey administered by the Brazilian Ministry of Labor and Employment (MTE); data collection initiated in 1986 and reached complete coverage of formal establishments in 1994.
- Fines are levied on late, incomplete, or inaccurate reports; MTE conducts frequent checks, particularly on earnings adherence to minimum wage legislation.
- RAIS contains anonymized, time-invariant person, establishment, and firm identifiers enabling longitudinal linkage.
- Reported earnings are gross and include regular salary payments, holiday bonuses, performance-based and commission bonuses, tips, and profit-sharing agreements; they exclude capital income and in-kind transfers.
- To control for labor supply variation, total earnings from an employment relationship in a year are divided by the duration of the job spell (e.g., seven months implies dividing by seven).
- Weekly hours only exist from 1994; main analysis focuses on adult males to limit unmeasured labor supply differences.
- Age is constructed from modal implied year of birth and coded into bins: less than 18, 18–24, 25–29, 30–39, 40–49, and more than 49 years old.
- Years of schooling are set to modal yearly value and made non-decreasing across years; four education groups defined: primary school (four years), middle school (seven years), high school (12 years), and college (12 or more years).
- Occupation classification uses pre-2003 CBO at one-digit level; sector uses pre-2003 CNAE at two-digit level; 2003–2012 classifications made consistent via IBGE conversion tables.
- Firm size measure is number of full-time equivalent workers during the reference year, computed as total worker-months divided by 12.
- Sample cleaning excludes invalid firm or worker identifiers and observations with missing earnings, dates, education, or age, dropping less than one percent of the original population.
- One observation per worker-year is kept by choosing the highest-paying among all longest employment spells; average number of jobs per year is 1.2 with no trend.
- Analysis restricts to male workers age 18–49 to match literature focusing on prime age males and to limit changing female labor supply and early retirement issues.

### RAIS descriptive statistics (periods and panels)
- Subperiods used in AKM analysis: 1988–1992, 1992–1996, 1996–2000, 2000–2004, 2004–2008, 2008–2012.
- Statistics note: the number of worker-years and number of unique workers are reported in millions; earnings are in log multiples of the current monthly minimum wage; schooling in years; age is average age of bin. Means computed by period; standard deviations demeaned by year then pooled.

Panel A. ALL FORMAL SECTOR WORKERS (RAIS)
- 1988–1992: # Worker-EarningsAgeSchooling years — 165.54 1.3 1.10 0.86 31.90 11.44 7.65 4.43
- 1992–1996: 162.14 3.0 1.18 0.86 33.17 11.29 8.09 4.40
- 1996–2000: 174.64 6.7 1.19 0.84 33.67 11.26 8.61 4.26
- 2000–2004: 202.75 2.6 1.00 0.80 34.01 11.32 9.50 4.05
- 2004–2008: 254.26 2.4 0.81 0.74 34.21 11.45 10.27 3.78
- 2008–2012: 326.57 5.7 0.71 0.71 34.46 11.62 10.81 3.51

Panel B. ADULT MALE WORKERS (age 18–49)
- 1988–1992: 78.42 3.5 1.23 0.86 30.80 8.05 7.17 4.25
- 1992–1996: 78.92 4.3 1.28 0.86 31.42 8.10 7.50 4.20
- 1996–2000: 83.92 6.5 1.26 0.83 31.54 8.17 7.94 4.07
- 2000–2004: 94.72 9.8 1.04 0.78 31.58 8.22 8.81 3.89
- 2004–2008: 112.93 4.0 0.85 0.72 31.65 8.24 9.60 3.67
- 2008–2012: 135.03 9.6 0.76 0.68 31.76 8.21 10.24 3.46

Panel C. ADULT MALE WORKERS AT LARGE MANUFACTURING AND MINING FIRMS (PIA)
- 1996–2000: 15.5 5.8 1.54 0.86 31.51 8.05 7.95 4.01
- 2000–2004: 16.7 6.3 1.28 0.83 31.30 8.13 8.87 3.84
- 2004–2008: 21.0 7.8 1.07 0.77 31.12 8.14 9.52 3.69
- 2008–2012: 23.9 9.0 0.96 0.73 31.26 8.08 10.16 3.51

- Notes from table: In the 1996–2000 subperiod, prime age adult males have 0.67 years less schooling than the overall sample; this gap falls to 0.57 years in the last subperiod.
- Adult males earn 4–13 log points more than the overall population; variance of log earnings slightly higher for adult males.
- PIA subpopulation adult males earned on average 28 log points more than all adult males in 1996–2000; this decreased to 20 log points in 2008–2012.
- PIA adult males display a 3 log point and 5 log point higher standard deviation of log earnings in 1996–2000 and 2008–2012, respectively.

### Data sources: PIA (firm characteristics) — collection, variables, and sample selection
- PIA contains firm financial characteristics from 1996–2012, constructed by IBGE from annual firm surveys in manufacturing and mining.
- The survey is mandatory for firms with more than 30 employees or above a revenue threshold and for an annual random sample of smaller firms; non-compliance is subject to a fine.
- Each firm has an anonymized unique identifier enabling linkage to RAIS.
- PIA variables include operational and non-operational revenues, costs, investment and capital sales, number of employees and payroll; nominal values converted to real using the IBGE CPI.
- Operational costs defined as cost of raw materials, intermediate inputs, electricity and other utilities; net revenues as gross sales net of returns, cancellations, and inventory changes.
- Value added = net revenues − intermediate inputs; value added per worker = value added / full-time equivalent workers. This is the paper’s main measure of firm productivity.
- The productivity measure does not control for capital intensity due to lack of direct capital stock data; alternative variants adjusting for worker input quality yield similar results.
- Sample selection focuses on the deterministic set of relatively larger firms to exploit panel dimension and avoid excessive attrition.

### PIA descriptive statistics (1996–2012 subperiods)
- Subperiods: 1996–2000, 2000–2004, 2004–2008, 2008–2012.
- Notes: number of firm-years and unique firms reported in thousands; firm size is log number of full-time equivalent employees; value added per worker is log of real value added per worker; means and standard deviations weighted by number of full-time employees.

Table 2 summary (means and standard deviations)
- 1996–2000: # Firm-years # Unique firms — 110.5 34.8; Firm size Mean St.d. — 6.41 1.74; Value added p.w. Mean St.d. — 11.15 1.13
- 2000–2004: 130.6 40.9; Firm size Mean St.d. — 6.35 1.79; Value added p.w. Mean St.d. — 11.19 1.32
- 2004–2008: 156.5 48.8; Firm size Mean St.d. — 6.57 1.90; Value added p.w. Mean St.d. — 11.22 1.34
- 2008–2012: 176.8 55.8; Firm size Mean St.d. — 6.69 1.98; Value added p.w. Mean St.d. — 11.30 1.31

- Between the first and last period, the number of firms in PIA increased by 60 percent.
- The average log firm size increased by 28 log points and average log real value added per worker grew by 15 log points.
- The standard deviation of firm size monotonically increases; the standard deviation of value added per worker first increases then falls in the last subperiod.
- All firm-level results are weighted by the number of full-time equivalent workers.

### Inequality trends and role of firms (overview)
- The paper provides an overview of Brazil’s rapid decrease in earnings inequality using the sample of prime age males (age 18–49).
- The decrease in inequality occurred throughout a large part of the earnings distribution.
- The authors provide suggestive evidence that firms are an important source of inequality and a factor behind the decrease in inequality in Brazil.

*Source: RAIS and PIA data descriptions and summary statistics as presented in the provided content unit.*

### 4.1    The evolution of earnings inequality

### 4.1    The evolution of earnings inequality

### Evolution of variance of log earnings (1988–2012)
- Between 1996 and 2012, the variance of log earnings in Brazil decreased by 28 log points or 39 percent, from 0.73 to 0.44.
- Comparison note: the US experienced an increase in the variance of log annual earnings of 30 log points from 1967–2005 (Heathcote et al., 2010).
- Figure specification note: Statistics computed for males of age 18–49. Source: RAIS.

### Compression across the earnings distribution (percentile ratios)
- Percentile ratios are plotted as log percentile ratios normalized to zero in 1996.
- Two main empirical facts:
  - The compression was widespread throughout the earnings distribution, reaching as high as the 75th percentile.
  - The amount of compression gradually decreases as one moves further up in the distribution.
- Exact changes in log percentile ratios from 1996 to 2012:
  - log 90–50 percentile ratio falls by 26 log points.
  - log 50–10 percentile ratio decreases by 38 log points.
  - log 50–5 percentile ratio decreases by 53 log points.
- Observed pattern at the top:
  - Less compression above the 75th percentile.
  - Some divergence among the top 10 percent, with the log 95–50 percentile ratio falling slightly less than the log 90–50 ratio.
- Figure specification note: Statistics computed for males of age 18–49. Source: RAIS.
- Appendix B: shows there was wide-spread real earnings growth over this period.

### Role of worker observables (education, demographics)
- The inequality decrease was not driven by the rapid educational attainment expansion and demographic changes during this period.
- Within a Mincer framework (Appendix A), most of the decrease is orthogonal to worker observables.
- This weak explanatory power from worker-side observables motivates studying the role of firms.

### Decomposition: between-firm versus within-firm variance
- Decomposition identity (as in Fortin et al., 2011; Song et al., 2016):
  - Var(y_ijt) (overall) = Var(y_jt) (between firms) + Var(y_ijt | i ∈ j) (within firms)
  - Where y_ijt denotes log earnings of worker i employed by firm j in year t, and ȳ_jt denotes average log earnings in firm j in year t.
- Empirical findings (Brazil, 1988–2012):
  - There is significant variability in earnings within firms, but an even greater amount of earnings inequality between firms.
  - Between-firm inequality decreased by 18 log points or 42 percent from 1988–2012.
  - Within-firm inequality dropped by 11 log points or 36 percent from 1988–2012.
- Interpretation caveat:
  - Raw decompositions cannot distinguish whether firm differences reflect firm pay policies or recruitment of systematically higher-paid workers; panel methods are used subsequently to identify firm pay policies.

### Sectoral patterns in between- and within-firm variance
- Large between-firm pay compressions occurred in all Brazilian sectors except government.
- Figure specification note: Statistics computed for males of age 18–49. Source: RAIS.
- Appendix C: shows similar patterns obtain in every region of Brazil, among small and large firms, and among unproductive and productive firms—though less so at the most productive firms.

### Empirical approach preview (motivation for panel fixed effects)
- To identify the importance of firm pay policies, the authors estimate high-dimensional fixed effects econometric models controlling for unobserved worker and firm heterogeneity, using RAIS linked employer-employee data.
- Estimations are performed separately in six subperiods covering:
  - 1988–1992, 1992–1996, 1996–2000, 2000–2004, 2004–2008, and 2008–2012.
- First-stage specification (two-way fixed effects):
  - log y_it = α_i + α_{J(i,t)} + Υ_t + ε_it
  - Assumption: E[ε_it | α_i, α_{J(i,t)}, Υ_t] = 0 (exogenous mobility).
- Identification relies on worker mobility across firms; summary of mobility statistics (Table 3):
  - # Unique workers (millions) and related mobility measures by subperiod:
    - 1988–1992: 23.1 ; Average # of jobs 1.56 ; % switchers 0.37
    - 1992–1996: 23.7 ; Average # of jobs 1.48 ; % switchers 0.33
    - 1996–2000: 25.6 ; Average # of jobs 1.45 ; % switchers 0.32
    - 2000–2004: 28.8 ; Average # of jobs 1.46 ; % switchers 0.32
    - 2004–2008: 33.0 ; Average # of jobs 1.54 ; % switchers 0.36
    - 2008–2012: 38.6 ; Average # of jobs 1.66 ; % switchers 0.42
  - Note: Number of unique workers in millions. A switcher is defined as a worker who is associated with two or more employers during the period. Source: RAIS.
- Coverage and connectedness:
  - The largest connected set of firms and workers in each subperiod covers 97–98 percent of all workers in each subperiod (Table 13, Appendix F).
- Tests and assumptions:
  - The authors investigate exogenous mobility by examining gains/losses for workers switching between firm-effect quartiles and the distribution of average residuals across worker and firm effects.
  - Variance decomposition used in diagnostics:
    - Var(log y_it) = Var(a_i) + Var(a_{J(i,t)}) + Var(Y_t) + 2Cov(a_i, a_{J(i,t)}) + 2Cov(a_i, Y_t) + 2Cov(a_{J(i,t)}, Y_t) + Var(e_it)
  - Sampling error in estimated effects may overstate variances and induce negative bias in covariances; the authors assume such error is constant over time so changes over time remain valid.

*Source: wp17278 - 4.1    The evolution of earnings inequality (RAIS-based analysis).*

### 6.1    AKM decomposition

### 6.1 AKM decomposition

### Variance decomposition and trends
- Table 4 reports AKM variance decomposition for six five-year subperiods between 1988 and 2012. Results are weighted by worker-years.
- Key series depicted in Figure 5: variance of raw earnings, variance of estimated worker effects, variance of firm effects. Variance of year effects and covariance terms are small and excluded from the figure.
- Variance of log earnings by period (Table 4):
  - 1988–1992: 0.75 (100.0%)
  - 1992–1996: 0.75 (100.0%)
  - 1996–2000: 0.69 (100.0%)
  - 2000–2004: 0.62 (100.0%)
  - 2004–2008: 0.53 (100.0%)
  - 2008–2012: 0.47 (100.0%)
  - Change 1996–2012: -0.23 (100.0%)
- Variance of worker effects (Table 4):
  - 1988–1992: 0.38 (50.8%)
  - 1992–1996: 0.37 (48.9%)
  - 1996–2000: 0.34 (48.8%)
  - 2000–2004: 0.33 (52.6%)
  - 2004–2008: 0.30 (56.4%)
  - 2008–2012: 0.27 (58.4%)
  - Change 1996–2012: -0.07 (29.0%)
- Variance of firm effects (Table 4):
  - 1988–1992: 0.16 (21.1%)
  - 1992–1996: 0.17 (23.3%)
  - 1996–2000: 0.16 (23.7%)
  - 2000–2004: 0.13 (20.4%)
  - 2004–2008: 0.09 (17.4%)
  - 2008–2012: 0.07 (16.0%)
  - Change 1996–2012: -0.09 (39.6%)
- Other decomposition elements (Table 4):
  - Variance of year effects: ranges 0.02 (2.2%) to 0.00 (0.1%); Change 1996–2012: 0.00 (-0.3%)
  - 2×Cov. worker and firm effects: ranges 0.13 (17.3%) to 0.09 (18.4%); Change 1996–2012: -0.05 (23.2%)
  - Variance of residual: ranges 0.08 (10.1%) to 0.04 (7.6%); Change 1996–2012: -0.02 (8.7%)
- Sample size and fit (Table 4):
  - # worker years: 77.3, 77.3, 81.5, 91.8, 110.0, 132.2 (by period)
  - # firms (in millions): 0.96, 1.02, 1.21, 1.41, 1.71, 2.16
  - R2: 0.90, 0.90, 0.92, 0.92, 0.93, 0.92
- Two principal results:
  - Worker heterogeneity is the single most important determinant of earnings inequality: worker fixed effects share rises from 49 percent in 1996–2000 to 58 percent in 2008–2012 (Table 4 shares).
  - Firm effects fall more than proportionately over time: between 1996–2000 and 2008–2012, firm effects variance falls from 16 to seven log points while person effects fall from 34 to 27 log points. The correlation between worker and firm effects stays around 0.3, so the covariance term falls in line with standard deviations.

---

### 6.2 The link between firm effects and firm characteristics

### Comparison: manufacturing & mining subpopulation (PIA) versus overall population
- PIA coverage and comparison (Table 5):
  - The PIA subpopulation includes 2.6–2.9 percent of firms but covers 18–19 percent of employment.
  - Variance of log earnings (% of population estimate) in PIA:
    - 1996–2000: 0.74 (106.9%)
    - 2000–2004: 0.70 (113.0%)
    - 2004–2008: 0.60 (114.2%)
    - 2008–2012: 0.53 (112.6%)
    - Change 1996–2012: -0.22 (95.1%)
  - Variance of worker effects (% of pop. estimate) in PIA:
    - 1996–2000: 0.37 (109.8%)
    - 2008–2012: 0.32 (116.6%)
    - Change 1996–2012: -0.05 (81.3%)
  - Variance of firm effects (% of pop. estimate) in PIA:
    - 1996–2000: 0.14 (86.8%)
    - 2008–2012: 0.06 (81.4%)
    - Change 1996–2012: -0.08 (91.2%)
- Conclusion: inequality trends are similar in the PIA subpopulation compared to the overall population.

### Regression of firm effects on firm characteristics (Table 6 and Table 7)
- Main predictors: average log value added per worker (value added p.w.) and average log firm size. All regressions weighted by worker-years.
- Key findings (Table 6, Panel A–C summaries):
  - A one log point more productive firm pays a given worker 0.26 log points more in 1996–2000 (Panel A, no controls). This is interpreted as a one standard deviation more productive firm paying over 30 percent more to a given worker.
  - The return to firm size is largely explained by its correlation with productivity: controlling for value added per worker, firm size pays only marginally better.
  - A linear term in average log value added per worker explains around half of the variation in the firm component of pay. With subsector controls, R2 exceeds 0.6.
- Decline in pass-through over time (Table 6, Table 7):
  - Pass-through from value added p.w. to firm pay falls substantially between 1996–2000 and 2008–2012.
  - 1996–2000: value added p.w. coefficient ≈ 0.256 (Panel A, Table 6); 2008–2012: value added p.w. coefficient ≈ 0.140 (Panel A, Table 6). (See Panel-specific entries for exact period-by-period coefficients.)
  - Table 7 (controlling for differential pass-through by firm size) period-specific value added p.w. coefficients:
    - 1996–2000: 0.158 (Baseline), 0.139 (State), 0.142 (Sector)
    - 2000–2004: 0.124, 0.116, 0.106
    - 2004–2008: 0.082, 0.075, 0.062
    - 2008–2012: 0.058, 0.053, 0.040
  - Value added p.w. × Firm size interaction: 0.012 (Baseline), 0.009 (State), 0.010 (Sector) (Table 7)
  - Worker-years in Table 7 regressions: 77.1 (in millions). R2: 0.553 (Baseline), 0.627 (State), 0.648 (Sector).
- Quantitative contribution to inequality decline (Figure 7):
  - Between 1996–2000 and 2008–2012, the variance explained by observable firm characteristics decreases by six log points, accounting for roughly 30 percent of the overall decrease in the variance of log earnings over this period. All of this decrease is driven by changes in the pass-through from productivity to pay.
- Heterogeneity by firm size:
  - Changes in pass-through are not driven solely by shifts in the firm size distribution; conditional on firm size the estimated fall in pass-through remains comparable (Table 7).
  - Figure 8 shows the relationship between AKM firm effects and value added p.w. by firm size quartile: the relationship flattens over time across quartiles.

---

### 6.3 The link between worker effects and worker characteristics

### Regression of worker effects on age and education (Table 8)
- Dependent variable: estimated worker effect a_i. Education estimates are relative to less than middle school (< 7 years). Age estimates are relative to age 18–24. Number of worker-years in millions. Weighted by worker-years.
- Estimated age coefficients (relative to 18–24), examples from Table 8:
  - Age 25–29: ranges ~0.194 to 0.158 across periods and with/without state controls.
  - Age 30–39: ranges ~0.399 to 0.299 across periods and with/without state controls.
  - Age 40–49: ranges ~0.533 to 0.436 across periods and with/without state controls.
- Estimated education coefficients (relative to < middle school), examples from Table 8:
  - Middle school: ranges ~0.204 to 0.084 across periods and with/without state controls.
  - High school: ranges ~0.617 to 0.246 across periods and with/without state controls.
  - College or more: ranges ~1.235 to 1.030 across periods and with/without state controls.
- R2 values for regressions of worker effects on observables:
  - R2 ranges from 0.365 to 0.423 (Table 8 depending on period and state controls).
- Contributions to variance decline (Figure 9):
  - Observable worker characteristics jointly explain 37–42 percent of the variance in the worker component of pay.
  - Education and age account for roughly 50 percent of the decline in dispersion in the worker component between 1996–2000 and 2008–2012, and this contribution is driven entirely by changes in returns (not distribution) according to the counterfactuals presented.
- Specific declines in relative pay (text):
  - Workers aged 40–49 experienced a decrease in pay relative to those aged 18–24 of eight log points (or 15 percent) between 1996–2000 and 2008–2012.
  - Middle school, high-school and college graduates saw decreases of five, 19 and nine log points, respectively, relative to those with only primary school over the same period.

---

### 6.4 Support for the AKM model

### Robustness checks and residual patterns
- Two robustness exercises following Card et al. (2013) were replicated using RAIS data.
- Event-study of switchers (Figure 10):
  - Figure 10 shows average earnings of workers who switch firms up to two years prior and two years after the switch for the latest subperiod (1996–2000 and 2008–2012). Switchers are classified by firm-effect quartile of pre- and post-transition employers.
  - Gains from switching up are similar to losses from switching down, consistent with the AKM specification. Similar results hold in other subperiods and across other quartiles.
- Residual structure (Figure 11):
  - Figure 11 shows average estimated residual by decile of worker and firm effects for 1996–2000 and 2008–2012.
  - Some evidence of misspecification at extremes:
    - Lowest decile of worker effects displays systematically positive residuals at lowest-paying firms; highest decile shows opposite pattern.
    - Lowest firm-effect decile shows systematic positive mean residuals for the lowest-paid workers, and vice versa for highest-paid workers.
  - Magnitudes of these errors are modest; beyond the lowest deciles errors do not exhibit a strong systematic relationship.
- Notes:
  - Bonhomme et al. (2016) caution that models lacking log additivity may still satisfy the switchers test; interpret event-study results with that qualification.
  - Appendix E (referenced) investigates exporters vs non-exporters and finds both groups experienced a similar fall in the pass-through.

*Source: RAIS and PIA results as presented in "6.1 AKM decomposition" from wp17278 - 6.1 AKM decomposition.*

### 6.5    Potential explanations behind decreasing inequality in Brazil

### 6.5    Potential explanations behind decreasing inequality in Brazil

### Key empirical findings
- Changes in the pass-through from worker and firm characteristics to pay was a major driver behind Brazil’s inequality decrease from 1996 to 2012.
- The pass-through from worker ability and firm productivity to pay "collapsed", producing:
  - a lower worker skill pay premium, and
  - a lower firm productivity pay premium.
- The overwhelming majority of the inequality decrease took place within states and detailed industry classifications.
- The pass-through from productivity to pay fell:
  - within firm size groups,
  - for exporters and non-exporters, and
  - for a balanced panel of surviving firms.
- Worker observables jointly explain about 45 percent of the overall variance in log earnings.
- Between 1996–2012 the variance of log labor earnings decreased by 28 log points (as reported in the conclusion).

### Promising explanations discussed
- Rapid expansion in educational attainment
  - By increasing the labor supply of higher education groups, this may have contributed towards falling education premiums.
  - Caveat noted: most of Brazil’s inequality decrease was due to forces on the firm side, so it is less clear why changes in the market for education would affect firm-specific pay components.
- Changes in bargaining over firm-specific surplus
  - If the bargaining process became more centralized over this period, earnings could become less dependent on firm-specific performance and more on regional, industry or aggregate negotiation outcomes.
  - Higher baseline pay and lower dependence on firm-specific (worker-specific) performance could explain rising real earnings and a lower pass-through from firm productivity (worker ability) to pay.
  - The authors note the potential value of using detailed data on union bargaining agreements to study this mechanism further.
- Federal minimum wage increases
  - Brazil’s federal minimum wage rose by 119 percent in real terms from 1996 to 2012.
  - Because minimum wage legislation cuts across sectors and states, it is a promising candidate to explain the decrease in inequality by affecting how firm-specific surplus is split without directly changing the distribution of worker and firm characteristics.
  - Challenge: explaining why inequality fell throughout large parts of the earnings distribution; spillover effects would need to be quantified (Engbom and Moser (2017a) pursue this via an equilibrium search model estimating spillovers).

### Evidence arguing against compositional explanations
- The decline in pass-through occurred within narrow groups (within states, within detailed industries, within firm size groups, exporters/non-exporters, and surviving firms), suggesting a story that goes beyond direct compositional change in observable worker or firm fundamentals.
- Measures of worker ability and firm productivity remained positively correlated with pay throughout the period; the distribution of those fundamentals did not compress sufficiently to explain the observed fall in earnings dispersion.
- Therefore, the primary mechanism identified is weaker mapping (lower pass-through) from fundamentals to pay rather than compression in the underlying distributions of fundamentals.

### Implications for theory and future research
- Any candidate theory of Brazil’s inequality decrease should:
  - generate pay differences between firms for identical workers that are strongly positively correlated with firm productivity, and
  - generate a compression in firm pay differences over time not through compression in the distribution of firm productivity, but through a weaker link between productivity and pay.
- On the worker side, a successful explanation should generate compression primarily through declining returns to ability and not through compression in the underlying ability distribution.
- Investigating changes in the nature of wage setting (including minimum wage effects and bargaining centralization) and other labor market institutions is a promising avenue for future research.
- The authors highlight follow-up work (Engbom and Moser, 2017a) that develops an equilibrium search model and estimates minimum wage spillover effects consistent with the stylized facts.

*Source: IMF working paper section 6.5, "Potential explanations behind decreasing inequality in Brazil."*

### Section 5 suggests that worker observables explain 37–40 percent of the variance ofthe worker com-

### wp17278 - Section 5 suggests that worker observables explain 37–40 percent of the variance ofthe worker com-

### Worker observables and variance decomposition (Mincer vs AKM)
- Worker observables explain 37–40 percent of the variance of the worker component of earnings (which constitutes 50–60 percent of the overall variance).
- Difference between Mincer and AKM frameworks:
  - Mincer: allows observables and their returns to vary by year.
  - AKM: uses averages over the subperiod.
  - Conceptual implication: part of the return to education and experience is due to a selection into better paying firms.
- Fraction of earnings dispersion explained by worker observables does not change much over time.
- Worker observables and sector controls explain close to 23 percent of the fall in inequality.
- Decomposition highlights:
  - Age and education account for roughly 20 percent of the variance of log earnings and six percent of the decline.
  - Inequality between occupations remains relatively constant around six to eight percent of total variance.
  - Fraction of total inequality explained by differences in means across sectors falls from seven to three percent — this accounts for 15 percent of the overall decline in variance.
  - Covariances between the explanatory variables remain stable around 12–14 of total variance.
- Conclusion: Even when controlling for detailed worker characteristics, more than half of the level of inequality as well as its decline is residual in nature.

### Education attainment, returns, and inequality decline
- Workforce attainment changes (1996 → 2012):
  - High school attainment: from 16 percent in 1996 to 49 percent in 2012.
  - College attainment: from eight percent in 1996 to ten percent in 2012.
- Changes in premiums relative to less than middle school (log points):
  - High school premium: from 0.72 in 1996 to 0.22 in 2012.
  - College premium: from 1.32 in 1996 to 1.23 in 2012.
- Net effect: The push towards higher educational levels, particularly among initially lowest education groups, was largely offset by a decline in the returns to education.
- Consistency: This is consistent with findings of a lower gradient of worker effects in worker observables, including education (noted in Section 6).

### Earnings growth decomposition — individual vs firm average earnings (percentiles)
- Method: Linked employer-employee RAIS data to quantify role of changes in firm average pay for earnings growth across income groups.
- Periods analyzed: 1988–1996 (mild inequality increase) and 1996–2012 (rapid inequality decline).
- Key empirical features:
  - Real labor earnings growth occurred at almost every percentile in both periods.
  - Individual earnings growth:
    - Slopes mildly upward in initial earnings percentiles from 1988–1996 (earnings inequality increased a little).
    - Slopes pronouncedly downward from 1996–2012 (earnings inequality decreased significantly).
  - Individual earnings growth throughout the distribution is almost entirely explained by the growth of firm average earnings among those groups.
  - There were no significant changes in within-firm earnings inequality.
- Interpretation: Suggestive evidence that firms played an important role in explaining earnings inequality dynamics during both 1988–1996 and 1996–2012.

### Variance between and within firms by region, firm size, and productivity
- Findings common across dimensions:
  - Compression in pay across firms is present across all Brazilian regions, firm size terciles, and productivity terciles.
  - Dominance of between-firm compression is present in every region.
- Regions: North, Northeast, Southeast, South, Central−East all display substantial compression of pay over time; between-firm compression dominates.
- Firm size terciles:
  - Substantial compression across firms within small, medium, and large firms.
- Productivity terciles:
  - Substantial compression across firms within low-productivity and medium-productivity firms.
  - The decline is less pronounced among high productivity firms.
- Data note: Statistics computed for males of age 18–49. Source: RAIS and PIA.

### Robustness of AKM results
- Longer subperiods (nine-year: 1988–1996, 1996–2004, 2004–2012) — Table 9:
  - Variance of log earnings: 0.77 (100.0%), 0.70 (100.0%), 0.54 (100.0%); Change 1996–2012: -0.23 (100.0%).
  - Variance of worker effects: 0.35 (45.9%), 0.32 (46.0%), 0.29 (53.6%); Change 1996–2012: -0.06 (27.6%).
  - Variance of firm effects: 0.170 (21.9%), 0.15 (21.7%), 0.09 (16.3%); Change 1996–2012: -0.08 (35.3%).
  - Variance of year effects: 0.03 (3.3%), 0.00 (0.2%), 0.00 (0.2%); Change 1996–2012: -0.03 (10.9%).
  - 2×Cov. worker and firm: 0.15 (19.7%), 0.15 (21.2%), 0.11 (20.5%); Change 1996–2012: -0.04 (17.7%).
  - Variance of residual: 0.10 (12.8%), 0.07 (10.3%), 0.05 (9.6%); Change 1996–2012: -0.05 (20.6%).
  - # worker-years: 155.6, 175.9, 249.4.
  - # firms: 1.45, 1.93, 2.80.
  - R2: 0.87, 0.90, 0.90.
- Conclusion from longer subperiods: Worker effects explain a few percentage points less of overall variation (offset by larger residual variance); share from firm effects remains roughly the same. Main conclusion holds: worker effects explain most of the level of inequality while firm effects explain a disproportionate share of the decline.

### Hourly wages robustness (last four subperiods)
- Data limitation: No hours prior to 1994; contracted weekly hours available after 1994. Top 75 percent report 44 hours a week (conditional on working) for prime age males.
- Table 10 (hourly wage rate, 1996–2000, 2000–2004, 2004–2008, 2008–2012; Change 1996–2012):
  - Variance of log wages: 0.73 (101.4%), 0.65 (99.2%), 0.55 (96.1%), 0.49 (93.8%); Change: -0.24 (120.9%).
  - Variance of worker effects: 0.34 (94.8%), 0.33 (91.8%), 0.29 (88.4%), 0.27 (86.9%); Change: -0.07 (145.5%).
  - Variance of firm effects: 0.18 (108.3%), 0.14 (110.4%), 0.11 (111.7%), 0.09 (111.0%); Change: -0.09 (105.9%).
  - Variance of residual: 0.05 (99.0%), 0.05 (98.1%), 0.04 (97.0%), 0.03 (96.9%); Change: -0.02 (102.8%).
  - # worker-years: 80.1, 90.2, 108.2, 130.0.
  - # firms: 1.20, 1.40, 1.70, 2.15.
  - R2: 0.93 across subperiods.
- Interpretation: Inequality in hourly wages falls by four log points more than inequality in monthly income. This is primarily due to a larger fall in the variance of the worker component of pay; the fall in the firm component is in line with earlier findings. Main conclusion remains: compression in the firm component contributed most to the overall fall.

### Robustness of productivity-pay relationship and exporter status
- Balanced panel of firms (present in PIA in all four periods 1996–2012):
  - Results are in line with earlier Table 6.
  - Slightly steeper relationship between value added per worker and firm effects in the balanced panel.
  - Similar decline in magnitude of the coefficient over time across specifications.
- Exporting status regressions (Table 12):
  - Positive relationship between value added per worker and firm effects remains after controlling for exporting variables; coefficient declines over time.
  - Exporting firms pay a higher premium than non-exporting firms, but the premium among exporting firms decreases with export intensity.
  - Difference between exporting and non-exporting firms decreases from 13–25 log points to 2–7 log points once controlling for productivity differences.
  - Positive relationship between value added per worker and firm effects is present among both exporting and non-exporting firms and has flattened for both groups.
- Export intensity definition: revenues from exports divided by total revenues.
- Implication: Patterns motivate focus on firm productivity dimension as a key driver.

### Largest connected set representativeness
- Largest connected set covers 97–98 percent of the overall population; statistics are very similar to the overall adult male population.
- Table 13 (summary statistics relative to adult males; values reported as absolute with proportions in parentheses):
  - Worker-years and # Workers reported across subperiods with proportions ~96.9–98.6 percent.
  - Example entries (selected):
    - 1988–1992: Earnings 77.3 (98.6%), Age 23.1 (98.2%), Schooling 1.24 (100.9%), Mean age 30.82 (100.1%), Mean schooling 8.04 (100.0%).
    - 2008–2012: Earnings 132.2 (97.9%), Age 38.6 (97.5%), Schooling 0.77 (101.0%), Mean age 31.74 (99.9%), Mean schooling 8.20 (99.9%).
- Note: Earnings are in log multiples of the monthly minimum wage; schooling is years of education; age is assigned as the average of a worker’s age bin. Source: RAIS.

*Source: wp17278 - Section 5 suggests that worker observables explain 37–40 percent of the variance ofthe worker com- (RAIS and PIA data as presented in the provided content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17278.pdf_
