The Reform Was Selective: How Brazil Created a Natural Comparison
This chapter examines the institutional design and research comparison behind Brazil’s Plano Brasil Maior payroll-tax reform. The practical question is whether a reported employment increase reflects the reform’s change in labor costs rather than a difference that was already unfolding across sectors. Understanding who was treated, who was not, and how the timing was used will let readers interpret the later employment, wage, and profit estimates—and keep the role of local labor-market concentration in view.
One reform, different costs for different firms
Brazil’s corporate tax system gave firms more than one route for paying taxes. Under the standard regime, firms paid several contributions and a flat payroll tax of 20 percent on wages, alongside other corporate taxes. A separate system called SIMPLES consolidated several taxes and social-security contributions into a single payment based on gross revenue for eligible small and medium-sized firms. The PBM did not directly change the labor costs of firms in SIMPLES. Its direct intervention was aimed at standard-regime firms in selected sectors.
Beginning in 2011, the PBM changed the base used to calculate the social-security contribution for those selected sectors. Instead of paying the 20 percent payroll contribution, eligible standard-regime firms paid between 1 and 2 percent of gross revenue, with export earnings excluded from the revenue base. In ordinary language, the government moved part of the charge from the firm’s wage bill to its revenue. That distinction matters because a firm with many employees relative to its revenue could receive a much larger effective labor-cost reduction than a firm with a smaller wage bill.
The replacement contribution also means that the policy was not a pure payroll-tax waiver. Revenue-based taxation can discourage output, while lower payroll taxation can encourage labor use. The paper calculates that the payroll-tax effect dominated for almost all sectors in the main analysis, and that the overall burden fell by close to zero to 84 percent depending on the sector. The weighted average reduction was about 51 percent of the prior burden. A few sectors could experience a small increase rather than a reduction. The careful conclusion is therefore that the PBM created a large, uneven labor-cost shock—not that every treated employer received the same subsidy.
The change in tax base also helps explain why treatment intensity is economically meaningful. Under the old arrangement, the contribution rose with the wage bill. Under the new arrangement, it rose with revenue instead. Two firms with the same payroll could therefore experience different changes if their revenue, labor intensity, or export share differed. Conversely, a sector with a high share of labor costs could see a much larger reduction in the cost of adding an employee. Later comparisons by expected burden reduction are consequently not arbitrary “strong” and “weak” policy groups: they approximate how much the reform changed the incentive to use labor in each sector. They still remain grouped estimates, not a promise that the response increases in one exact proportion for every firm.
This distinction matters for the concentration question as well. A cost reduction that is large for one sector and small for another changes the pressure on employers unevenly, while a labor market may contain firms facing different tax regimes at the same time. The policy can therefore alter both the absolute cost of labor and the competitive position of a treated firm. Reading the PBM as a single national tax-rate change would hide the relative-cost channel that the later analysis connects to local employer concentration.
The reform’s eligibility rules created two further boundaries. Some firms or sectors were included through lists of product codes under the Mercosur Common Nomenclature, or NCM, rather than through their economic sector. The researchers could not observe the share of a firm’s revenue coming from eligible products, so they could not measure the extent of the relief reliably for those firms. The main analysis therefore focuses on broad economic sectors with sector-based eligibility and excludes categories touched by the problematic product criterion. The paper reports no statistically distinguishable effects for the NCM-exposed group in a separate appendix analysis, but those sectors do not provide the clean treatment comparison used for the headline result.
The remaining reform was rolled out gradually. It began in December 2011 with sectors such as software, call centers, and hotels, then expanded through later legislative acts. By the end of the period studied, 65 four-digit ISIC sectors had entered the new regime. The timing difference is not just a historical detail. It gives the researchers observations before and after implementation for the same sector, as well as comparisons with sectors that had not yet been treated or were never treated.
The selective design also connects the tax rules to local labor-market concentration. A treated standard-regime firm could compete for workers against SIMPLES firms or firms in sectors not yet covered by the reform. The PBM therefore changed one employer’s cost relative to some of its potential competitors rather than moving every employer in a local market together. This relative change is what allows the paper to ask whether the response was stronger where workers had more competing employers to choose from.
From administrative records to a comparison path
The main outcome data come from RAIS, Brazil’s matched employer–employee administrative census of formal employment. Employers report information on firms and workers, including sector, tax regime, admission and dismissal dates, education, age, sex, and December wages. The dataset is reported annually, but the admission and dismissal dates allow the researchers to reconstruct the number of people employed each month. That monthly reconstruction is important because the PBM entered sectors at specific months rather than at one common annual date.
The main sample contains 219 four-digit sectors. Sixty-five eventually received the sector-based PBM treatment, while 154 never received it during the relevant design. The resulting panel has 15,768 sector-month observations. It covers about 20.4 million workers and 1.9 million firms, roughly half of total employment and active firms in 2014. These figures describe the weight of the sample; they do not mean that the PBM applied to half of Brazil’s employers. Firms in SIMPLES are left out of the principal directly treated sample, and the sector selection excludes several areas whose eligibility could not be measured cleanly.
RAIS is especially useful for this question because it is not a small survey of selected workplaces. The Brazilian government uses the records to administer benefits, and employers have incentives to report accurately. The data connect a firm’s sector and tax regime to individual employment spells, which lets the researchers observe both the aggregate sector outcome and the timing of hiring and separation. At the same time, the strength of the records creates a boundary: RAIS is a census of formal work, not a census of every economic activity. A rise in RAIS employment can represent new formal work, the registration of previously informal work, or movement into the observed employer population. It is a strong measure of the formal margin, not a complete measure of all work.
The sample also makes the comparison internally coherent. The never-treated sectors help estimate common month effects and broad-sector seasonality even though they do not themselves identify a treatment effect through a change in tax status. The treated sectors supply variation in implementation timing and burden intensity. This arrangement is useful precisely because it does not ask the reader to compare unrelated countries or distant historical periods; it asks whether a sector’s path changed when its own eligibility began, relative to the paths of sectors operating under the same national conditions.
The research design is an event study. Instead of asking only whether treated sectors had higher employment after the policy than before it, the researchers line up each sector’s outcomes around its own implementation month. They then compare those event-time changes with never-treated sectors, while accounting for differences that do not change over time within a sector and shocks common to all sectors in a given month.
This alignment gives the coefficients a practical meaning. A coefficient before implementation asks whether a sector that will later receive the reform was already moving differently from the comparison sectors at that point in the timeline. A coefficient after implementation asks whether its outcome is now above or below the comparison path, after the fixed differences and common shocks have been accounted for. The design can therefore show both anticipation or pre-existing divergence and the timing of the response. It does not require every sector to have the same employment level; it requires the comparison of changes to be credible enough for the timing of the gap to be informative.
A sector fixed effect absorbs stable differences such as long-standing technology, geography, or market structure. A month fixed effect absorbs nationwide changes in the economy. The design also includes month effects interacted with broad economic-sector categories so that seasonal patterns specific to, say, services or construction do not get mistaken for a tax effect. In some specifications, the researchers add time interactions with pre-reform characteristics that help predict when a sector entered the PBM, including the average age of workers, the share of men, and average employees per firm.
The most important diagnostic is the pre-treatment path. If employment in sectors that entered earlier was already rising or falling differently before implementation, a post-treatment gap could reflect that trajectory rather than the reform. The paper reports no differential pre-trend in employment before PBM implementation. That result supports the comparison: the treated sectors do not appear to have started diverging in the measured outcome just before the tax change.
But a flat pre-trend is a diagnostic, not a certificate of random assignment. The timing of implementation is correlated with some baseline sector characteristics. Earlier-treated sectors had, in some specifications, more younger and male workers and slightly smaller firms. The authors control for the characteristics that predict timing, but unmeasured political or economic selection could still matter if it was connected to future sector-specific changes that the model does not capture.
The paper also addresses a technical problem created by staggered timing. A conventional two-way fixed-effects estimate can combine comparisons in ways that are difficult to interpret when some sectors are already treated while others are just becoming treated. The authors re-estimate the main outcomes with interaction-weighted, difference-in-differences-with-multiple-periods, and imputation approaches designed for staggered adoption. The employment and wage results are in line with the baseline. That agreement reduces concern that the central result is an artifact of one estimator, although it does not remove the more substantive limits of the sample or policy setting.
The study period is similarly deliberate. The main analysis ends in 2014. Starting in 2015, the federal government began to reverse parts of the reform, change revenue-based rates, make participation optional in some circumstances, and later exclude several sectors. Brazil also entered a severe economic downturn in 2015–2016. Extending the sample without accounting for those changes would mix the original treatment with a different policy regime and a different macroeconomic environment. The paper reports qualitatively similar and somewhat larger results when later data are included, but uses the earlier window for its main interpretation.
These choices define what the event study can establish. It can compare the path of formal sector employment around a staggered, sector-based cost change and test whether a pre-existing differential trend is visible. It cannot show that the PBM was randomly assigned, that every industry would respond in the same way, or that the measured employment effect is entirely new work rather than some combination of firm expansion, entry, formalization, and tax-regime movement. It also cannot make local labor-market concentration disappear as a possible source of heterogeneous response; concentration is part of the relationship the paper tests after establishing the average effect.
The comparison is therefore stronger than a simple before-and-after story but narrower than a universal policy experiment. The distinction is crucial. If employment rises after the tax change, the next question is not merely whether the graph slopes upward. It is whether the timing, intensity, treatment boundary, and absence of a pre-trend make the labor-cost explanation more credible than the alternatives. The answer will still need to account for wages, firm behavior, and the concentration of employers competing for workers.
With the institutional rules and comparison now clear, the headline estimates can be read on their proper scale. The next result concerns the two immediate margins: how much employment changed after implementation, and whether the lower cost appeared in the measured wages of new hires.