Sectoral labor productivity in Brazil is computed as the quotient of Gross Value Added (GVA) to total hours worked for each sector. Annual data on sectoral Gross Value Added (GVA) are drawn from the National Accounts, presented by the Instituto Brasileiro de Geografia e Estatística (IBGE). IBGE reports GVA for twelve sectors: 1) Agriculture, 2) Extractive Industry, 3) Manufacturing, 4) Electricity, Gas, Water, and Sewage, 5) Construction, 6) Commerce, 7) Transportation, Storage, and Mail, 8) Information Services, 9) Financial Services, 10) Real Estate, 11) Other Services, and 12) Public Administration, Health, Education, and Social Security. We combine sectors 2 and 3, sectors 4 and 5, sectors 8 and 9, and sectors 10 and 11 in order to reduce the influence of extreme outliers and to make data trends visually comparable while preserving the conceptual integrity of each sector to the greatest degree possible.¹ The resulting eight aggregated sectors employed throughout this study are: 1) Agriculture, 2) Industry (Manufacturing and Extractive), 3) Construction and Utilities, 4) Commerce, 5) Transportation, Storage, and Distribution, 6) Financial and Information Services, 7) Real Estate and Other Services, and 8) Public Administration, Health, Education, and Social Security.

Data on total hours worked per sector, as well as a range of additional variables employed in our analysis, are drawn from the Pesquisa Nacional por Amostra de Domicílios (PNAD), sponsored by IBGE. Data are pooled over the years 1996 to 2014. Data from Brazil’s Northern region are excluded due to lack of data on rural areas of this region before 2004.² The four-digit economic Activity codes from the PNAD were grouped into the eight sectors described above, and incorporate the methodological changes introduced in the Classificação Nacional de Atividades Econômicas (CNAE) after the year 2000.

Income data from the PNAD were deflated to 2014 constant values using the Índice Nacional de Preços ao Consumidor, provided by IPEA. GVA values were deflated to 2014 constant values using the World Bank’s GDP deflator. Deflated values in Brazilian Reals were then converted to Purchasing Power Parity (PPP) constant 2014 US dollars using data from the IMF’s 2014 World Economic Outlook report.³ Conversion to US dollars was intended to facilitate international comparisons.

Sectoral variation in the productivity-wage relationship may be the result of variation in institutional and market forces among different types of economic activities. Alternatively, it may be the result of differences in average worker characteristics across sectors. To distinguish between these alternative dynamics, we fit a regression model to pooled annual data from the PNAD for years 1996-2012.⁴ The objective is to measure the impact of sectoral labor productivity on individual hourly wages. We limit our sample to working adults between the ages of 15 and 65. The dependent variable is the natural logarithm of the i-th individual’s labor income per hour (w), which is assumed to be a function of the i-th individual’s personal characteristics (vector x), j-th economic sector’s characteristics (z) in the state where that individual is employed, and national minimum wage (v) in year t:

Factor r _i captures unobserved regional heterogeneity and is controlled by fixed effects using 26 binary variables to represent Brazil’s 27 federal units (UF). The factor c _t represents economic, political and/or institutional cycles affecting wages. These cycles are controlled for by 4 binary variables representing five federal governments over the sample period (1996-1998: first term of Fernando Henrique Cardoso; 1999-2002: second term of Fernando Henrique Cardoso; 2003-2006: first term of Luiz Inácio Lula da Silva; 2007-2010: second term of Luiz Inácio Lula da Silva; 2011-2014: first term of Dilma Rousseff). ε ijt is an idiosyncratic error term. Coefficients β, ( and ( express the net impact of the explanatory variable on the log of hourly wage. Regression estimates are computed by Ordinary Least Squares (OLS) with heteroscedasticity consistent standard errors.

This is a hierarchical approach, where the minimum wage v only varies by year (level 1), economic sectors’ characteristics vary by sector/state and year (z, level 2) and individuals’ characteristics vary for each individual i of the sector/state s and year t (x, level 3). In a comparative model, we replace the variable minimum wage and cycles c _t with temporal fixed effects, using 16 binary variables to represent the 17 years included in the sample. The intention is to estimate the model under alternative specifications. The set of individuals’ and sectors’ characteristics used in our analysis is described in Table 1. More detailed descriptive statistics are presented in Appendix A.

Table 1:
Variables and mean values

The main advantage of this pooled hierarchical strategy is that it allows for wider variability of the independent variables. This is particularly important for structural characteristics, which only vary across sectors (e.g. productivity), and/or years (e.g. minimum wage). In turn, the main disadvantage of the pooled models is that variability may differ across years (heteroscedasticity), resulting in biased estimates for the standard errors. This problem was avoided by estimating standard errors robust to heteroscedasticity (Wooldridge, 2003, p.258).

Since explanatory variables in vectors x and z are measured in different scales, we cannot directly compare coefficients in β or 𝜹 in order to identify variables with the largest impact on w. Consequently, standardized coefficients were estimated which permit measurement of explanatory variable effects on w in standard deviation units, rather than in terms of the original units of 𝒙 and 𝒛 (or w). Standard deviation units are the resultant parameters when all variables are standardized to a mean of 0 and a variance of 1, and are estimated by (Wooldridge, 2003):

Where 𝛽 𝑗 is the OLS estimator, 𝜎 𝑗 is the sample standard deviation for the j-th explanatory variable, and 𝜎 𝑤 is the sample standard deviation for the dependent variable ln w.

First, baseline regressions were estimated on all pooled observations using Equation 1. Baseline regressions establish economy-wide coefficient parameters for subsequent comparison with sectoral results. Furthermore, baseline results provide a sensitivity analysis of alternative model specifications (progressively controlling for time (or political-economic cycle) fixed effects and institutional factors). We next estimate Equation 1 for each economic sector individually, drawing upon only those observations that fall into each economic sector, successively. Sectoral results illustrate variations in the productivity-wage relationship between different economic activities.

Brazil successfully managed to bring its hyperinflationary experience to an end in 1994 with the Plano Real under the presidency of Fernando Cardoso. Nevertheless, the side effect of a series of initiatives to control inflation and the growing public deficit (such as privatization, reduction of tariff barriers, and strong valorization of the Real), coupled with simultaneous international crises in Mexico (1994), Asia (1997), and Russia (1998), led to an increase in unemployment, stagnation in productivity, and declines in real wages over this period, as observed in Figure 1 (Dedecca, 2005; Gennari, 2002). Markets balked briefly again in 2002 over the election of Luiz Inácio Lula da Silva of the Workers Party to the presidency, but growth soon returned in full force with the international commodity boom of the 2000s. Fueled primarily by China’s explosive growth, international commodity prices rose sharply for major Brazilian exports such as iron and soybeans, improving Brazil’s terms of trade and attracting foreign investment (Giambiagi, 2011).

Figure 1
Brazil: overall labor productivity and wage growth (1996-2014)
Source: Figure constructed by author from PNAD (1996-2014) and National Accounts

From 2003 onwards, Lula’s government capitalized on this commodity boom by facilitating the expansion of consumer credit and fostering income redistribution through significant real increases in the minimum wage and spending programs like Bolsa Familia. Unemployment shrank, the number of formal salaried jobs multiplied, and income inequality and poverty experienced historic declines (IPEA, 2012). This period was accompanied by a rise in real wages that outstripped growth in labor productivity, as illustrated by the notable rise in real wages relative to productivity in Figure 1. This development suggests that, for the Brazilian economy as a whole, rising real wages over the 2003-2013 period may have been more the result of minimum wage increases and formalization of the labor market than of structural development toward more productive employment. This apparent failure to capitalize on the commodity boom of the 2000s to promote structural productivity gains bodes poorly for prospects of further sustained wage growth in Brazil (Maia; Menezes, 2014).

Finally, the decline in both productivity and wages after 2013, as observed in Figure 1, marks the onset of Brazil’s current economic crisis. The slowing of China’s commodity boom and the crash in oil prices in 2014 reversed Brazil’s terms of trade, leading to a rising deficit that in turn provoked pro-cyclical austerity measures. These recent dynamics have corresponded with declines in real wages and productivity levels.

Examining the Brazilian economy on a sectoral level, the wide divergence in productivity levels among different economic activities becomes clear (Figure 2).⁵ For instance, in 1996 an average hour of labor in the lowest-productivity sector, agriculture, produced US$4.93 in value (in constant 2014 US dollars), while an average hour of labor in the highest productivity sector, financial and information services, produced US$39.67. From 1996 onwards, these sectoral productivity values display a convergent tendency, such that the difference between the lowest productivity sector in 2014 (commerce) and highest (real estate and other services) was only 43% of the difference in 1996.

Figure 2
Brazil: labor productivity (value added per labor hour) by economic sector, absolute and relative growth (1996-2014)
Source: Figures constructed by author from PNAD (1996-2014)

This convergence is the result of two sectoral developments. The first, the surprising decline in productivity for Financial and Information Services, is the combined result of real productivity trends and changing occupational distributions within the sector. In real terms, the financial services subsector suffered a 40% productivity loss between 1998 and 2014. This dynamic could be related to changes in Brazilian banks’ “earnings spread” between lending and borrowing rates, which fell sharply with the decline in interest rates over this period, bringing value-added down with it (Freitas; Prates, 2016; Afonso; Köhler; Freitas, 2009).⁶ Parallel to these real value-added losses in the financial sector, the activity composition of the information services subsector evolved toward lower-productivity activities such as telemarketing and “other business services.” Employment in these lower-productivity components of information services grew by 37% over the 1996-2014 period (relative to overall employment growth of 24%), thus coming to occupy an inflated proportion of total sectoral employment by 2014 (PNAD, 2016).

The second trend leading to sectoral productivity convergence is the dramatic rise in agricultural productivity since 1996. Labor productivity in agriculture (historically Brazil’s lowest productivity sector) grew by 121.7% over the 1996-2014 period, with an annual growth rate of 4.8%. This rapid growth was the result of a number of interrelated dynamics. Rural labor shortages incentivized investments in labor-augmenting technologies, rural workers’ average education levels increased, new transgenic seeds boosted yields, and domestic and international research and extension improved innovation and technology adoption. Overarching these trends, agribusiness assumed an expanded role in capital-intensive, monoculture production for export, especially on the newly opened Cerrado frontier (Buainain, 2014; Guanziroli, 2014; Maia; Sakamoto, 2016; Gasques et al, 2014)

Two additional trends should be noted in Figure 2. Firstly, Real Estate and Other Services’ high levels of “productivity” are misleading, since much of real estate’s value derives from accumulated capital (past labor embodied in property). Thus, when dividing the large value gains enjoyed by real estate during Brazil’s sustained property appreciation by the small number of hours worked in this sector, the resulting labor productivity values appear surprisingly high. These values should be treated with due caution.

Secondly, Figure 2 highlights the ongoing plight of industry in Brazil. Industrial labor productivity in Brazil largely stagnated since 2003, grew at a very modest 2.2% annually over the 1996-2014 period, and actually declined since 2011. This poor performance in productivity was accompanied by a decline in industrial employment as a share of total national employment from 25% in 1985 to just 17% in 2013, as well as by declining industrial value-added as a share of GDP (Jardim and Perin, 2016). These trends suggest that Brazil may be experiencing a process of deindustrialization, as traditional manufacturers confront stiff international competition, unfavorable macroeconomic policies (exchange rate valorization), and the re-articulation of production processes along transnational value chains (Hiratuka; Sarti, 2015). Nevertheless, other sectors of Brazilian industry, such as extractive industries and technology-intensive industries, continued to exhibit dynamic growth until 2013, suggesting that “deindustrialization” only partially describes the continuing evolution of Brazil’s productive structure (Jardim; Perin, 2016).

With the exception of public employment, real wages declined across all sectors throughout the late 1990s and early 2000s. The year 2003 appears to be an inflection point, at which all sectors reverse negative income trends and begin enjoying real income gains, which are sustained until 2013. Over the 1996-2014 period, agriculture exhibits the strongest income gains (37.1% total increase), and financial and information services exhibit the greatest losses (9.3% total decrease) (Figure 3).⁷

Figure 3:
Brazil: monthly labor income by economic sector, absolute and relative growth (1996-2014)
Source: Figures constructed by author from PNAD (1996-2014)

Wages for most sectors roughly follow productivity trends for the corresponding sector, with three noteworthy exceptions. Firstly, despite monotonic declines in productivity in the financial and information services sector, wages for this sector nevertheless pick up after 2003, suggesting a rise in the relative wage (an increasing share of average value-added going to wages). This divergence may be the result of the information asymmetries and bargaining power dynamics described above. Financial and information services jobs are more likely to be career-track, and thus subject to the type of seniority-biased incentive structuring identified as a potential source of wage-productivity divergence by Biesebroeck (2015). Furthermore, jobs in this sector require higher education levels, potentially allowing educated workers (still relatively scarce in the Brazilian labor market) to exert greater bargaining power over employers.⁸ Finally, despite the erosion of union power in the financial sector throughout the 1990s and early 2000s, labor organization in this sector remains high relative to other areas, further improving workers’ bargaining power and driving up relative wage (PNAD, 2016).

In the second incidence of productivity-wage divergence, public employees’ wages realize a total increase of 30.1% between 1996 and 2014, despite total productivity growth of only 6.9% over this period. It is important to note that productivity in the public sector must be interpreted carefully. Because output in the government sector is measured as equal to the total value of inputs, the appreciation of relative wages in the public sector indicates only that salaries constitute a higher share of total costs in the sector, not that public workers are necessarily retaining a greater proportion of their value-added (Boyle, 2006). Alternatively, this appreciation of the relative wage may indeed be a consequence of the significant bargaining power imparted on public employees by institutional protections and benefits, high levels of labor organization, and the relatively high education levels of these workers.

In the third notable instance of productivity-wage divergence, the low wage-levels of workers in the real estate and other services sector illustrates the extent to which value added per hour in this sector, as displayed in Figure 2, is in fact constituted by property rents rather than true labor productivity.

In sum, sectoral labor productivity and real wages both evolved in significant ways over the 1996 to 2014 period. To what degree were these developments related? Did gains in productivity drive gains in wages, as neoclassical labor economics would predict, or were institutional and/or personal worker characteristics more determinative of earnings levels? In order to disentangle these alternative scenarios, estimates of micro-level wage equations are presented in the following section.

We first estimate alternative baseline specifications (Equation 1), and report results in Table 2. All model specifications fit relatively well to the PNAD pooled sample, with coefficients of determination above 46%. Standard errors are robust to heteroscedasticity and estimates are stable and significant over multiple specifications.

Signs and magnitudes of control variables are as expected from the literature. The coefficient of 0.089 for schooling in Model 3 indicates that a one-year increase in an employed individual’s years of schooling, all else equal, is associated with a 9.3% (e^0.089(1=0.092) increase in hourly earnings. The coefficient of (0.274 for female indicates that, all else equal, women’s average hourly wages are 24.0% (e ^(0.274 (1=(0.240) lower than men’s, indicating extensive gender inequalities in the Brazilian labor market. The coefficient of (0.114 for non-white indicates that, all else equal, black and indigenous workers present an average hourly wage that is 10.8% (e ^(0.114 (1=(0.108) lower than white and Asian workers, indicating the presence of strong racial inequalities in employment. The positive coefficient estimates for age and negative estimates for age² indicates an inverted-U relation between wages and working experience⁹.

Table 2:
Baseline regression estimates for the dependent variable ln of hourly wage

The coefficient estimates of productivity, 0.123 in Models 1 and 2, indicate that a 1% increase in each observed worker’s state- and activity-specific labor productivity is associated with a 0.123% increase in real hourly wages for the average worker. This impact increases only slightly when we control for unionization and formal employment in Model 3 (0.166%). While positive and significant, this elasticity appears to be low by international standards (Peeters; Den Reijer, 2011).

Institutional factors also play a significant role in determining wage levels. The coefficient value of minimum wage, 0.251 in Model 2 (0.238 in Model 3), indicates that a 1% increase in the national real minimum wage is associated with a 0.251% increase in real wages (0.238% in Model 3). The coefficient value of 1.019 for union in Model 3 indicates that a 1 percentage point increase in the proportion of organized workers in an observed worker’s state is associated with a 1.019% increase in real wages for that worker, all else equal. Finally, the coefficient of 0.394 for formal indicates that a 1 percentage point increase in the proportion of formalized workers in an observed worker’s state is associated with a 0.394% increase in real wages for that worker.

Estimates for the standardized coefficients allow us to understand to what extent relative change (one standard deviation) in an explanatory variable affects relative variation in the dependent variable. The similar estimates for log of productivity (0.122), unionized (0.098), and formal (0.089) suggest that real wages were equally affected by both relative variations in labor productivity and institutional factors. This finding suggests that remuneration in the Brazilian labor market (both absolutely and relatively between sectors) is not merely determined by market-driven forces such as labor productivity, but also by structural characteristics, many of which are activity-specific. The association between relative changes in the minimum wage and in the hourly wage is slightly lower (0.066), but this result may hide important differences between labor- and capital-intensive sectors. Thus, we next estimate sectoral regressions in order to assess the activity-specific nature of wage-determinants.

Table 3 reports sectoral regression results, generated by fitting the wage-regression model developed above to seven pooled datasets, each of which include only individuals employed in the respective economic sector.¹⁰ Standard errors are robust to heteroscedasticity, and estimates, with some exceptions, are significant at the 1% level. Estimates for the standardized coefficients are presented in Appendix C.

Signs and magnitudes of social control variables are again as expected, though there is notable sectoral variation. Results indicate that returns to schooling are higher than the baseline average in financial and information services (e^0.131(1=14,0% per year of education), and public administration and services (e^0.119(1=12.7%). Returns to formal education are lowest in agriculture (e^0.043(1=4.4%) and real estate and other services (e^0.054(1=5.5%).¹¹ These results indicate that returns to education are higher in more skill-intensive sectors, suggesting the presence of skill-biased technological change and job polarization (increasing separation between high- and low-paying activities) as potential generators of both wage differentiation and persistent inequality among sectors (Berman et al., 1997; Autor, 2010).

Table 3:
Sectoral regression estimates for the dependent variable log of hourly wage

The gender wage-penalty varies significantly by sector. The negative wage impact of being female is greatest in industry (e^(0.351(1=(29.6%) and real estate and other services (e^(0.324(1=(27.6%), and smallest in construction (e^(0.019(1=(1.9%) and agriculture (e^(0.201(1=(18.2%). The effects of race on wages appear to be more robust and consistent over economic sectors. Race-related wage penalties range from a minimum impact of (5.8% for non-whites in agriculture, to a maximum impact of (15.5% for non-whites in Financial & Information Services.

The variable of focus in this study, log of productivity, exhibits significant variation over economic sectors, though its impact is universally positive. All else equal, a 1% increase in the level of labor productivity in an observed worker’s state and economic sector has the largest positive impact on real wages in real estate and other services (+0.34%), commerce (+0.20%), public services (+0.19%), and industry (+0.15%). Productivity has lower impact on wages in financial and information services (+0.10%), construction and utilities (+0.09%), and agriculture (insignificant).

The relatively large positive impacts measured in real estate and commerce may be due to the prevalence of work-on-commission in these sectors. Many workers earn according to their sales performance, and should thus exhibit high levels of association between labor productivity and earnings. The relatively large impact of labor productivity on wages in the public sector is likely the result of labor productivity accounting methods. As noted before, public sector workers’ value-added is calculated directly from inputs, thus eliciting a positive association between productivity and wages. The labor productivity-wage relationship is potentially relatively low in the financial and information services sector due to information asymmetries inherent to these activities. In banking, marketing, research, and other activities in this sector, firms find it difficult to assess employees’ true productivity levels. In contrast, productivity impacts on wages in industry are relatively high due, presumably, to the facility with which firms can measure employees’ productivity and by the degree to which employers regulate employees through employment contracts (Rose; O’Reilly, 1998). Finally, the impacts of labor productivity on wages are smallest in construction and agriculture. This may be due in part to the generally low levels of worker qualification and the substantial influence of institutional variables such as minimum wage in these sectors. It is important to note that agriculture exhibits especially high levels of heterogeneity in firm and worker productivity, causing low sector-level elasticity to obscure noteworthy real productivity and wage gains among modern agricultural practices (Maia; Sakamoto, 2014).

Institutional variables also exert strong and significant effects on real wages at the sectoral level. Results indicate that a 1 percentage point increase in union participation in an observed worker’s state is significantly negatively associated with real wages in agriculture ((0.30%) and commerce ((0.34%), and significantly positively associated with real wages in public services (+0.63%), construction & utilities services (+0.38%) and financial and information services (+0.29%). This divergent behavior may reflect divergent levels of labor organization across sectors. For example, agriculture and commerce are characterized by low levels of labor organization (14% and 15% respectively in 2014), while financial and information services and public services are characterized by high levels of labor organization (25% and 27%, respectively).¹² Furthermore, financial and information services and public services are characterized by higher barriers to entry, high-skilled (difficult to replace) jobs, and institutional protections that facilitate labor organization and action (IPEA, 2009).

As a result, an increase in labor organization in unorganized sectors (agriculture and commerce) may have the effect of pulling resources away from the large majority of non-organized workers, thus resulting in the negative coefficient reported in Table 3. In organized sectors (financial and information services and public services), an increase in labor organization benefits a large enough proportion of workers to pull up sectoral wages as a whole, resulting in the positive coefficient measured for these sectors.

The impacts of formal (measuring the proportion of formalized workers in each state) are significant and positive across all sectors.¹³ Since worker formalization primarily impacts labor markets by requiring firms to pay minimum wage, this variable should exhibit the largest impact on sectors characterized by high proportions of minimum-wage employment, i.e. sectors with low average wages. Indeed, results in Table 3 indicate that a 1 percentage point increase in state-level formalization is associated with the largest wage gains (+0.64%) in agriculture, which presents the lowest wage levels of any economic sector, and the smallest wage gains (+0.21%) in real estate and other services and financial and information services, the latter of which presents the highest wage levels of any economic sector. Despite low average wages in real estate and other services, the net impact of formalization on wages in this sector is likely low due to high within-sector heterogeneity-ranging from extremely high wage activities, such as real estate, to low wage activities, such as cleaning and security services.

The substantial increases of the minimum wage over the sample period primarily favored low-wage sectors: agriculture and real estate and other services. For each 1% variation in the minimum wage, average wages in these sectors increased by 0.50% and 0.46%, respectively. In turn, the impact of minimum wage on high-skill jobs, such as in financial and information services, are negligible.

Overall, the standardized coefficients in Appendix C highlight that labor productivity has played an important role in increasing hourly wages. Relative changes in labor productivity exerted larger effects on average wages in sectors characterized by productivity-based pay, i.e. commerce and real estate and other services. Nevertheless, institutional factors such as minimum wage also significantly explain the dynamics of average wages in Brazil, especially in low-paying sectors. Relative variation in the minimum wage exerted the largest impact on hourly wages in agriculture, real estate and other services, commerce, and construction and other utilities.

ekatovich@wisc.edu gori@unicamp.br